List of Feature Definition Files

Features are organized and bundled into Feature Definition Files (FDF). Each of the 92 FDFs contains several single features which can be selected or de-selected individually. Moreover, FDFs are sorted in different categories with respect to the character and nature of the contained features. Accent features are mostly equivalent to binary structural markers for positions on which tonal events might be perceived as more salient than others. Auxiliary features are no features in the strict sense, but are useful for advanced and combined computations. Tone Formation features deal with articulations (staccato, legato etc.) and annotated modulations . Contour features try to capture aspects of a melody contour in time vs. pitch space. Interval features derive from differences of pitch and pitch class values. They come in four different levels of exactness: semitone intervals, fuzzy intervals (aka ‘refined contour’), Parson’s Codeyear (aka ‘contour’), as well as interval classes. Metadata features access annotated metadata of melodies. Meter features deal with metrical annotations and derived transformations. Rhythm deals with the time componente of melodies mostly durations and inter-onset intervals. Sequence features come in three different flavours: interval, pitch and rhythm. They try to catch aspects of sequentiality, such as bi- and trigrams and runlengths. Finally, Structure features are based on self-similarities of melodies on the level of individual phrases.

Note

FDFs are mostly of two different varieties: Those bundling single, scalar-valued features, and those containing vector or matrix-valued data. Vector-valued features are often either histograms or event-based features, which return for each event a certain value, e.g., a list of raw values of a certain transformation. For a better understanding of the transformation underlying many features, a look into basic tranformations might be helpful.

Auxiliary

Feature Bundle
Melody Export
Sections IDs and Labels

MLA

Feature Bundle
MLA Exports

Metadata

Feature Bundle
Solo Metadata
EsAC Metadata

Tone Formation

Feature Bundle
Loudness
Articulation
Modulation

Accents

Interval jump accents

  • Description: Large pitch jumps are thought to convey accents, i.e., have a higher salience. However, no consensus is reached what means ‘large’ nor if it should be absolute or relative jumps nor if the tone before or after the jump or both should receive the accent. Some basic variants are collected here, but further research is needed to settle the issue.
  • Feature Definition File Name: ACCENTS_JUMPS
  • Category: Accents

Included Features:

Label Type Length Description
jumpaft3 Integer [0,1] N Raw vector of accents for tones following a large pitch jump of at least 3 semi-tones (either direction).
jumpaft4 Integer [0,1] N Raw vector of accents for tones following a large pitch jump of at least 4 semi-tones (either direction).
jumpaft5 Integer [0,1] N Raw vector of accents for tones following a large pitch jump of at least 5 semi-tones (either direction).
jumpbef3 Integer [0,1] N Raw vector of accents for tones before a large pitch jump of at least 3 semi-tones (either direction).
jumpbef4 Integer [0,1] N Raw vector of accents for tones before a large pitch jump of at least 4 semi-tones (either direction).
jumpbef5 Integer [0,1] N Raw vector of accents for tones before a large pitch jump of at least 5 semi-tones (either direction).
jumpbea3 Integer [0,1] N Raw vector of accents for tones before and after a large pitch jump of at least 3 semi-tones (either direction).
jumpbea4 Integer [0,1] N Raw vector of accents for tones before and after a large pitch jump of at least 4 semi-tones (either direction).
jumpbea5 Integer [0,1] N Raw vector of accents for tones before and after a large pitch jump of at least 5 semi-tones (either direction).
jumploc Integer [0,1] N Raw vector of accents for tones after a pitch interval that is at least 1 semi-tone larger than the previous interval.
jumploc2 Integer [0,1] N Raw vector of accents for tones after a pitch interval that is at least 2 semi-tone larger than the previous interval.
thom Real [0, 1] N Raw vector of accents according to Thomassen’s algorithm (1982), which is based on the seven possible pitch direction patterns that can be formed by 2-interval chains (3-note patterns). Values are not binary, but probabilities.
thom_thr Inetger[0,1] N Raw vector of Thomassen accents with exceed a certain threshold (.75 in this case).

Duration accents (Abs. IOI)

  • Description: Durations accent are based on the music-psychological observation, that for short-long (SL) patterns the second tone is perceived as accented (more pronounced, more salient). Included here are several accent features that operationalize the intuitive notion of short-long in several different ways based on IOI classification according to a absolute reference time of 0.5 s.
  • Feature Definition File Name: ACCENTS_DURATION_ABS
  • Category: Accents

Included Features:

Label Type Length Description
longpr_abs Integer [0,1] N Raw accent vector of tones with an higher (longer) IOI class than the previous one.
long2pr_abs Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes away from that of the previous one.
longmod_abs Integer [0,1] N Raw accent vector of tones with a higher/longer IOI class than the most frequent (mode) IOI class.
long2mod_abs Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes away from the most frequent (mode) IOI class.
longmod_win5_abs Integer [0,1] N Raw accent vector of tones with an IOI class longer than mode IOI class of the past 5 IOIs.
long2mod_win5_abs Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes longer than the mode IOI class of the past 5 IOIs.

Structural accents (phrases)

  • Description: Structurally important positions can give rise to accents (and vice versa). Currently, only phrases markers are implemented here.
  • Feature Definition File Name: ACCENTS_STRUCTURE
  • Category: Accents

Included Features:

Label Type Length Description
phrasbeg Integer [0,1] N Raw vector of markers for phrase beginnings (if available).
phrasend Integer [0,1] N Raw vector of markers for phrase endings (if available).
phrasbor Integer [0,1] N Raw vector of markers for phrase beginning & endings (if available).

Duration accents (Rel. IOI)

  • Description: Durations accent are based on the music-psychological observation, that for short-long (SL) patterns the second tone is perceived as accented (more pronounced, more salient). Included here are several accent features that operationalize the intuitive notion of short-long in several different ways, but always based on IOI classification according to a relative reference time of beat duration.
  • Feature Definition File Name: ACCENTS_DURATION_REL
  • Category: Accents

Included Features:

Label Type Length Description
longpr_rel Integer [0,1] N Raw accent vector of tones with an higher (longer) IOI class than the previous one.
long2pr_rel Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes away from that of the previous one.
longmod_rel Integer [0,1] N Raw accent vector of tones with a higher/longer IOI class than the most frequent (mode) IOI class.
long2mod_rel Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes away from the most frequent (mode) IOI class.
longmod_win5_rel Integer [0,1] N Raw accent vector of tones with an IOI class longer than mode IOI class of the past 5 IOIs.
long2mod_win5_rel Integer [0,1] N Raw accent vector of tones with an IOI class at least two classes longer than the mode IOI class of the past 5 IOIs.

Pitch extrema accents

  • Description: Extreme pitch values (maxima and minima) are likely positions to stand out from the context. Some variations in operationalisation are possible (e.g., with respect to cambiatas and trillers).
  • Feature Definition File Name: ACCENTS_PITCH_EXTREMA
  • Category: Accents

Included Features:

Label Type Length Description
pextrem Real [0,1] N Raw accent vector of pitch extrema (no restrictions).
pextrmf Real [0,1] Var Raw accent vector of pitch extrema (excluding proper cambiata).
pextrst Real [0,1] N Raw accent vector of pitch extrema (sensu Steinbeck, with at least two intervals before and after the extrema leading strictly to it).

Metrical accents (on/offbeat)

  • Description: Metrical accents are one the most fundamental accents. Classically, meter is partly defined as an underlying regular accent structure. Syncopations are likely to give rise to accents as well, since they violate metrical expectations. Here, on-beat as well as off-beat (i.e., syncopation) accents collected.
  • Feature Definition File Name: ACCENTS_METER
  • Category: Accents

Included Features:

Label Type Length Description
beat1 Integer [0,1] N Raw vector of accents on the primary accent (first beat) of a bar.
beat3 Integer [0,1] N Raw vector of accents on the secondary accent of a bar, if present, e.g. in on the 3rd beat in 4/4-measures (hence the name).
beat13 Integer [0,1] N Raw vector of accents on the primary and secondiry accent of a bar.
beatall Integer [0,1] N Raw vector of accents on all beat positions in a bar.
sync1 Integer [0,1] N Raw vector of accents on all syncopations right before the primary accent in a bar (‘anticipated 1’).
sync3 Integer [0,1] N Raw vector of accents on all syncopations right before the secondary accent in a bar (‘anticipated 3’).
sync13 Integer [0,1] N Raw vector of accents on all syncopations right before primary and secondary accents in a bar (‘anticipated 1s and 3s’).
sync1234 Integer [0,1] N Raw vector of accents on all syncopations right before all beats in a bar.
syncall Integer [0,1] N Raw vector of accents on all syncopations on every sub-beat metrical level (i.e., excluding half-beat level).
swing_markers Integer [0,1] N Raw vector of structural markers for swing candidates.

Duration accents (Raw IOI)

  • Description: Durations accent are based on the music-psychological observation, that for short-long (SL) patterns the second tone is perceived as accented (more pronounced, more salient). Icnluded here are several accent features that operationalize the intuitive notion of short-long in several differnet ways based on raw IOIs.
  • Feature Definition File Name: ACCENTS_DURATION
  • Category: Accents

Included Features:

Label Type Length Description
longpr Integer [0,1] N Raw accent vector of tones with longer IOI than the previous tone.
long2pr Integer [0,1] N Raw accent vector of tones with an at least twice longer IOI than the previous tone.
longmod Integer [0,1] N Raw accent vector of tones with a longer IOI than mode IOI of all tones.
long2mod Integer [0,1] N Raw accent vector of tones with an at least twice longer IOI than mode IOI of all tones.
longmod_win5 Integer [0,1] N Raw accent vector of tones with IOI that is at least 41% longer than mean of the past 5 IOIs.
long2mod_win5 Integer [0,1] N Raw accent vector of tones with IOI that is at least twice as long than mean of the past 5 IOIs.

Harmony accents

  • Description: Structural marker for in-chord or out-of-chord tones (only for melodies with annotated chords).
  • Feature Definition File Name: ACCENTS_HARMONY
  • Category: Accents

Included Features:

Label Type Length Description
triad Integer [0,1] Var Raw markers of chord tones (excluding upper structures).
inchord Integer [0,1] Var Raw markers of chord tones (including upper structures).
outchord Integer [0,1] Var Raw marker vector of non-chord tones.

Auxiliary

Melody Export

  • Description: Exports melody as lists of onset, duration, pitch, and metrical positions.
  • Feature Definition File Name: GENERAL_MELODY_RAW
  • Category: Auxiliary

Included Features:

Label Type Length Description
onset Real N Onsets (in seconds) of events.
norm_onset Real N Normalized onsets [0-1] of events.
first_onset Real 1 First onset (in seconds) of events.
last_offset Real 1 Last offset (in seconds) of events.
duration Real N Duration (in seconds) of events.
pitch Real N Raw MIDI pitch values of events.
metrical_position String N Metrical positions of events (MP-dot-notation).
phraseid String Var ID of containing phrase.
offset Real N Offsets (in seconds) of events.

Sections IDs and Labels

  • Description: Returns sequences of raw phrase IDs, form labels, chorus IDs, chord labels and chord types.
  • Feature Definition File Name: GENERAL_SECTION_IDS
  • Category: Auxiliary

Included Features:

Label Type Length Description
phrase_id_raw Integer N_P Sequence of event-wise phrase IDs.
chorus_id_raw Integer Var Sequence of event-wise chorus IDs.
form_labels_raw Integer Var Sequence of event-wise form labels (as annotated).
chords_raw Integer Var Sequence of event-wise chord symbols (as annotated and normalized).
chord_types_raw Integer Var Sequence of event-wise chord type labels (triad type + seventh).
chord_types_sections Integer Var Type labels for chord sections (triad type + seventh).

Contour

Waviness/Zig-Zagity.

  • Description: This feature set contains features related to (pitch-)extremal points in a melody. A local extremum (or turning) point can be either a minimum or maximum. Hence, pitches right before and right after an extremal point are both lower or equal (maximum) or higher or equal (minimum). For this module only true extremal points are counted, where the pitches must be strictly higher or lower to both sides. The percentage of extremal points can be regarded as a measure for zig-zagity (or waviness) of a melody. The extreme case are two alternatine pitches, where each and every point is a local extremum resulting in an extrema ratio of 1.
  • Feature Definition File Name: PITCH_WAVINESS
  • Category: Contour

Included Features:

Label Type Length Description
extrema_ratio Real 1 Percentage of extremal points.
extrema_count Integer 1 Number of extremal points.
note_count Integer 1 Number of notes.

Contour Descriptors

  • Description: Calculates various contour descriptors. First, Huron contour in three different varieties is supported: Code, Reduced Code and Numeric. Huron contour is calculated on a sequence of pitches by size comparison of the first pitch to the mean inner pitch (MIP, i.e., mean of MIDI pitches except the first and last element), and the MIP to the last pitch. There are three possible relations between two numbers: Greater (>), equal (=), or less (<). Combining two comparisons gives a set of nine different contour values: descending (>>), descending-horizontal (>=), concave (<>), horizontal-descending (=>), horizontal (==), horizontal-ascending (=<), convex (><), ascending-horizontal (<=), and ascending (<<). Short mnemonic codes for them are desc, desc-hor, concave, hor-desc, hor, hor-asc, convex, asc-hor, asc. The reduced code is a mapping of the mixed-horizontal category to their non-horizontal partner, i.e., desc-hor and hor-desc become desc and asc-hor and hor-asc become asc, which leaves 5 categories. The numeric values range from -4 to 4 in the same order. (Cf. Huron (1994), The Melodic Arch in Western Folksongs.). Second, gradient countour measures the pitch gradient \Delta p/\Delta T between extreme points. Third, Abesser contour is similar to Huron contour, but uses more stable estimators and is more suitable for longer sequences.
  • Feature Definition File Name: CONTOUR
  • Category: Contour

Included Features:

Label Type Length Description
huron_contour_code String 1 Huron contour codes. One of desc, desc-hor, concave, hor-desc, hor, hor-asc, convex, asc-hor, asc.
huron_contour_redcode String 1 Reduced Huron contour codes. One of desc, concave, hor, convex, asc.
huron_contour_num Integer [-4:+4] 1 Numeric value of Huron contour.
gradient_contour Real Var List of gradients \Delta p/\Delta T between extreme values of note track.
abesser_contour_code String 1 Abesser contour codes. One of desc, desc-hor, concave, hor-desc, hor, hor-asc, convex, asc-hor, asc.
abesser_contour_redcode String 1 Reduced Abesser contour codes. One of desc, concave, hor, convex, asc.
abesser_contour_num Integer [-4:+4] 1 Numeric value for Abesser contour.

Interval

Parson’s Code Single Densities

  • Description: Normalized histogram densities, mode, entropy and Zipf coefficient for Parson’s code. Values: down (-1), repeat/unisone (0), up (+1).
  • Feature Definition File Name: PARSON_FEATURES
  • Category: Interval

Included Features:

Label Type Length Description
parsons_hist_descending Real 1 Relative frequency for down (-1).
parsons_hist_constant Real 1 Relative frequency for repeat (0).
parsons_hist_ascending Real 1 Relative frequency for up (+1).
parsons_mode Real Var Mode of Parson’s Code distribution (can be list!).
parsons_entropy Real 1 Normalized entropy of Parson’s Code distribution.
parsons_zipf Real 1 Zipf coefficient of Parson’s Code distribution.

Semitone Interval Single Features

  • Description: Calculates set of basic statistical sample descriptors (mean, median, mode, min, max, range, var, std) for interval and absolute interval distribution.
  • Feature Definition File Name: INT_FEATURES
  • Category: Interval

Included Features:

Label Type Length Description
int_mean Real 1 Mean of interval distribution.
int_median Real 1 Median of interval distribution.
int_var Real 1 Variance of of interval distribution.
int_std Real 1 Standard deviation of interval distribution.
int_min Real 1 Minimum of of interval distribution.
int_max Real 1 Maximum of of interval distribution.
int_range Real 1 Range of of interval distribution.
int_mode Real Var Mode of interval distribution (can be list!).
int_zipf Real 1 Zipf coefficient of interval distribution.
int_entropy Real 1 Entropy of interval distribution (bits).
abs_int_mean Real 1 Mean of absolute interval distribution.
abs_int_median Real 1 Median of absolute interval distribution.
abs_int_var Real 1 Variance of absolute interval distribution.
abs_int_std Real 1 Standard deviation of interval distribution.
abs_int_min Real 1 Minimum of absolute interval distribution.
abs_int_max Real 1 Maximum of absolute interval distribution.
abs_int_range Real 1 Range of absolute interval distribution.
abs_int_mode Real 1 Mode of interval absolute distribution (can be list!).
abs_int_zipf Real 1 Zipf coefficient of absolute interval distribution.
abs_int_entropy Real 1 Entropy of absolute interval distribution (bits).
abs_int_sum Real 1 Sum of absolute intervals (= distance moved in pitch space).

Parson’s Code Distribution (“Contour”)

  • Description: Histogram and raw vector of Parson’s Code (contour). Values: down (-1), repeat/unisone (0), up (+1).
  • Feature Definition File Name: PARSON_HIST
  • Category: Interval

Included Features:

Label Type Length Description
parsons_raw Integer [-1:+1] N Raw output (Parson’s code values of all melody notes).
parsons_hist_values Integer [-1:+1] 3 Histogram bins for Parson’s code.
parsons_hist_counts Real 3 Histogram bin counts (abs. frequencies).

Semitone Interval Distribution

  • Description: Histogram and raw vector of semitone intervals, either limited to the range of -20 to 20 semitones or full range (only raw and with appended zero element).
  • Feature Definition File Name: INT_HIST
  • Category: Interval

Included Features:

Label Type Length Description
int_raw Integer [-127:+127] N-1 ist of semitone intervals.
int_hist_values Integer [-20:+20] 41 Bins of semitone intervals [-20:20].
int_hist_counts Integer 41 Absolute frequencies of semitone intervals.
int_raw_full_pad Integer [-127:+127] N List of semitone intervals with appended 0.

Fuzzy Interval Distribution (Refined Contour)

  • Description: Histogram and raw vector of fuzzy intervals (aka refined contour). Values: big jump up/down (+/-4), jump up/down (+/-3), leap up/down (+/-1), step up/down (+/-1), repetition (0).
  • Feature Definition File Name: FUZZYINT_HIST
  • Category: Interval

Included Features:

Label Type Length Description
fuzzyint_raw Integer [-4:4] N-1 Raw output (Fuzzy interval values over the full melody).
fuzzyint_hist_values Integer [-4:4] 5 Histogram bins.
fuzzyint_hist_counts Real 5 Histogram bin counts (abs. frequencies).

Fuzzy Interval (Refined Contour) Single Features

  • Description: Histogram counts, mode, entropy and Zipf coefficient of distribution of all 9 fuzzy interval classes (FUZZYINT). Values: big jump up/down (+/-4), jump up/down** (+/-3), leap up/down (+/-1), step up/down (+/-1), repetition (0).
  • Feature Definition File Name: FUZZYINT_FEATURES
  • Category: Interval

Included Features:

Label Type Length Description
fuzzyint_hist_09_big_jump_down Real 1 Relative frequency big jump down (-4).
fuzzyint_hist_08_jump_down Real 1 Relative frequency jump down (-3).
fuzzyint_hist_07_leap_down Real 1 Relative frequency leap down ( -2).
fuzzyint_hist_06_step_down Real 1 Relative frequency step down (-1).
fuzzyint_hist_05_repeat Real 1 Relative frequency repetition (0).
fuzzyint_hist_04_step_up Real 1 Relative frequency step up (+1).
fuzzyint_hist_03_leap_up Real 1 Relative frequency leap up (+2).
fuzzyint_hist_02_jump_up Real 1 Relative frequency jump up (+3).
fuzzyint_hist_01_big_jump_up Real 1 Relative frequency big jump up (+4).
fuzzyint_mode Real Var Mode of FUZZYINT distribution (can be list!).
fuzzyint_zipf Real 1 Zipf coefficient of FUZZYINT distribution.
fuzzyint_entropy Real 1 Normalized entropy of FUZZYINT distribution.

Interval Class Distribution

  • Description: Histogram and raw vector of intervals classes. Interval classes are minimal differences between pitches classes (i.e., length of shortest path on the circle) and have values from 0 to 6.
  • Feature Definition File Name: INTCLASS_HIST
  • Category: Interval

Included Features:

Label Type Length Description
ic_raw Integer [0:+6] N-1 List of interval classes.
ic_hist_values Integer [0:6] 7 Bins of Interval classes [0:6].
ic_hist_counts Integer 7 Absolute frequencies of interval classes.
ic_raw_pad Integer [0:+6] N List of interval classes with appended 0.

Interval Class Single Features

  • Description: Normalized histogram count, and basic statistics of interval class distribution. Interval classes are minimal differences between pitches classes (i.e., length of shortest path on the circle) and have values from 0 to 6.
  • Feature Definition File Name: INTCLASS_FEATURES
  • Category: Interval

Included Features:

Label Type Length Description
ic_mean Real 1 Mean of interval class distribution.
ic_median Real 1 Median of interval class distribution.
ic_var Real 1 Variance of of interval class distribution.
ic_std Real 1 Standard deviation of interval class distribution.
ic_min Real 1 Minimum of of interval class distribution.
ic_max Real 1 Maximum of of interval class distribution.
ic_range Real 1 Range of of interval class distribution.
ic_mode Real Var Mode of interval class distribution (can be list!).
ic_zipf Real 1 Zipf coefficient of interval class distribution.
ic_entropy Real 1 Entropy of interval class distribution (bits).
ic_hist_density_0 Real 1 Relative frequency IC = 0.
ic_hist_density_1 Real 1 Relative frequency IC = 1.
ic_hist_density_2 Real 1 Relative frequency IC = 2.
ic_hist_density_3 Real 1 Relative frequency IC = 3.
ic_hist_density_4 Real 1 Relative frequency IC = 4.
ic_hist_density_5 Real 1 Relative frequency IC = 5.
ic_hist_density_6 Real 1 Relative frequency IC = 6.

Parson’s Code Up/Down Ratio

  • Description: Ratio of number of ascending and descending intervals in a melody (ignoring unisones).
  • Feature Definition File Name: PARSON_UP_DOWN_RATIO
  • Category: Interval

Included Features:

Label Type Length Description
ratio_ascending_descending Real [0,1] 1 Ratio of ascending to descending intervals.

MLA

MLA Exports

  • Description: Exports several mid-level analysis (MLA) related features (section event-based).
  • Feature Definition File Name: MLA_RAW
  • Category: MLA

Included Features:

Label Type Length Description
MLA_raw String Var Raw list of mid-level units.
MLA_main_type String Var Raw list of main types of mid-level units.
MLA_full_type String Var Raw list of full types of mid-level units.
MLA_backref String Var Raw list of back references for mid-level units.
MLA_glue String Var Raw list of glue operators of mid-level units.
MLA_modifier String Var Raw list of modifiere for mid-level units.
MLA_length String Var Raw list of idea length as number of notes.
MLA_duration String Var Raw list of idea duration in secs onset-offset.
MLA_duration_bars String Var Raw list of idea duration in decimal bar time.
MLA_main_dir String Var Raw list of main directions of mid-level units.

Metadata

Solo Metadata

  • Description: Export all metadata for Jazzomat solos (currently only a subset).
  • Feature Definition File Name: GENERAL_SOLO_METADATA
  • Category: Metadata

Included Features:

Label Type Length Description
performer String 1 Performer.
title String 1 Title.
titleaddon String 1 Add-on for title.
solopart String 1 Running number of a solo in a piece.
full_title String 1 Full title (including solo part and title addon).
instrument String 1 Instrument of the soloist. One of ts, ts-c, as, bs, bsx, cbsx, ss, sss, cl, bcl, acl, tp, tpt, flgn, tb, fl, cor, ptp, frhn, ob, voc or empty.
style String 1 Style of piece. One of TRADITIONAL, SWING, BEBOP, COOL, HARDBOP, POSTBOP, FREE, FUSION, OTHER, MIX or empty.
avgtempo String 1 Avg. Tempo (bpm).
tempoclass String 1 Tempo class. One of SLOW, MEDIUM SLOW, MEDIUM, MEDIUM UP, UP or empty.
rhythmfeel String 1 Basic rhythmic feel/groove style. One of TWOBEAT, SWING, BALLAD, LATIN, FUNK or empty.
key String 1 Key or tonal center of the piece.
filename_sv String 1 Filename of originating SV project file.
status String 1 Status of transcription. One of PREFINAL, FINAL, DRAFT, ASSIGNED or empty.
composer String 1 Composer of the piece.
tonality_type String 1 Tonality type of the composition. One of FUNCTIONAL, MODAL, COLOR, FREE, BLUES, JAZZ-BLUES or empty.
genre String 1 Genre of the composition. One of TRADITIONAL, BLUES, GREAT AMERICAN SONGBOOK, WORMS, ORIGINAL, RIFF or empty.
form String 1 Form of the composition.
signature String 1 Signature(s) of solo.
mbzid String 1 Music Brainz ID.
recordingdate String 1 Recording date(s) of containing record.
recordingyear String 1 Recording year of containing record.
recordbib String 1 Bibliography of record.
recordtitle String 1 Title of containing record.
label String 1 Record label of containing record.
solostart String 1 Start time of solo in full track.
soloend String 1 End time of solo in full track.
lineup String 1 Line-up of track.
chord_changes String 1 Chord changes as compact string.
chorus_count String 1 Number of full choruses in the solo.
harmony_template String 1 Underlying harmony template of composition.
solo_time String 1 Start/End of solo in surrounding track.

EsAC Metadata

  • Description: Export all metadata for EsAC tunes.
  • Feature Definition File Name: GENERAL_ESAC_METADATA
  • Category: Metadata

Included Features:

Label Type Length Description
collection String 1 Collection.
esac_title String 1 Orignal EsAC title.
region String 1 Region of origin.
esac_key String 1 Annotated key (tonic).
unit Integer (2,4,8,16,32) 1 Annotated minimal rhythmic unit.
signature String 1 Annotated signature (possible more than one, space separate, or FREE).
source String 1 Annotated key (tonic).
function String 1 Annotated (ritual) function.
comment String 1 Comment section.
cnr String 1 CNR section (with yet unknown meaning!).
tunefamily String 1 Tunefamily based on EsAC-Ids.
text String 1 Annotated lyrics.
melstring String 1 Original melody code.

Meter

Metrical Circle Map Markov Distributions

  • Description: Histogram for Metrical Circle Map (N=48) unigrams and bigrams (occurence and transition probabilites). See this document or here for more details.
  • Feature Definition File Name: MCM_HIST
  • Category: Meter

Included Features:

Label Type Length Description
mcm_unigram_values Array of Integer [0:47] 48 Histogram bins (MCM:48 values).
mcm_unigram_freqs Real [0,1] 48 Histogram bin densities (MCM:48 values).
mcm_bigram_values Array of Integer Var Histogram bins (all unique MCM bigrams).
mcm_bigram_freqs Real [0,1] Var Histogram bin densities (all unique MCM bigrams).

Metrical Circle Map Markov Features

  • Description: Histogram for Metrical Circle Map (N=48) unigrams and bigrams (occurence and transition probabilites, entropies and Zipf coefficients). See this document or here for more details.
  • Feature Definition File Name: MCM_FEATURES
  • Category: Meter

Included Features:

Label Type Length Description
mcm_mean_angle Real 1 Angle of circular mean of MCM distribution.
mcm_mean_length Real 1 Length of circular mean of MCM distribution.
mcm_var Real 1 Circular variance of MCM distribution (=1-length of circular mean).
mcm_std Real 1 Circular standard deviation of MCM distribution.
mcm_disp Real 1 Circular dispersion of MCM distribution.
mcm_mode Real Var Mode of MCM distribution (can be list!).
mcm_zipf Real 1 Zipf coefficient of MCM distribution.
mcm_entropy Real 1 Normalized entropy of MCM distribution.

Meter Raw Exports

  • Description: Exports several meter related features (event-based).
  • Feature Definition File Name: METER_RAW
  • Category: Meter

Included Features:

Label Type Length Description
metrical_position String N Metrical positions (in MPD-notation) of events.
bar Integer N Event bar numbers.
beat Integer N Event beat numbers.
tatum Integer N Event tatum positions.
durtatum Integer N Events durations measured in current tatums (handle with care!).
metrical_weight Integer [0:2] N Metrical weight of event (0 for any subbeat event, 1 for metrical weak beat events, 2 for metrical strong events).
mcm_48 Integer [0:47] N Position in metrical circle map with N=48 divisions.
metrical_weight_transition_code Integer [0:8] N-1 Bigrams of metrical weights coded with ternary numbers. Let w_i, w_{i+1} be the weights of notes i, i+1. Then the metrical weight transition value is 3w_i + w_{i+1}.
period Integer N Event-based periods.
division Integer N Event-based beat divisions.
syncopation Integer [0:1] N Marker if event is syncopated (0=not syncopated, 1=syncopated).
syncopicity Real [0,1] 1 Ratio of syncopated events to non-syncopated events.
metric_complexity Real [0,1] 1 Combined metric complexity after Frieler, in preparation.
metric_complexity_division Real [0,1] 1 Division-based metric complexity after Frieler, in preparation.
metric_complexity_compression Real [0,1] 1 Compression-based metric complexity after Frieler, in preparation.

Pitch

Pitch Features

  • Description: Calculates set of basic statistical sample descriptors (mean, median, mode, min, max, range, var, std).
  • Feature Definition File Name: PITCH_FEATURES
  • Category: Pitch

Included Features:

Label Type Length Description
pitch_mean Real 1 Mean of PITCH distribution.
pitch_median Real 1 Median of PITCH distribution.
pitch_var Real 1 Variance of of PITCH distribution.
pitch_std Real 1 Standard deviation of PITCH distribution.
pitch_min Real 1 Minimum of of PITCH distribution.
pitch_max Real 1 Maximum of of PITCH distribution.
pitch_range Real 1 Range of of PITCH distribution.
pitch_mode Real Var Mode of PITCH distribution (can be list!).
pitch_zipf Real 1 Zipf coefficient of PITCH distribution.
pitch_entropy Real 1 Entropy of PITCH distribution (bits).

Tessitura Normalized Pitch

  • Description: Raw vector of scaled MIDI pitches with respect to pitch range.
  • Feature Definition File Name: TESS_NORM_PITCH
  • Category: Pitch

Included Features:

Label Type Length Description
tess_norm_pitch Real [0, 1] N Tessatura-scaled pitch values.

Chordal Diatonic Pitch Class Distribution

  • Description: Histogram and raw vector of chordal diatonic pitch classes (CDPC).
  • Feature Definition File Name: CDPC_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
cdpc_raw Integer N Raw output of note-wise CDPC values.
cdpc_hist_bins String Var Histogram bins (all unique CDPC values occuring in a melody).
cdpc_hist_density Real Var Histogram bin counts (absolute frequencies).
cdpc_raw_all Integer N Raw output of note-wise CDPC values including NC values.

Number of Different Pitch Classes.

  • Description: Number of different pitch classes.
  • Feature Definition File Name: PC_NUM_UNIQUE_PC
  • Category: Pitch

Included Features:

Label Type Length Description
number_of_unique_pc Integer 1 Average number of different pitch classes.

Tonal Pitch Class Single Features

Included Features:

Label Type Length Description
tpc_hist_density_00 Real 1 Relative frequency TPC = 0.
tpc_hist_density_01 Real 1 Relative frequency TPC = 1.
tpc_hist_density_02 Real 1 Relative frequency TPC = 2.
tpc_hist_density_03 Real 1 Relative frequency TPC = 3.
tpc_hist_density_04 Real 1 Relative frequency TPC = 4.
tpc_hist_density_05 Real 1 Relative frequency TPC = 5.
tpc_hist_density_06 Real 1 Relative frequency TPC = 6.
tpc_hist_density_07 Real 1 Relative frequency TPC = 7.
tpc_hist_density_08 Real 1 Relative frequency TPC = 8.
tpc_hist_density_09 Real 1 Relative frequency TPC = 9.
tpc_hist_density_10 Real 1 Relative frequency TPC = 10.
tpc_hist_density_11 Real 1 Relative frequency TPC = 11.
tpc_circ_mean_angle Real 1 Angle of circular mean of TPC distribution.
tpc_circ_mean_length Real 1 Length of circular mean of TPC distribution.
tpc_circ_var Real 1 Circular variance of TPC distribution (=1-length of circular mean).
tpc_circ_std Real 1 Circular standard deviation of TPC distribution.
tpc_circ_disp Real 1 Circular dispersion of TPC distribution.
tpc_entropy Real 1 Normalized entropy of TPC distribution.

Extended Chordal Diatonic Pitch Class Distribution

  • Description: Histogram and raw vector of extended multi-digit chordal diatonic pitch classes (CDPCX).
  • Feature Definition File Name: CDPCX_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
cdpcx_raw Integer N Raw output of note-wise CDPCX values.
cdpcx_hist_bins Integer N Histogram bins (all unique CDPCX values occuring in a melody).
cdpcx_hist_density Real N Histogram bin counts (abs. frequencies).
cdpcx_raw_all Integer N Raw output of note-wise CDPCX values including NC values.

Chordal Diatonic Pitch Class Features

  • Description: Normalized histogram counts, mode, entropy and Zipf coefficient of CDPC distribution.
  • Feature Definition File Name: CDPC_FEATURES
  • Category: Pitch

Included Features:

Label Type Length Description
cdpc_density_1 Real 1 Relative frequency CDPC = ‘1’.
cdpc_density_2 Real 1 Relative frequency CDPC = ‘2’.
cdpc_density_3 Real 1 Relative frequency CDPC = ‘3’.
cdpc_density_4 Real 1 Relative frequency CDPC = ‘4’.
cdpc_density_5 Real 1 Relative frequency CDPC = ‘5’.
cdpc_density_6 Real 1 Relative frequency CDPC = ‘6’.
cdpc_density_7 Real 1 Relative frequency CDPC = ‘7’.
cdpc_density_T Real 1 Relative frequency CDPC = ‘T’.
cdpc_density_B Real 1 Relative frequency CDPC = ‘B’.
cdpc_density_L Real 1 Relative frequency CDPC = ‘L’.
cdpc_density_< Real 1 Relative frequency CDPC = ‘<’.
cdpc_density_> Real 1 Relative frequency CDPC = ‘>’.
cdpc_mode Real Var Mode of CDPC distribution (can be list!).
cdpc_entropy Real 1 Normalized entropy of CDPC distribution.
cdpc_zipf Real 1 Zipf coefficient of CDPC distribution.

Pitch Class Single Features

Included Features:

Label Type Length Description
pc_hist_density_00_C Real 1 Relative frequency PC = C.
pc_hist_density_01_C# Real 1 Relative frequency PC = C#.
pc_hist_density_02_D Real 1 Relative frequency PC = D.
pc_hist_density_03_Eb Real 1 Relative frequency PC = Eb.
pc_hist_density_04_E Real 1 Relative frequency PC = E.
pc_hist_density_05_F Real 1 Relative frequency PC = F.
pc_hist_density_06_F# Real 1 Relative frequency PC = F#.
pc_hist_density_07_G Real 1 Relative frequency PC = G.
pc_hist_density_08_Ab Real 1 Relative frequency PC = Ab.
pc_hist_density_09_A Real 1 Relative frequency PC = A.
pc_hist_density_10_Bb Real 1 Relative frequency PC = Bb.
pc_hist_density_11_B Real 1 Relative frequency PC = B.
pc_circ_mean_angle Real 1 Angle of circular mean of PC distribution.
pc_circ_mean_length Real 1 Length of circular mean of PC distribution.
pc_circ_var Real 1 Circular variance of PC distribution (=1-length of circular mean).
pc_circ_std Real 1 Circular standard deviation of PC distribution.
pc_circ_disp Real 1 Circular dispersion of PC distribution.
pc_entropy Real 1 Normalized entropy of PC distribution.

Average Number of Pitch Classes per Phrase

  • Description: Average number of different pitch classes per phrase.
  • Feature Definition File Name: PC_AV_NUM_UNIQUE_PC_PHRASES
  • Category: Pitch

Included Features:

Label Type Length Description
mean_number_of_unique_pc Real [0:11] 1 Average number of different pitch classes per phrase.

Extended Chordal Diatonic Pitch Class Single Features

  • Description: Normalized histogram counts, mode, entropy and Zipf coefficient of CDPCX distribution.
  • Feature Definition File Name: CDPCX_FEATURES
  • Category: Pitch

Included Features:

Label Type Length Description
cdpcx_density_1 Real 1 Relative frequency CDPCX = ‘1’.
cdpcx_density_2 Real 1 Relative frequency CDPCX = ‘2’.
cdpcx_density_3 Real 1 Relative frequency CDPCX = ‘3’.
cdpcx_density_4 Real 1 Relative frequency CDPCX = ‘4’.
cdpcx_density_5 Real 1 Relative frequency CDPCX = ‘5’.
cdpcx_density_6 Real 1 Relative frequency CDPCX = ‘6’.
cdpcx_density_7 Real 1 Relative frequency CDPCX = ‘7’.
cdpcx_density_T Real 1 Relative frequency CDPCX = ‘T’.
cdpcx_density_B Real 1 Relative frequency CDPCX = ‘B’.
cdpcx_density_L Real 1 Relative frequency CDPCX = ‘L’.
cdpcx_density_< Real 1 Relative frequency CDPCX = ‘<’.
cdpcx_density_> Real 1 Relative frequency CDPCX = ‘>’.
cdpcx_density_b2 Real 1 Relative frequency CDPCX = ‘-‘ (b2).
cdpcx_density_b6 Real 1 Relative frequency CDPCX = ‘%’ (b6).
cdpcx_mode Real Var Mode of CDPCX distribution (can be list!).
cdpcx_entropy Real 1 Normalized entropy of CDPCX distribution.
cdpcx_zipf Real 1 Zipf coefficient of CDPCX distribution.

Tonal Pitch Class Distribution

  • Description: Histogram and raw vector of tonal pitch classes (TPC).
  • Feature Definition File Name: TPC_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
tpc_raw Integer N Raw output of note-wise TPC values.
tpc_hist_bins Integer [0:11] 12 Histogram bins (all unique TPC values occuring in a melody).
tpc_hist_counts Integer 12 Histogram bin count values (abs. frequencies).

Pitch Class Distribution

  • Description: Histogram and raw vector of pitch classes.
  • Feature Definition File Name: PC_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
pc_raw Integer [0:11] N Raw output (values for all melody notes).
pc_hist_values Integer [0:11] 12 Histogram bins.
pc_hist_dens Real 12 Histogram bin count (abs. frequencies).

Chordal Pitch Class Feature

Included Features:

Label Type Length Description
cpc_hist_density_00 Real 1 Relative frequency CPC = 0.
cpc_hist_density_01 Real 1 Relative frequency CPC = 1.
cpc_hist_density_02 Real 1 Relative frequency CPC = 2.
cpc_hist_density_03 Real 1 Relative frequency CPC = 3.
cpc_hist_density_04 Real 1 Relative frequency CPC = 4.
cpc_hist_density_05 Real 1 Relative frequency CPC = 5.
cpc_hist_density_06 Real 1 Relative frequency CPC = 6.
cpc_hist_density_07 Real 1 Relative frequency CPC = 7.
cpc_hist_density_08 Real 1 Relative frequency CPC = 8.
cpc_hist_density_09 Real 1 Relative frequency CPC = 9.
cpc_hist_density_10 Real 1 Relative frequency CPC = 10.
cpc_hist_density_11 Real 1 Relative frequency CPC = 11.
cpc_circ_mean_angle Real 1 Angle of circular mean of CPC distribution.
cpc_circ_mean_length Real 1 Length of circular mean of CPC distribution.
cpc_circ_var Real 1 Circular variance of CPC distribution (=1-length of circular mean).
cpc_circ_std Real 1 Circular standard deviation of CPC distribution.
cpc_circ_disp Real 1 Circular dispersion of CPC distribution.
cpc_zipf Real 1 Zipf coefficient of CPC distribution.
cpc_entropy Real 1 Normalized entropy of CPC distribution.

Chordal Pitch Class Distribution

  • Description: Histogram and raw vector of chordal pitch classes (CPC).
  • Feature Definition File Name: CPC_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
cpc_raw Integer [0:11] N Raw output (CPC values of all melody notes).
cpc_raw_all Integer [-1:11] N Raw output (CPC values of all melody notes, including undefined CPCs, set to -1).
cpc_hist_bins Integer [0:11] 12 Histogram bins.
cpc_hist_densities Real 12 Histogram bin count (absolute frequencies).

Pitch Distribution

  • Description: Histogram and raw vector of raw MIDI pitches.
  • Feature Definition File Name: PITCH_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
pitch_raw Integer [0:127] N Raw output (values of all melody notes).
pitch_hist_values Integer [0:127] 128 Histogram bins.
pitch_hist_counts Real 128 Histogram bin counts (abs. frequencies).

Chordal Pitch Class Circle Map Features (Experimental)

  • Description: Histogram for Chordal Pitch Class Circle Map (N=12) unigrams and bigrams (occurence and transition probabilites).
  • Feature Definition File Name: CPCCM_HIST
  • Category: Pitch

Included Features:

Label Type Length Description
cpccm_unigram_values Array of Integer [0:11] 12 Histogram bins (CPCCM:12 values).
cpccm_unigram_freqs Real [0,1] 12 Histogram bin densities (CPCCM:12 values).
cpccm_bigram_values Array of Integer Var Histogram bins (all unique CPCCM bigrams).
cpccm_bigram_freqs Real [0,1] Var Histogram bin densities (all unique CPCCM bigrams).

Rhythm

Variability Indices (Duration)

  • Description: Normalized pairwise variability index, which measures the mean contrast of adjacents durations. The formula is \frac{1}{N-1}\sum_{i=1}^{N-1} 2\left| \frac{d_{i+1}-d_{i}}{d_{i+1}+d_{i}}\right|. It is the mean of the absolute value of ratios of duration difference to duration mean. This index originated in linguistic and is sometimes used by rhythm researchers as a variability measure. The coefficient of variation is another common index for variability. It is defined as the ratio of standard deviation and mean. A simple biased estimator is the corresponding ratio of sample standard deviation and sample mean. The coefficient of variation is only sensical for ratio scales, which applies to durations since they only assum positive values. The formula is: c_V = \frac{\sigma(d_{i})}{\mu(d_{i})}.
  • Feature Definition File Name: VI_DUR
  • Category: Rhythm

Included Features:

Label Type Length Description
nPVI_dur Real [0,1] 1 Normalized pairwise variability index.
CV_dur Real 1 Coefficient of variation.

Microtiming Features

  • Description: Features related to micro-timing in a solo.
  • Feature Definition File Name: GENERAL_MICROTIMING
  • Category: Rhythm

Included Features:

Label Type Length Description
swing_ratios Real Var Swing ratios of beats with a binary sub-division as the ratio duration of the first to the second eighth in a beat. Straight binary eigths have a value of 1 (1:1), whereas triplet eighths get 2 and dotted eighths + sixteenth get 3:1.
mean_swing_ratio Real 1 Mean value of swing ratios.
median_swing_ratio Real 1 Median of swing ratios.
std_swing_ratio Real 1 Standard deviation of swing ratio.
swing_shapes Real Var Swing shapes are the loudness difference between first and second eighth in a beat with binary subdivision.
mean_swing_shape Real 1 Mean value of swing shapes.
median_swing_shape Real 1 Median of swing shapes.
std_swing_shape Real 1 Standard deviation of swing shapes.
number_binary_beats Real 1 Number of fully occupied binary beats.
proportion_binary_beats Real 1 Proportion of fully occupied binary beats.
nominal_metrical_onsets Real N Nominal metrical onsets according to metrical annotation and beat track.
diff_nominal_metrical_onsets Real N Difference of onsets to nominal metrical onsets.
abs_diff_nominal_metrical_onsets Real N Absolute difference of onsets to nominal metrical onsets.

Duration Ratios

  • Description: Raw vector of duration ratios as well as duration ratio classification (-1: shorter, 0: equal, +1: longer) using thresholds 1.49 and .81.
  • Feature Definition File Name: DURRATIO_HIST
  • Category: Rhythm

Included Features:

Label Type Length Description
dur_ratios Float N-2 Raw output of duration ratios.
dur_ratio_classes Integer [-1:1] N-2 Classified duration ratios.

Inter-onset Interval Class Single Features

  • Description: Normalzed histogram counts, mode,entropy and Zipf coefficent for all 5 duration classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2). Reference time-span is either local beat duration (“relative mode”) or 0.5 sec (“absolute mode”).
  • Feature Definition File Name: IOICLASS_FEATURES
  • Category: Rhythm

Included Features:

Label Type Length Description
ioiclass_abs_hist_01_very_short Real 1 Relative frequency very short (absolute mode).
ioiclass_abs_hist_02_short Real 1 Relative frequency short (absolute mode).
ioiclass_abs_hist_03_medium Real 1 Relative frequency medium (absolute mode).
ioiclass_abs_hist_04_long Real 1 Relative frequency long (absolute mode).
ioiclass_abs_hist_05_very_long Real 1 Relative frequency very long (absolute mode).
ioiclass_rel_hist_01_very_short Real 1 Relative frequency very short (relative mode).
ioiclass_rel_hist_02_short Real 1 Relative frequency short (relative mode).
ioiclass_rel_hist_03_medium Real 1 Relative frequency medium (relative mode).
ioiclass_rel_hist_04_long Real 1 Relative frequency long (relative mode).
ioiclass_rel_hist_05_very_long Real 1 Relative frequency very long (relative mode).
ioiclass_abs_mode Real Var Mode of absolute inter-onset interval class distributions (can be list).
ioiclass_abs_entropy Real 1 Entropy of absolute inter-onset interval class distributions.
ioiclass_abs_zipf Real 1 Zipf coefficient of absolute durations class distributions.
ioiclass_rel_mode Real Var Mode of relative inter-onset interval classdistributions (can be list).
ioiclass_rel_entropy Real 1 Entropy of relative inter-onset interval class distributions.
ioiclass_rel_zipf Real 1 Zipf coefficient of relative inter-onset interval class distributions.

Inter-onset Interval Class Distribution

  • Description: Histogram and raw vector of IOI classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2). Reference time-span is either local beat duration (“relative mode”) or 0.5 sec (“absolute mode”).
  • Feature Definition File Name: IOICLASS_HIST
  • Category: Rhythm

Included Features:

Label Type Length Description
ioiclass_abs_raw Integer N Raw output of IOI classes of all melody notes (absolute mode).
ioiclass_abs_hist_values Integer [-2:2] 5 Histogram bin values (absolute mode).
ioiclass_abs_hist_dens Integer 5 Histogram bin counts (absolute mode).
ioiclass_rel_raw Integer N Raw output of IOI classes of all melody notes (relative mode).
ioiclass_rel_hist_values Integer [-2:2] 5 Histogram bin values (relative mode).
ioiclass_rel_hist_counts Integer 5 Histogram bin counts (relative mode).

Event Densities

  • Description: Events per second and per bar.
  • Feature Definition File Name: GENERAL_EVENT_DENSITY
  • Category: Rhythm

Included Features:

Label Type Length Description
event_density Real 1 Event density (notes per seconds).
total_duration Real 1 Total duration in seconds.
metrical_event_density Real 1 Event density (notes per bar). WARNING: Might not be working properly with meter changes.
total_duration_bar Real 1 Total duration in fractional bar units (e.g., 6 Quarter notes in 4/4 timing are 1.5 bars long).
number_notes Integer 1 Number of events.

Duration Class Single Features

  • Description: Normalized histogram counts, mode, entropy and Zipf cofficient for distribution if all 5 duration classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2) in relative and absolute mode. Reference duration is either local beat duration (‘relative mode’) or 0.5 sec (‘absolute mode’).
  • Feature Definition File Name: DURCLASS_FEATURES
  • Category: Rhythm

Included Features:

Label Type Length Description
durclass_abs_hist_01_very_short Real 1 Relative frequency very short.
durclass_abs_hist_02_short Real 1 Relative frequency short.
durclass_abs_hist_03_medium Real 1 Relative frequency medium.
durclass_abs_hist_04_long Real 1 Relative frequency long.
durclass_abs_hist_05_very_long Real 1 Relative frequency very long.
durclass_rel_hist_01_very_short Real 1 Relative frequency very short.
durclass_rel_hist_02_short Real 1 Relative frequency short.
durclass_rel_hist_03_medium Real 1 Relative frequency medium.
durclass_rel_hist_04_long Real 1 Relative frequency long.
durclass_rel_hist_05_very_long Real 1 Relative frequency very long.
durclass_abs_mode Real Var Mode of absolute duration class distributions (can be list).
durclass_abs_entropy Real 1 Entropy of absolute duration class distributions.
durclass_abs_zipf Real 1 Zipf coefficient of absolute duration class distributions.
durclass_rel_mode Real Var Mode of relative duration class distributions (can be list).
durclass_rel_entropy Real 1 Entropy of relative duration class distributions.
durclass_rel_zipf Real 1 Zipf coefficient of relative duration class distributions.

Duration Class Distribution

  • Description: Histogram and raw vector of duration classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2). Reference duration is either local beat duration (‘relative mode’) or 0.5 sec (‘absolute mode’).
  • Feature Definition File Name: DURCLASS_HIST
  • Category: Rhythm

Included Features:

Label Type Length Description
durclass_abs_raw Integer N Raw output of duration classes (absolute mode).
durclass_abs_hist_values Integer [-2:2] 5 Histogram bin values (absolute mode).
durclass_abs_hist_counts Integer 5 Histogram bin counts (abs. frequencies, absolute mode).
durclass_rel_raw Integer N Raw output of duration classes (relative mode).
durclass_rel_hist_values Integer [-2:2] 5 Histogram bin values (relative mode).
durclass_rel_hist_counts Integer 5 Histogram bin counts (abs. frequencies, relative mode).

Variability Indices (IOI)

  • Description: Normalized pairwise variability index, which measures the mean contrast of adjacents inter-onset intervals. The formula is \frac{1}{N-1}\sum_{i=1}^{N-1} 2\left| \frac{I_{i+1}-I_{i}}{I_{i+1}+I_{i}}\right|. It is the mean of the absolute value of ratios of IOI difference to IOI mean. This index originated in linguistic and is sometimes used by rhythm researchers as a variability measure. The coefficient of variation is another common index for variability. It is defined as the ratio of standard deviation and mean. A simple biased estimator is the corresponding ratio of sample standard deviation and sample mean. The coefficient of variation is only sensical for ratio scales, which applies to durations since they only assum positive values. The formula is: c_V = \frac{\sigma(I_{i})}{\mu(I_{i})}.
  • Feature Definition File Name: VI_IOI
  • Category: Rhythm

Included Features:

Label Type Length Description
nPVI_ioi Real [0,1] 1 Normalized pairwise variability index.
CV_ioi Real 1 Coefficient of variation.

IOI Ratios

  • Description: Raw vector of IOI ratios as well as IOI ratio classification (-1: shorter, 0: equal, +1: longer) using thresholds 1.49 and .81.
  • Feature Definition File Name: IOIRATIO_HIST
  • Category: Rhythm

Included Features:

Label Type Length Description
ioi_ratios Float N-2 Raw output of IOI ratios.
ioi_ratio_classes Integer [-1:1] N-2 Classified IOI ratios.

Tempo Features

  • Description: Statistical descriptors of tempo distribution (mean, stddev, min, max, range) for inter-beat intervals (IBI) measured in seconds or BPM counts (beats per minute). BPM=60/IBI.
  • Feature Definition File Name: GENERAL_TEMPO
  • Category: Rhythm

Included Features:

Label Type Length Description
mean_tempo Real 1 Mean tempo of a melody measured in seconds (average IBI).
std_tempo Real 1 Standard devitation of IBI distribution in seconds.
ibi_range Real 1 Range of IBI distribution (max-min) in seconds.
ibi_min Real 1 Minimum of IBI distribution (min) in seconds.
ibi_max Real 1 Maximum of IBI distribution (max) in seconds.
mean_tempo_bpm Real 1 Mean tempo of a melody measured in BPM (beats per minute, BPM = 60/IBI).
std_tempo_bpm Real 1 Standard devitation of BPM distribution.
bpm_range Real 1 Range (max.min) of BPM distribution.
bpm_min Real 1 Minimum of BPM distribution.
bpm_max Real 1 Maximum of BPM distribution.
beat_track Real Var Raw annotated beat positions.

Sequence/Interval

Interval Class Bigrams

  • Description: Histogram features for intervals classes (IC) bigrams: all unique IC bigrams and all IC bigrams occuring at least twice.
  • Feature Definition File Name: INTCLASS_2GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
ic_bigram_values Array of Integer Var Histogram bins (all unique IC bigrams).
ic_bigram_counts Integer Var Histogram bin counts (all unique IC bigrams).
ic_bigram_patterns Array of Integer Var Histogram bins (all unique IC bigrams occuring at least twice).
ic_bigram_pattern_counts Integer Var Histogram bin counts (all unique IC bigrams occuring at least twice).
ic_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
ic_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Fuzzy Interval Trigrams

  • Description: Histogram features for fuzzy intervals (FUZZYINT, also known as refined contour) trigrams: all unique FUZZYINT trigrams and all FUZZYINT trigrams occuring at least twice.
  • Feature Definition File Name: FUZZYINT_3GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
fuzzyint_trigram_values Array of Integer [-4:4] Var Histogram bins (all unique FUZZYINT trigrams).
fuzzyint_trigram_counts Integer Var Histogram bin count (all unique FUZZYINT trigrams).
fuzzyint_trigram_pattern Array of Integer [-4:4] Var Histogram bins (all unique FUZZYINT trigrams occuring at least twice).
fuzzyint_trigram_pattern_counts Integer Var Histogram bin count (all unique FUZZYINT trigrams occuring at least twice).

Average Run-lengths of arpeggios

  • Description: Run-lengths and means of arpeggios with intervals of size +3/+4 or -3/-4 (fuzzy interval classes leap up/down.).
  • Feature Definition File Name: ARPEGGIOS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_length_arpeggio_ascending Real 1 Mean length of ascending arpeggios.
mean_length_arpeggio_descending Real 1 Mean length of descending arpeggios.
mean_length_arpeggio Real 1 Mean length of arpeggios (mixed up/down thirds).

Average Run-lengths of Chromatic passages

  • Description: Run-lengths and means of chromatic sequences with intervals of size +1 or -1.
  • Feature Definition File Name: STEP_SEQUENCES
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_length_step_ascending Real 1 Mean length of ascending step passages.
mean_length_step_descending Real 1 Mean length of descending step passages.
mean_length_step Real 1 Mean length of step passages (up and down steps).

Fuzzy Interval Bigrams

  • Description: Histogram features for fuzzy intervals (FUZZYINT, also known as refined contour) bigrams: all unique FUZZYINT bigrams and all FUZZYINT bigrams occuring at least twice.
  • Feature Definition File Name: FUZZYINT_2GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
fuzzyint_bigram_values Array of Integer [-4:4] Var Histogram bins (all unique FUZZYINT bigrams).
fuzzyint_bigram_counts Integer Var Histogram bin count (all unique FUZZYINT bigrams).
fuzzyint_bigram_pattern Array of Integer [-4:4] Var Histogram bins (all unique FUZZYINT bigrams occuring at least twice).
fuzzyint_bigram_pattern_counts Integer Var Histogram bin count (all unique FUZZYINT bigrams occuring at least twice).
fuzzyint_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
fuzzyint_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Average Run-lengths of Chromatic Passages

  • Description: Run-lengths and means of chromatic sequences with intervals of size +1 or -1.
  • Feature Definition File Name: CHROMATIC_SEQUENCES
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_length_chromatic_ascending Real 1 Mean length of ascending chromatic passages.
mean_length_chromatic_descending Real 1 Mean length of descending chromatic passages.
mean_length_chromatic_mixed Real 1 Mean length of chromatic passages (up and down combined).

Parson’s Code Bigrams

  • Description: Histogram features for Parson’s Code (PARSON) bigrams: all unique PARSON bigrams and all PARSON bigrams occuring at least twice.
  • Feature Definition File Name: PARSON_2GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
parson_bigrams_values Array of Integer [-1:+1] Var Histogram bins (all unique PARSON bigrams).
parson_bigrams_hist Integer Var Histogram bin count (all unique PARSON bigrams).
parson_bigrams_pattern Array of Integer [-1:+1] Var Histogram bins (all unique PARSON bigrams occuring at least twice).
parson_bigrams_pat_freq Integer Var Histogram bin count (all unique PARSON bigrams occuring at least twice).
parsons_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
parsons_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Percentage of Chromatic Passages

  • Description: Computes the percentage of chromatic passages (interval sequences with values +1 or -1 with at least three notes), in the set of all passages (sequences of constant interval).
  • Feature Definition File Name: INT_CHROMATIC_SEQUENCES_RATIO
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
ratio_chromatic_sequences Real 1 Percentage of of chromatic passages.

Average Run-length of Parson’s Code Segments

  • Description: Means of run-length of segments of notes with constant interval direction.
  • Feature Definition File Name: PARSON_CONST_DIRECTION_AV_LEN
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_segment_length_constant_interval_direction Real 1 Mean run-length of unisone sequences.
mean_segment_length_constant_positive_interval_direction Real 1 Mean run-length of strictly ascending sequences.
mean_segment_length_constant_negative_interval_direction Real 1 Mean run-length of strictly descending sequences.

Average Run-length of Chromatic Passages

  • Description: Mean run-length of sequences with intervals of size +1 or -1.
  • Feature Definition File Name: INT_CHROMATIC_SEQUENCES_AV_LEN
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_length_chromatic_sequences Real 1 Mean run-length of chromatic passages.

Average Run-lengths of Parsons Sequences

  • Description: Run-lengths and means of sequences of all (Parson’s) intervals directions.
  • Feature Definition File Name: PARSON_SEQUENCES
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
mean_length_seq_ascending Real 1 Mean length of ascending passages.
mean_length_seq_descending Real 1 Mean length of descending passages.
mean_length_seq_repitition Real 1 Mean length of repetitions.

Semitone Interval Trigrams

  • Description: Histogram features for semitone intervals (INT) trigrams: all unique INT trigrams and all INT trigrams occuring at least twice.
  • Feature Definition File Name: INT_3GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
int_trigram_values Array of Integer Var Histogram bins (all unique INT trigrams).
int_trigram_counts Integer Var Histogram bin counts (all unique INT trigrams).
int_trigram_patterns Array of Integer Var Histogram bins (all unique INT trigrams occuring at least twice).
int_trigram_pattern_counts Integer Var Histogram bin count (all unique INT trigrams occuring at least twice).

Parson’s Code Trigrams

  • Description: Histogram features for Parson’s Code (PARSON) trigrams: all unique PARSON trigrams and all PARSON trigrams occuring at least twice.
  • Feature Definition File Name: PARSON_3GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
parson_3grams_values Array of Integer [-1:+1] Var Histogram bins (all unique PARSON trigrams).
parson_3grams_hist Integer Var Histogram bin count (all unique PARSON trigrams).
parson_3grams_pattern Array of Integer [-1:+1] Var Histogram bins (all unique PARSON trigrams occuring at least twice).
parson_3grams_pat_freq Integer Var Histogram bin count (all unique PARSON trigrams occuring at least twice).

Semitone Interval Bigrams

  • Description: Histogram features for semitone intervals (INT) bigrams: all unique INT bigrams and all INT bigrams occuring at least twice.
  • Feature Definition File Name: INT_2GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
int_bigram_values Array of Integer Var Histogram bins (all unique INT bigrams).
int_bigram_counts Integer Var Histogram bin counts (all unique INT bigrams).
int_bigram_patterns Array of Integer Var Histogram bins (all unique INT bigrams occuring at least twice).
int_bigram_pattern_counts Integer Var Histogram bin counts (all unique INT bigrams occuring at least twice).
int_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
int_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Interval Class Trigrams

  • Description: Histogram features for interval class (IC) trigrams: all unique IC trigrams and all IC trigrams occuring at least twice.
  • Feature Definition File Name: INTCLASS_3GRAMS
  • Category: Sequence/Interval

Included Features:

Label Type Length Description
ic_trigram_values Array of Integer Var Histogram bins (all unique IC trigrams).
ic_trigram_counts Integer Var Histogram bin counts (all unique IC trigrams).
ic_trigram_patterns Array of Integer Var Histogram bins (all unique IC trigrams occuring at least twice).
ic_trigram_pattern_counts Integer Var Histogram bin count (all unique IC trigrams occuring at least twice).

Sequence/Pitch

Pitch Class Bigrams

  • Description: Histogram features for pitch class (PC) bigrams: all unique PC bigrams and all PC bigrams occuring at least twice.
  • Feature Definition File Name: PC_2GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
pc_2grams_values Array of Integer [0:11] Var Histogram bins (all unique PC bigrams).
pc_2grams_hist Integer Var Histogram bin count (all unique PC bigrams).
pc_2grams_pattern Array of Integer [0:11] Var Histogram bins (all unique PC bigrams occuring at least twice).
pc_2grams_pat_freq Integer Var Histogram bin count (all unique PC bigrams occuring at least twice).
pc_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
pc_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Chordal Diatonic Pitch Class Trigrams

  • Description: Histogram features for Chordal Diatonic Pitch Class (CDPC) trigrams: all unique CDPC trigrams and all CDPC trigrams occuring at least twice.
  • Feature Definition File Name: CDPC_3GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
cdpc_trigrams List of Strings Var Histogram bins (all unique CDPC 3-grams).
cdpc_trigram_counts Integer Var Histogram bin count (all unique CDPC 3-grams).
cdpc_trigram_pattern List of String Var Histogram bins (all unique CDPC 3-grams occuring at least twice).
cdpc_trigram_pattern_counts Integer Var Histogram bin count (all unique CDPC 3-grams occuring at least twice).

Chordal Diatonic Pitch Class Bigrams

  • Description: Histogram features for Chordal Diatonic Pitch Class (CDPC) bigrams: all unique CDPC bigrams and all CDPC bigrams occuring at least twice.
  • Feature Definition File Name: CDPC_2GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
cdpc_bigrams String Var Histogram bins (all unique CDPC bigrams).
cdpc_bigram_counts String Var Histogram bin counts (all unique CDPC bigrams).
cdpc_bigram_patterns String Var Histogram bins (all unique CDPC bigrams occuring at least twice).
cdpc_bigram_pattern_counts Integer Var Histogram bin counts (all unique CDPC bigrams occuring at least twice).
cdpc_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
cdpc_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Pitch Class Trigrams

  • Description: Histogram features for pitch class (PC) trigrams: all unique PC trigrams and all PC trigrams occuring at least twice.
  • Feature Definition File Name: PC_3GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
pc_3grams_values Array of Integer [0:11] Var Histogram bins (all unique PC trigrams).
pc_3grams_hist Integer Var Histogram bin count (all unique PC trigrams).
pc_3grams_pattern Array of Integer [0:11] Var Histogram bins (all unique PC trigrams occuring at least twice).
pc_3grams_pat_freq Integer Var Histogram bin count (all unique PC trigrams occuring at least twice).

Tonal Pitch Class Bigrams

  • Description: Histogram features for raw MIDI pitch (TPC) bigrams: all unique TPC bigrams and all TPC bigrams occuring at least twice.
  • Feature Definition File Name: TPC_2GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
tpc_bigrams_values Array of Integer [0:11] Var Histogram bins (all unique TPC bigrams).
tpc_bigrams_hist Integer Var Histogram bin count (all unique TPC bigrams).
tpc_bigrams_pattern Array of Integer [0:11] Var Histogram bins (all unique TPC bigrams occuring at least twice).
tpc_bigrams_pat_freq Integer Var Histogram bin count (all unique TPC bigrams occuring at least twice).
tpc_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
tpc_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Pitch Bigrams

  • Description: Histogram features for raw MIDI pitch (PITCH) bigrams: all unique PITCH bigrams and all PITCH bigrams occuring at least twice.
  • Feature Definition File Name: PITCH_2GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
pitch_bigrams_values Array of Integer [0:127] Var Histogram bins (all unique PITCH bigrams).
pitch_bigrams_hist Integer Var Histogram bin count (all unique PITCH bigrams).
pitch_bigrams_pattern Array of Integer [0:127] Var Histogram bins (all unique PITCH bigrams occuring at least twice).
pitch_bigrams_pat_freq Integer Var Histogram bin count (all unique PITCH bigrams occuring at least twice).
pitch_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
pitch_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Pitch Pattern Lengths and Distances

  • Description: Average length and occurence distance between exact pitch patterns. Uses the edit distance based self-similarity matrix alogorithm.
  • Feature Definition File Name: PITCH_PATTERN_DIST_LEN_FEATURES
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
mean_pattern_length Real 1 Mean length of exact pitch patterns.
mean_pattern_distance Real 1 Mean distance between exact pitch patterns.

Pitch Trigrams

  • Description: Histogram features for raw MIDI pitch (PITCH) trigrams: all unique PITCH trigrams and all PITCH trigrams occuring at least twice.
  • Feature Definition File Name: PITCH_3GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
pitch_3grams_values Array of Integer [0:127] Var Histogram bins (all unique PITCH trigrams).
pitch_3grams_hist Integer Var Histogram bin count (all unique PITCH trigrams).
pitch_3grams_pattern Array of Integer [0:127] Var Histogram bins (all unique PITCH trigrams occuring at least twice).
pitch_3grams_pat_freq Integer Var Histogram bin count (all unique PITCH trigrams occuring at least twice).

Chordal Pitch Class Bigrams

  • Description: Histogram features for Chordal Pitch Class (CPC) bigrams: all unique CPC bigrams and all CPC bigrams occuring at least twice.
  • Feature Definition File Name: CPC_2GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
cpc_bigrams String Var Histogram bins (all unique CPC bigrams).
cpc_bigram_counts String Var Histogram bin counts (all unique CPC bigrams).
cpc_bigram_patterns String Var Histogram bins (all unique CPC bigrams occuring at least twice).
cpc_bigram_pattern_counts Integer Var Histogram bin counts (all unique CPC bigrams occuring at least twice).
cpc_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
cpc_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Tonal Pitch Class Trigrams

  • Description: Histogram features for tonal pitch class (TPC) trigrams: all unique TPC trigrams and all TPC trigrams occuring at least twice.
  • Feature Definition File Name: TPC_3GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
tpc_3grams_values Array of Integer [0:11] Var Histogram bins (all unique TPC trigrams).
tpc_3grams_counts Integer Var Histogram bin count (all unique TPC trigrams).
tpc_3grams_pattern Array of Integer [0:11] Var Histogram bins (all unique TPC trigrams occuring at least twice).
tpc_3grams_pattern_counts Integer Var Histogram bin count (all unique TPC trigrams occuring at least twice).

Chordal Pitch Class Trigrams

  • Description: Histogram features for Chordal Pitch Class (CPC) trigrams: all unique CPC trigrams and all CPC trigrams occuring at least twice.
  • Feature Definition File Name: CPC_3GRAMS
  • Category: Sequence/Pitch

Included Features:

Label Type Length Description
cpc_trigrams List of Strings Var Histogram bins (all unique CPC 3-grams).
cpc_trigram_counts Integer Var Histogram bin count (all unique CPC 3-grams).
cpc_trigram_pattern List of String Var Histogram bins (all unique CPC 3-grams occuring at least twice).
cpc_trigram_pattern_counts Integer Var Histogram bin count (all unique CPC 3-grams occuring at least twice).

Sequence/Rhythm

Duration Class Bigrams

  • Description: Histogram features for (absolute) duration classes (DURCLASS) bigrams: all unique DURCLASS bigrams and all DURCLASS bigrams occuring at least twice.
  • Feature Definition File Name: DURCLASS_2GRAMS
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
durclass_bigrams String Var Histogram bins (all unique DURCLASS bigrams).
durclass_bigram_counts Integer Var Histogram bin count (all unique DURCLASS bigrams).
durclass_bigram_pattern String Var Histogram bins (all unique DURCLASS bigrams occuring at least twice).
durclass_bigram_pattern_count Integer Var Histogram bin count (all unique DURCLASS bigrams occuring at least twice).
durclass_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
durclass_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Duration Class Trigrams

  • Description: Histogram features for (absolute) duration classes (DURCLASS) trigrams: all unique DURCLASS trigrams and all DURCLASS trigrams occuring at least twice. Reference time-span is 0.5s.
  • Feature Definition File Name: DURCLASS_3GRAMS
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
durclass_trigram_values String Var Histogram bins (all unique DURCLASS trigrams).
durclass_trigram_hist Integer Var Histogram bin count (all unique DURCLASS trigrams).
durclass_trigram_pattern String Var Histogram bins (all unique DURCLASS trigrams occuring at least twice).
durclass_trigram_pattern_counts Integer Var Histogram bin count (all unique DURCLASS trigrams occuring at least twice).

Duration Class Run-lengths

  • Description: Mean run-length of duration classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2). Reference time-span is either local beat duration (“relative mode”) or 0.5 sec (“absolute mode”).
  • Feature Definition File Name: DURCLASS_RUNLENGTH
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
durclass_mean_seg_len Real 1 Mean run-length averaged over all classes.
mean_seg_len_01_very_short Real 1 Mean run-length for very short events.
mean_seg_len_02_short Real 1 Mean run-length for short events.
mean_seg_len_03_medium Real 1 Mean run-length for medium long events.
mean_seg_len_04_long Real 1 Mean run-length for long events.
mean_seg_len_05_very_long Real 1 Mean run-length for very long events.
seg_len Integer Var Raw vector of run-lengths for all classes.

Inter-onset Interval Class Bigrams

  • Description: Histogram features for rel. IOI classes (IOICLASS) bigrams: all unique IOICLASS bigrams and all IOICLASS bigrams occuring at least twice.
  • Feature Definition File Name: IOICLASS_2GRAMS
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
ioiclass_bigrams String Var Histogram bins (all unique IOICLASS bigrams).
ioiclass_bigram_counts Integer Var Histogram bin count (all unique IOICLASS bigrams).
ioiclass_bigram_pattern String Var Histogram bins (all unique IOICLASS bigrams occuring at least twice).
ioiclass_bigram_pattern_count Integer Var Histogram bin count (all unique IOICLASS bigrams occuring at least twice).
ioiclass_bigram_entropy Real 1 Unnormalized Entropy of bigram distribution in bits.
ioiclass_bigram_entropy_norm Real [0,1] 1 Normalized Entropy of bigram distribution.

Inter-onset Interval Class Run-lengths

  • Description: Mean run-length of inter-onset interval classes (very short: -2, short: -1, medium: 0, long: 1, very long: 2). Reference time-span is either local beat duration (“relative mode”) or 0.5 sec (“absolute mode”).
  • Feature Definition File Name: IOICLASS_RUNLENGTH
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
aic_mean_seg_len Real 1 Mean run-length averaged over all classes (absolute mode).
aic_mean_seg_len_01_very_short Real 1 Mean run-length for very short events (absolute mode).
aic_mean_seg_len_02_short Real 1 Mean run-length for short events (absolute mode).
aic_mean_seg_len_03_medium Real 1 Mean run-length for medium long events (absolute mode).
aic_mean_seg_len_04_long Real 1 Mean run-length for long events (absolute mode).
aic_mean_seg_len_05_very_long Real 1 Mean run-length for very long events (absolute mode).
aic_seg_len Integer Var Raw vector of run-lengths for all classes (absolute mode).
ric_mean_seg_len Real 1 Mean run-length averaged over all classes (relative mode).
ric_mean_seg_len_01_very_short Real 1 Mean run-length for very short events (relative mode).
ric_mean_seg_len_02_short Real 1 Mean run-length for short events (relative mode).
ric_mean_seg_len_03_medium Real 1 Mean run-length for medium long events (relative mode).
ric_mean_seg_len_04_long Real 1 Mean run-length for long events (relative mode).
ric_mean_seg_len_05_very_long Real 1 Mean run-length for very long events (relative mode).
ric_seg_len Integer Var Raw vector of run-lengths for all classes (relative mode).

Inter-onset Interval Class Trigrams

  • Description: Histogram features for (absolute) IOI classes (IOICLASS) trigrams: all unique IOICLASS trigrams and all IOICLASS trigrams occuring at least twice. Reference time-span is 0.5s.
  • Feature Definition File Name: IOICLASS_3GRAMS
  • Category: Sequence/Rhythm

Included Features:

Label Type Length Description
ioiclass_trigram_values String Var Histogram bins (all unique IOICLASS trigrams).
ioiclass_trigram_hist Integer Var Histogram bin count (all unique IOICLASS trigrams).
ioiclass_trigram_pattern String Var Histogram bins (all unique IOICLASS trigrams occuring at least twice).
ioiclass_trigram_pattern_counts Integer Var Histogram bin count (all unique IOICLASS trigrams occuring at least twice).

Structure

Form Structure of Phrases

  • Description: Calculates form strings based on self-similarity between phrases using edit distance based on interval and duration classes.
  • Feature Definition File Name: GENERAL_PHRASE_FORM
  • Category: Structure

Included Features:

Label Type Length Description
form_by_interval String 1 Phrase form structure based on self-similarity matrix of interval edit distances.
form_by_durclass String 1 Phrase form structure based on self-similarity matrix of duration class edit distances.

Self-similarity Features

  • Description: Features derived from self similarity-matrix between phrases. These are basically statistical descriptors (max, median, stddev, range, entropy) either of the set of adjacent or non-adjacent cells in the self-similarity matrix. For instance, a high median of adjacent cells could indicate developing variation, the same value for non-adjacent cells hints at long-range correlations, i.e. form thinking.
  • Feature Definition File Name: GENERAL_SELF_SIM_MATRIX_PHRASES_FEATURES
  • Category: Structure

Included Features:

Label Type Length Description
adjacent_phrase_similarity_max Real [0,1] 1 Maximum of adjacent similarity values.
adjacent_phrase_similarity_median Real [0,1] 1 Median of adjacent similarity values.
adjacent_phrase_similarity_std Real [0,1] 1 Standard deviaation of adjacent similarity values.
adjacent_phrase_similarity_range Real [0,1] 1 Range of adjacent similarity values.
adjacent_phrase_similarity_entropy Real [0,1] 1 Entropy of adjacent similarity values.
nonadjacent_phrase_similarity_median Real [0,1] 1 Median of non-adjacent similarity values.
nonadjacent_phrase_similarity_std Real [0,1] 1 Standard deviation of non-adjacent similarity values.
nonadjacent_phrase_similarity_range Real [0,1] 1 Range of non-adjacent similarity values.
nonadjacent_phrase_similarity_entropy Real [0,1] 1 Maximum of adjacent similarity values.

Self-similarity Matrix of Phrases

  • Description: Similarity between phrases using edit distance based similarity on pitch values.
  • Feature Definition File Name: GENERAL_SELF_SIM_MATRIX_PHRASES
  • Category: Structure

Included Features:

Label Type Length Description
ssm Real [0,1] N_P \times N_P Interval-based self-similarity matrix of phrases.
ssm_thresh Real [0,1] N_P \times N_P Thresholded interval-based self-similarity matrix of phrases.

Tone Formation

Loudness

  • Description: Loudness (resp. intensity) related values for single tone events have been extracted from jazz solos by means of sophisticated MIR techniques and are stored in the WJazzD SQLITE database. Currently, five different values are available: Maximum, median and standard deviation of loudness, relative position of loudness peak and temporal centroid. Might not be avaibable for every solo.
  • Feature Definition File Name: GENERAL_LOUDNESS
  • Category: Tone Formation

Included Features:

Label Type Length Description
loudness_max Float [-120, 0] N Maximum loudness values per tone.
loudness_med Real [-120, 0] N Median loudness values per tone.
loudness_sd Real N Standard deviation of loudness per tone.
loudness_rel_peak_pos Real [0, 1] N Relative position of maximum loudness value, duration of tone equals 1.
loudness_temp_cent Real [0, 1] N Normed temporal centroid of loudness curve per tone, duration of tone equals 1.
loudness_s2b Real N Ratio of solo loudness (median) to backing track loudness (median).

Articulation

  • Description: Articulation of notes as measured by the ratio of duration and inter-onset interval. (Be careful at phrase endings, though). The higher the ratio, the more legato, the lower the ratio, the more staccato.
  • Feature Definition File Name: GENERAL_ARTICULATION
  • Category: Tone Formation

Included Features:

Label Type Length Description
ioi_raw Real N-1 Raw interonset-interval values.
duration_raw Real N Raw durations.
articulation_raw Real N-1 Ratios of durations to interonset-intervals.
art_mean Real 1 Mean of articulation distribution.
art_median Real 1 Median of articulation distribution.
art_var Real 1 Variance of of articulation distribution.
art_std Real 1 Standard deviation of articulation distribution.
art_min Real 1 Minimum of of articulation distribution.
art_max Real 1 Maximum of of articulation distribution.
art_range Real 1 Range of of articulation distribution.

Modulation

  • Description: F0-modulation of notes. Possible annoated values are fall-off, slide, vibrato, bend, straight, or an empty string, which might indicate that either no modulation is present or that no articulation was annotated. There are furthermore three automatically extracted value available: f0_mod_range_cents (Modulations in cents), f0_mod_freq_hz (Modulation frequency in Hz), f0_median_dev (Median devation from nominal 12TET pitch in cents).
  • Feature Definition File Name: GENERAL_F0MODULATION
  • Category: Tone Formation

Included Features:

Label Type Length Description
modulation String N Annonated modulation per tone.
modulation_short String N Annonated modulation per tone (short label).
f0_mod_range_cents Real N Extracted modulation range (cents).
f0_mod_freq Real N Extracted modulation range (Hz).
f0_median_dev Real N Extracted median deviation from nominal 12TET pitch (cents).