douglasdemoura / asymut

ASyMuT - Automatic System for Music Transcription

ASyMuT - Automatic System for Music Transcription (and analysis)

ASyMuT is a system for automatic music transcription (i.e., WAV to MIDI) and analysis.

Preliminary conventions:

n : natural number
z : integer number
r : "real" number (actually decimal <~ floating-point)
r+ : non-negative "real" number
s : contiguous string of characters without spaces
p : one among many predefined strings for the given parameter (multiple-choice or "combo")
< > : mandatory parameter
[ ] : optional parameter
[ ] ... [ ] : variable number of optional parameters

Mandatory parameters:

Parameter: --stdft_win_len <n>
Abbreviation: -w <n>
Description: Short-Time Discrete Fourier Transform Window Length (a.k.a Frame Size)
Associate variable: unsigned int fft_prop->window->length
Observations: It is possible to use the “k”“ multiplier (e.g. “-w 4k” instead of “-w 4096”)

Parameter: --input_file <s>
Abbreviation: -i <s>
Description: Input File Name
Associate variable: char stream->file_name[FILENAME_MAX]

Optional parameters:

Parameter: --stdft_apod_fun <p> [r] ... [r]
Abbreviation: -a <p>
Description: STDFT Apodization Function (a.k.a tapering or smoothing function)
Associate variable: Apodization_Function APOD_FUN
Possible values: rectangular, triangular, hamming, hann, blackman, blackman-harris, nuttall3..nuttall12 , nuttall14..nuttall15, gaussian, hanning-poisson, helie_a_w1, helie_a_w6

Parameter: --stdft_win_step <n>
Abbreviation: -s <n>
Description: STDFT Window Step (a.ka. window hop or frame stride)
Associate variable: unsigned int fft_prop->step

Parameter: --zero_padding_ratio <n>
Abbreviation: -z <n>
Description: Zero Padding Ratio
Associate variable: unsigned short fft_prop->zero_padding_ratio

Parameter: --min_abs_f0 <r+>
Description: Minimum Absolute F0 (in Hertz)
Associate variable: double MIN_ABSOLUTE_F0

Parameter: --min_rel_f0 <r+>
Description: Minimum Relative F0 (in Frequency Resolution units)
Associate variable: double MIN_F0_TO_FREQ_RES_RATIO
Observations: setting to 0 disables this constraint

Parameter: --max_abs_f0 <r+>
Description: Maximum Absolute F0 (in Hertz)
Associate variable: double MAX_ABSOLUTE_F0

Parameter: --max_bands <n>
Description: Maximum Critical Bands
Associate variable: unsigned char MAX_CRITICAL_BANDS

Parameter: --min_interonset_gap <r+>
Description: Minimum Interonset Gap (i.e., Minimum Distance Between Onsets) (in seconds)
Associate variable: double MIN_DISTANCE_BETWEEN_ONSETS

Parameter: --min_onset_win <r+>
Description: Onset Threshold Window Minimum Length (in seconds)
Associate variable: double ONSET_THRESHOLD_ WINDOW_ MIN_LENGTH

Parameter: --max_onset_win <r+>
Description: Onset Threshold Window Maximum Length (in seconds)
Associate variable: double ONSET_THRESHOLD_ WINDOW_ MAX_LENGTH

Parameter: --percentile <r+>
Abbreviation: -p <r+>
Description: Onset Threshold Percentile (in per-unit)
Associate variable: double ONSET_THRESHOLD_PERCENTILE
Observação: Rounded to nearest discrete possibility

Parameter: --scale_factor <r>
Abbreviation: -f <r>
Description: Onset Threshold Percentile Scaling Factor
Associate variable: double ONSET_THRESHOLD_PERCENTILE_SCALING_FACTOR
Observação: threshold = const_part + scale_factor * percentile(threshold_win)

Parameter: --const_part <r>
Abbreviation: -c <r>
Description: Onset Threshold Constant Part
Associate variable: double ONSET_THRESHOLD_CONSTANT_PART

Parameter: --max_delay <r+>
Description: Maximum After Onset Delay Before F0 Evidence (in seconds)
Associate variable: double MAX_AFTER_ONSET_DELAY_BEFORE_F0_EVIDENCE

Parameter: --max_gap <r+>
Description: Maximum F0 Evidence Gap Inside Note (in seconds)
Associate variable: double MAX_F0_EVIDENCE_GAP_INSIDE_NOTE

Parameter: --min_duration <r+>
Description: Minimum Note Duration (i.e., minimum note evidence time)
Associate variable: double MIN_NOTE_DURATION

Parameter: --ref_a4_freq <r+>
Description: A4 Reference Frequency (in Hertz)
Associate variable: double FREQ_REF_A4

Parameter: --pow_norm_level
Abbreviation: -n
Description: Power Normalization Level (in dB)
Associate variable: double POWER_NORMALIZATION_LEVEL

Parameter: --lowest_note <n>
Description: Pitch Range Lowest Note
Associate variable: unsigned char PITCH_RANGE_LOWEST_NOTE

Parameter: --highest_note <n>
Description: Pitch Range Highest Note
Associate variable: unsigned char PITCH_RANGE_HIGHEST_NOTE

Parameter: --f0_method <p> [p] [r] ... [r]
Description: F0 Estimation Method
Associate variable: F0_Estimation_Method ESTIMATION_METHOD
Possible values: (max_index, hps, hsc, bw_hsc, fft_fft) [mag|pow|lp|wd], klapuri

Parameter: --unpred_method <p> [r] ... [r]
Abbreviation: -u
Description: Unpredictability Estimation Method
Associate variable: Unpredictability_Estimation_Method UNPRED_METHOD
Possible values: complex, lp, ilp, sam, ph, new

Parameter: --klap_ic <r+>
Description: Klapuri Multiple F0 Estimation Iteration Control parameter
Associate variable: double KLAPURI_ITERATION_CONTROL

Parameter: --piano_roll
Description: Print Notes as Gnuplot Arrows (piano-roll like)
Associate variable: PRINT_NOTES_AS_GNUPLOT_ARROWS

Parameters that change the system purpose

Associate variable: Asymut_Operation_Mode OPERATION_MODE

Parameter: --transcribe
Description: Transcribe (i.e., default system operation mode)

Parameter: --print <p> | <p> <p> [p]
Description: Prints
Possible values: samples, apod_win_time, apod_win_freq, <spec|loc_max> <ph|uph|mag|pow|lp|wd|pn [ doubled_linebreaks]>, unpred, threshold, bands_response, bands_response_summary, bands_sum, onset, f0_estimate

Parameter: --ruminate <p>
Description: Ruminates (process but don’t print anything, essencialy for benchmarking)
Possible values: spec, unpred, f0_estimate

References

Undergraduate thesis

In Portuguese

Sistema Automático de Transcrição Melódica

Conference proceedings

In Portuguese

Um Sistema Automático de Transcrição Melódica conference proceeding

In English

Slides for the conference proceedings

A System for the Automatic Transcription of Melodies

Conference proceedings, focus on the Pitch Determination Algorithm (PDA)

Accurate and Efficient Fundamental Frequency Determination from Precise Partial Estimates

Related project

PDA extracted (only Hann window + Grandke), no external dependencies, input = samples (not audio files)

pda