Source code for music21.audioSearch

# -----------------------------------------------------------------------------
# Name:         audioSearch.py
# Purpose:      base subroutines for all audioSearching and score following
#               routines
#
# Authors:      Jordi Bartolome
#               Michael Scott Asato Cuthbert
#
# Copyright:    Copyright © 2011-2020 Michael Scott Asato Cuthbert
# License:      BSD, see license.txt
# -----------------------------------------------------------------------------
'''
Base routines used throughout audioSearching and score-following.

Requires numpy and matplotlib.  Installing scipy makes the process faster
and more accurate using FFT convolve.
'''
from __future__ import annotations

__all__ = [
    'AudioSearchException',
    'autocorrelationFunction',
    'decisionProcess',
    'detectPitchFrequencies',
    'getFrequenciesFromAudioFile',
    'getFrequenciesFromMicrophone',
    'getFrequenciesFromPartialAudioFile',
    'histogram',
    'interpolation',
    'joinConsecutiveIdenticalPitches',
    'normalizeInputFrequency',
    'notesAndDurationsToStream',
    'pitchFrequenciesToObjects',
    'prepareThresholds',
    'quantizeDuration',
    'quarterLengthEstimation',
    'recording',
    'scoreFollower',
    'smoothFrequencies',
    'transcriber',
]

import copy
import math
import pathlib
import wave
import warnings
import unittest

from typing import cast

# cannot call this base, because when audioSearch.__init__.py
# imports * from base, it overwrites audioSearch!
from music21 import base
from music21 import environment
from music21 import exceptions21
from music21 import features
from music21 import metadata
from music21 import note
from music21 import pitch
from music21 import scale
from music21 import stream

from music21.audioSearch import recording
from music21.audioSearch import transcriber

environLocal = environment.Environment('audioSearch')

audioChunkLength = 1024
recordSampleRate = 44100


[docs] def histogram(data, bins): # noinspection PyShadowingNames ''' Partition the list in `data` into a number of bins defined by `bins` and return the number of elements in each bin and a set of `bins` + 1 elements where the first element (0) is the start of the first bin, the last element (-1) is the end of the last bin, and every remaining element (i) is the dividing point between one bin and another. >>> data = [1, 1, 4, 5, 6, 0, 8, 8, 8, 8, 8] >>> outputData, bins = audioSearch.histogram(data, 8) >>> print(outputData) [3, 0, 0, 1, 1, 1, 0, 5] >>> bins [0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] >>> print([int(b) for b in bins]) [0, 1, 2, 3, 4, 5, 6, 7, 8] >>> outputData, bins = audioSearch.histogram(data, 4) >>> print(outputData) [3, 1, 2, 5] >>> print([int(b) for b in bins]) [0, 2, 4, 6, 8] ''' maxValue = max(data) minValue = min(data) lengthEachBin = (maxValue - minValue) / bins container = [] for i in range(int(bins)): container.append(0) for i in data: count = 1 while i > minValue + count * lengthEachBin: count += 1 container[count - 1] += 1 binsLimits = [minValue] count = 1 for i in range(int(bins)): binsLimits.append(minValue + count * lengthEachBin) count += 1 return container, binsLimits
[docs] def autocorrelationFunction(recordedSignal, recordSampleRateIn): # noinspection PyShadowingNames ''' Converts the temporal domain into a frequency domain. To do that, it uses the autocorrelation function, which finds periodicities in the signal in the temporal domain and, consequently, finds the frequency in each instant of time. >>> import wave >>> import numpy # you need to have numpy, scipy, and matplotlib installed to use this >>> wv = wave.open(str(common.getSourceFilePath() / ... 'audioSearch' / 'test_audio.wav'), 'r') >>> data = wv.readframes(1024) >>> samples = numpy.frombuffer(data, dtype=numpy.int16) >>> finalResult = audioSearch.autocorrelationFunction(samples, 44100) >>> wv.close() >>> print(finalResult) 143.6276... ''' # noinspection PyProtectedMember if 'numpy' in (bmi := base._missingImport): # len(_missingImport) > 0: raise AudioSearchException( 'Cannot run autocorrelationFunction without ' f'numpy installed (scipy recommended). Missing {bmi}') import numpy convolve = None try: with warnings.catch_warnings(): # scipy.signal gives ImportWarning warnings.simplefilter('ignore', ImportWarning) # numpy warns scipy that oldnumeric will be dropped soon. warnings.simplefilter('ignore', DeprecationWarning) # noinspection PyPackageRequirements from scipy.signal import fftconvolve as convolve # type: ignore except ImportError: # pragma: no cover warnings.warn('Running convolve without scipy -- will be slower', stacklevel=2) convolve = numpy.convolve recordedSignal = numpy.array(recordedSignal) correlation = convolve(recordedSignal, recordedSignal[::-1], mode='full') lengthCorrelation = len(correlation) // 2 correlation = correlation[lengthCorrelation:] difference = numpy.diff(correlation) # Calculates the difference between slots positiveDifferences = numpy.where(difference > 0)[0] if len(positiveDifferences) == 0: finalResult = 10 # Rest else: beginning = positiveDifferences[0] peak = numpy.argmax(correlation[beginning:]) + beginning vertex = interpolation(correlation, peak) finalResult = recordSampleRateIn / vertex return finalResult
[docs] def prepareThresholds(useScale=None): # noinspection PyShadowingNames ''' returns two elements. The first is a list of threshold values for one octave of a given scale, `useScale`, (including the octave repetition) (Default is a ChromaticScale). The second is the pitches of the scale. A threshold value is the fractional part of the log-base-2 value of the frequency. For instance if A = 440 and B-flat = 460, then the threshold between A and B-flat will be 450. Notes below 450 should be considered As and those above 450 should be considered B-flats. Thus, the list returned has one less element than the number of notes in the scale + octave repetition. If useScale is a ChromaticScale, `prepareThresholds` will return a 12-element list. If it's a diatonic scale, it'll have 7 elements. >>> pitchThresholds, pitches = audioSearch.prepareThresholds(scale.MajorScale('A3')) >>> for i in range(len(pitchThresholds)): ... print(f'{pitches[i]} < {pitchThresholds[i]:.2f} < {pitches[i + 1]}') A3 < 0.86 < B3 B3 < 0.53 < C#4 C#4 < 0.16 < D4 D4 < 0.28 < E4 E4 < 0.45 < F#4 F#4 < 0.61 < G#4 G#4 < 1.24 < A4 ''' if useScale is None: useScale = scale.ChromaticScale('C4') scPitches = useScale.pitches scPitchesRemainder = [] for p in scPitches: pLog2 = math.log2(p.frequency) scPitchesRemainder.append(math.modf(pLog2)[0]) scPitchesRemainder[-1] += 1 scPitchesThreshold = [] for i in range(len(scPitchesRemainder) - 1): scPitchesThreshold.append((scPitchesRemainder[i] + scPitchesRemainder[i + 1]) / 2) return scPitchesThreshold, scPitches
[docs] def interpolation(correlation, peak): # noinspection PyShadowingNames ''' Interpolation for estimating the true position of an inter-sample maximum when nearby samples are known. Correlation is a vector and peak is an index for that vector. Returns the x coordinate of the vertex of that parabola. >>> import numpy >>> f = [2, 3, 1, 6, 4, 2, 3, 1] >>> peak = int(numpy.argmax(f)) >>> peak # f[3] is 6, which is the max. 3 >>> audioSearch.interpolation(f, peak) 3.21428571... ''' if peak in (0, len(correlation) - 1): return peak vertex = (correlation[peak - 1] - correlation[peak + 1]) / ( correlation[peak - 1] - 2.0 * correlation[peak] + correlation[peak + 1]) vertex = vertex * 0.5 + peak return vertex
[docs] def normalizeInputFrequency(inputPitchFrequency, thresholds=None, pitches=None): # noinspection PyShadowingNames ''' Takes in an inputFrequency, a set of threshold values, and a set of allowable pitches (given by prepareThresholds) and returns a tuple of the normalized frequency and the pitch detected (as a :class:`~music21.pitch.Pitch` object) It will convert the frequency to be within the range of the default frequencies (usually C4 to C5), but the pitch object will have the correct octave. >>> audioSearch.normalizeInputFrequency(441.72) (440.0, <music21.pitch.Pitch A4>) If you will be doing this often, it's best to cache your thresholds and pitches by running `prepareThresholds` once first: >>> thresholds, pitches = audioSearch.prepareThresholds(scale.ChromaticScale('C4')) >>> for fq in [450, 510, 550, 600]: ... print(audioSearch.normalizeInputFrequency(fq, thresholds, pitches)) (440.0, <music21.pitch.Pitch A4>) (523.25113..., <music21.pitch.Pitch C5>) (277.18263..., <music21.pitch.Pitch C#5>) (293.66476..., <music21.pitch.Pitch D5>) ''' if ( (thresholds is None and pitches is not None) or (thresholds is not None and pitches is None) ): raise AudioSearchException( 'Cannot normalize input frequency if thresholds are given and ' + 'pitches are not, or vice versa') if thresholds is None: (thresholds, pitches) = prepareThresholds() inputPitchLog2 = math.log2(inputPitchFrequency) (remainder, octave) = math.modf(inputPitchLog2) octave = int(octave) for i in range(len(thresholds)): threshold = thresholds[i] if remainder < threshold: returnPitch = copy.deepcopy(pitches[i]) returnPitch.octave = octave - 4 # PROBLEM # returnPitch.inputFrequency = inputPitchFrequency name_note = pitch.Pitch(str(pitches[i])) return name_note.frequency, returnPitch # else: # above highest threshold returnPitch = copy.deepcopy(pitches[-1]) returnPitch.octave = octave - 3 returnPitch.inputFrequency = inputPitchFrequency name_note = pitch.Pitch(str(pitches[-1])) return name_note.frequency, returnPitch
[docs] def pitchFrequenciesToObjects(detectedPitchesFreq, useScale=None): # noinspection PyShadowingNames ''' Takes in a list of detected pitch frequencies and returns a tuple where the first element is a list of :class:~`music21.pitch.Pitch` objects that best match these frequencies and the second element is a list of the frequencies of those objects that can be plotted for matplotlib TODO: only return the former. The latter can be generated in other ways. >>> readPath = common.getSourceFilePath() / 'audioSearch' / 'test_audio.wav' >>> freqFromAQList = audioSearch.getFrequenciesFromAudioFile(waveFilename=readPath) >>> detectedPitchesFreq = audioSearch.detectPitchFrequencies( ... freqFromAQList, useScale=scale.ChromaticScale('C4')) >>> detectedPitchesFreq = audioSearch.smoothFrequencies(detectedPitchesFreq) >>> (detectedPitchObjects, listPlot) = audioSearch.pitchFrequenciesToObjects( ... detectedPitchesFreq, useScale=scale.ChromaticScale('C4')) >>> [str(p) for p in detectedPitchObjects] ['A5', 'A5', 'A6', 'D6', 'D4', 'B4', 'A4', 'F4', 'E-4', 'C#3', 'B3', 'B3', 'B3', 'A3', 'G3',...] ''' if useScale is None: useScale = scale.MajorScale('C4') detectedPitchObjects = [] (thresholds, pitches) = prepareThresholds(useScale) for i in range(len(detectedPitchesFreq)): inputPitchFrequency = detectedPitchesFreq[i] unused_freq, pitch_name = normalizeInputFrequency(inputPitchFrequency, thresholds, pitches) detectedPitchObjects.append(pitch_name) listPlot = [] i = 0 while i < len(detectedPitchObjects) - 1: name = detectedPitchObjects[i].name hold = i tot_octave = 0 while i < len(detectedPitchObjects) - 1 and detectedPitchObjects[i].name == name: tot_octave = tot_octave + detectedPitchObjects[i].octave i = i + 1 tot_octave = round(tot_octave / (i - hold)) for j in range(i - hold): detectedPitchObjects[hold + j - 1].octave = tot_octave listPlot.append(detectedPitchObjects[hold + j - 1].frequency) return detectedPitchObjects, listPlot
[docs] def getFrequenciesFromMicrophone(length=10.0, storeWaveFilename=None): ''' records for length (=seconds) a set of frequencies from the microphone. If storeWaveFilename is not None, then it will store the recording on disk in a wave file. Returns a list of frequencies detected. TODO -- find a way to test or at least demo ''' # noinspection PyProtectedMember if 'numpy' in base._missingImport: raise AudioSearchException( 'Cannot run getFrequenciesFromMicrophone without numpy installed') import numpy environLocal.printDebug('* start recording') storedWaveSampleList = recording.samplesFromRecording(seconds=length, storeFile=storeWaveFilename, recordChunkLength=audioChunkLength) environLocal.printDebug('* stop recording') freqFromAQList = [] for data in storedWaveSampleList: samples = numpy.frombuffer(data, dtype=numpy.int16) freqFromAQList.append(autocorrelationFunction(samples, recordSampleRate)) return freqFromAQList
[docs] def getFrequenciesFromAudioFile(waveFilename='xmas.wav'): ''' gets a list of frequencies from a complete audio file. Each sample is a window of audioSearch.audioChunkLength long. >>> audioSearch.audioChunkLength 1024 >>> readPath = common.getSourceFilePath() / 'audioSearch' / 'test_audio.wav' >>> freq = audioSearch.getFrequenciesFromAudioFile(waveFilename=readPath) >>> print(freq) [143.627..., 99.083..., 211.004..., 4700.313..., ...] ''' # noinspection PyProtectedMember if 'numpy' in base._missingImport: raise AudioSearchException( 'Cannot run getFrequenciesFromAudioFile without numpy installed') import numpy storedWaveSampleList = [] environLocal.printDebug('* reading entire file from disk') try: wv = wave.open(str(waveFilename), 'r') except IOError: raise AudioSearchException(f'Cannot open {waveFilename} for reading, does not exist') # modify it to read the entire file for i in range(int(wv.getnframes() / audioChunkLength)): data = wv.readframes(audioChunkLength) storedWaveSampleList.append(data) freqFromAQList = [] for data in storedWaveSampleList: samples = numpy.frombuffer(data, dtype=numpy.int16) freqFromAQList.append(float(autocorrelationFunction(samples, recordSampleRate))) wv.close() return freqFromAQList
[docs] def getFrequenciesFromPartialAudioFile(waveFilenameOrHandle='temp', length=10.0, startSample=0): ''' It calculates the fundamental frequency at every instant of time of an audio signal extracted either from the microphone or from an already recorded song. It uses a period of time defined by the variable "length" in seconds. It returns a list with the frequencies, a variable with the file descriptor, and the end sample position. >>> #_DOCS_SHOW readFile = 'pachelbel.wav' >>> sp = common.getSourceFilePath() #_DOCS_HIDE >>> readFile = sp / 'audioSearch' / 'test_audio.wav' #_DOCS_HIDE >>> fTup = audioSearch.getFrequenciesFromPartialAudioFile(readFile, length=1.0) >>> frequencyList, pachelbelFileHandle, currentSample = fTup >>> for frequencyIndex in range(5): ... print(frequencyList[frequencyIndex]) 143.627... 99.083... 211.004... 4700.313... 767.827... >>> print(currentSample) # should be near 44100, but probably not exact 44032 Now read the next 1 second: >>> fTup = audioSearch.getFrequenciesFromPartialAudioFile(pachelbelFileHandle, length=1.0, ... startSample=currentSample) >>> frequencyList, pachelbelFileHandle, currentSample = fTup >>> for frequencyIndex in range(5): ... print(frequencyList[frequencyIndex]) 187.798... 238.263... 409.700... 149.958... 101.989... >>> print(currentSample) # should be exactly double the previous 88064 ''' # noinspection PyProtectedMember if 'numpy' in base._missingImport: raise AudioSearchException( 'Cannot run getFrequenciesFromPartialAudioFile without numpy installed') import numpy if waveFilenameOrHandle == 'temp': waveFilenameOrHandle = environLocal.getRootTempDir() / 'temp.wav' if isinstance(waveFilenameOrHandle, pathlib.Path): waveFilenameOrHandle = str(waveFilenameOrHandle) if isinstance(waveFilenameOrHandle, str): # waveFilenameOrHandle is a filename waveFilename = waveFilenameOrHandle try: # noinspection PyUnusedLocal waveHandle = wave.open(waveFilename, 'r') except IOError: raise AudioSearchException(f'Cannot open {waveFilename} for reading, does not exist') else: # waveFilenameOrHandle is a file handle waveHandle = cast(wave.Wave_read, waveFilenameOrHandle) storedWaveSampleList = [] environLocal.printDebug('* reading file from disk a part of the song') for i in range(int(math.floor(length * recordSampleRate / audioChunkLength))): startSample = startSample + audioChunkLength if startSample < waveHandle.getnframes(): data = waveHandle.readframes(audioChunkLength) storedWaveSampleList.append(data) freqFromAQList = [] for data in storedWaveSampleList: samples = numpy.frombuffer(data, dtype=numpy.int16) freqFromAQList.append(autocorrelationFunction(samples, recordSampleRate)) endSample = startSample return (freqFromAQList, waveHandle, endSample)
[docs] def detectPitchFrequencies(freqFromAQList, useScale=None): # noinspection PyShadowingNames ''' Detects the pitches of the notes from a list of frequencies, using thresholds which depend on the useScale option. If useScale is None, the default value is the Major Scale beginning C4. Returns the frequency of each pitch after normalizing them. >>> freqFromAQList=[143.627689055, 99.0835452019, 211.004784689, 4700.31347962, 2197.9431119] >>> cMaj = scale.MajorScale('C4') >>> pitchesList = audioSearch.detectPitchFrequencies(freqFromAQList, useScale=cMaj) >>> for i in range(5): ... print(int(round(pitchesList[i]))) 147 98 220 4699 2093 ''' if useScale is None: useScale = scale.MajorScale('C4') (thresholds, pitches) = prepareThresholds(useScale) detectedPitchesFreq = [] for i in range(len(freqFromAQList)): # to find thresholds and frequencies inputPitchFrequency = freqFromAQList[i] unused_freq, pitch_name = normalizeInputFrequency(inputPitchFrequency, thresholds, pitches) detectedPitchesFreq.append(pitch_name.frequency) return detectedPitchesFreq
[docs] def smoothFrequencies( frequencyList: list[int|float], *, smoothLevels=7, inPlace=False ) -> list[int]|None: ''' Smooths the shape of the signal to avoid false detections in the fundamental frequency. Takes in a list of ints or floats. The second pitch below is certainly too low. It will be smoothed out: >>> inputPitches = [440, 220, 440, 440, 442, 443, 441, 470, 440, 441, 440, ... 442, 440, 440, 440, 397, 440, 440, 440, 442, 443, 441, ... 440, 440, 440, 440, 440, 442, 443, 441, 440, 440] >>> result = audioSearch.smoothFrequencies(inputPitches) >>> result [409, 409, 409, 428, 435, 438, 442, 444, 441, 441, 441, 441, 434, 433, 432, 431, 437, 438, 439, 440, 440, 440, 440, 440, 440, 441, 441, 441, 440, 440, 440, 440] Original list is unchanged: >>> inputPitches[1] 220 Different levels of smoothing have different effects. At smoothLevel=2, the isolated 220hz sample is pulling down the surrounding samples: >>> audioSearch.smoothFrequencies(inputPitches, smoothLevels=2)[:5] [330, 275, 358, 399, 420] Doing this enough times will smooth out a lot of inconsistencies. >>> audioSearch.smoothFrequencies(inputPitches, smoothLevels=28)[:5] [432, 432, 432, 432, 432] If inPlace is True then the list is modified in place and nothing is returned: >>> audioSearch.smoothFrequencies(inputPitches, inPlace=True) >>> inputPitches[:5] [409, 409, 409, 428, 435] Note that `smoothLevels=1` is the baseline that does nothing: >>> audioSearch.smoothFrequencies(inputPitches, smoothLevels=1) == inputPitches True And less than 1 raises a ValueError: >>> audioSearch.smoothFrequencies(inputPitches, smoothLevels=0) Traceback (most recent call last): ValueError: smoothLevels must be >= 1 There cannot be more smoothLevels than input frequencies: >>> audioSearch.smoothFrequencies(inputPitches, smoothLevels=40) Traceback (most recent call last): ValueError: There cannot be more smoothLevels (40) than inputPitches (32) Note that the system runs on O(smoothLevels * len(frequenciesList)), so additional smoothLevels can be costly on a large set. This function always returns a list of ints -- rounding to the nearest hertz (you did want it smoothed right?) * Changed in v6: inPlace defaults to False (like other music21 functions) and if done in Place, returns nothing. smoothLevels and inPlace became keyword only. ''' if smoothLevels < 1: raise ValueError('smoothLevels must be >= 1') numFreqs = len(frequencyList) if smoothLevels > numFreqs: raise ValueError( f'There cannot be more smoothLevels ({smoothLevels}) than inputPitches ({numFreqs})' ) dpf = frequencyList detectedPitchesFreq: list[float] if inPlace: detectedPitchesFreq = [float(f) for f in dpf] else: detectedPitchesFreq = [float(f) for f in copy.copy(dpf)] # smoothing beginning = 0.0 ends = 0.0 for i in range(smoothLevels): beginning = beginning + detectedPitchesFreq[i] ends = ends + detectedPitchesFreq[numFreqs - 1 - i] beginning = beginning / smoothLevels ends = ends / smoothLevels for i in range(numFreqs): # TODO: replace this O(i*smoothLevels) routine with an O(i) routine if i < int(math.floor(smoothLevels / 2.0)): detectedPitchesFreq[i] = int(beginning) elif i > numFreqs - int(math.ceil(smoothLevels / 2.0)) - 1: detectedPitchesFreq[i] = int(ends) else: change = 0.0 for j in range(smoothLevels): change += detectedPitchesFreq[i + j - int(math.floor(smoothLevels / 2.0))] detectedPitchesFreq[i] = change / smoothLevels out: list[int] = [int(round(f)) for f in detectedPitchesFreq] if not inPlace: return out else: for i in range(len(frequencyList)): frequencyList[i] = out[i] return None
# ------------------------------------------------------ # Duration related routines
[docs] def joinConsecutiveIdenticalPitches(detectedPitchObjects): # noinspection PyShadowingNames ''' takes a list of equally spaced :class:`~music21.pitch.Pitch` objects and returns a tuple of two lists, the first a list of :class:`~music21.note.Note` or :class:`~music21.note.Rest` objects (each of quarterLength 1.0) and a list of how many were joined together to make that object. N.B. the returned list is NOT a :class:`~music21.stream.Stream`. >>> readPath = common.getSourceFilePath() / 'audioSearch' / 'test_audio.wav' >>> freqFromAQList = audioSearch.getFrequenciesFromAudioFile(waveFilename=readPath) >>> chrome = scale.ChromaticScale('C4') >>> detectedPitchesFreq = audioSearch.detectPitchFrequencies(freqFromAQList, useScale=chrome) >>> detectedPitchesFreq = audioSearch.smoothFrequencies(detectedPitchesFreq) >>> (detectedPitches, listPlot) = audioSearch.pitchFrequenciesToObjects( ... detectedPitchesFreq, useScale=chrome) >>> len(detectedPitches) 861 >>> notesList, durationList = audioSearch.joinConsecutiveIdenticalPitches(detectedPitches) >>> len(notesList) 24 >>> print(notesList) [<music21.note.Rest quarter>, <music21.note.Note C>, <music21.note.Note C>, <music21.note.Note D>, <music21.note.Note E>, <music21.note.Note F>, <music21.note.Note G>, <music21.note.Note A>, <music21.note.Note B>, <music21.note.Note C>, ...] >>> print(durationList) [71, 6, 14, 23, 34, 40, 27, 36, 35, 15, 17, 15, 6, 33, 22, 13, 16, 39, 35, 38, 27, 27, 26, 8] ''' # initialization REST_FREQUENCY = 10 detectedPitchObjects[0].frequency = REST_FREQUENCY # detecting the length of each note j = 0 good = 0 bad = 0 valid_note = False total_notes = 0 total_rests = 0 notesList = [] durationList = [] while j < len(detectedPitchObjects): fr = detectedPitchObjects[j].frequency # detect consecutive instances of the same frequency while j < len(detectedPitchObjects) and fr == detectedPitchObjects[j].frequency: good = good + 1 # if more than 6 consecutive identical samples, it might be a note if good >= 6: valid_note = True # if we've gone 15 or more samples without getting something constant, # assume it's a rest if bad >= 15: durationList.append(bad) total_rests = total_rests + 1 notesList.append(note.Rest()) bad = 0 j = j + 1 if valid_note: durationList.append(good) total_notes = total_notes + 1 # doesn't this unnecessarily create a note that it doesn't need? # notesList.append(detectedPitchObjects[j - 1].frequency) should work n = note.Note() n.pitch = detectedPitchObjects[j - 1] notesList.append(n) else: bad = bad + good good = 0 valid_note = False j = j + 1 return notesList, durationList
[docs] def quantizeDuration(length): ''' round an approximately transcribed quarterLength to a better one in music21. Should be replaced by a full-featured routine in midi or stream. See :meth:`~music21.stream.Stream.quantize` for more information on the standard music21 methodology. >>> audioSearch.quantizeDuration(1.01) 1.0 >>> audioSearch.quantizeDuration(1.70) 1.5 ''' length = length * 100 typicalLengths = [25.00, 50.00, 100.00, 150.00, 200.00, 400.00] thresholds = [] for i in range(len(typicalLengths) - 1): thresholds.append((typicalLengths[i] + typicalLengths[i + 1]) / 2) finalLength = typicalLengths[0] for i in range(len(thresholds)): threshold = thresholds[i] if length > threshold: finalLength = typicalLengths[i + 1] return finalLength / 100
[docs] def quarterLengthEstimation(durationList, mostRepeatedQuarterLength=1.0): # noinspection PyShadowingNames ''' takes a list of lengths of notes (measured in audio samples) and tries to estimate what the length of a quarter note should be in this list. If mostRepeatedQuarterLength is another number, it still returns the estimated length of a quarter note, but chooses it so that the most common note in durationList will be the other note. See example 2: Returns a float -- and not an int. >>> durationList = [20, 19, 10, 30, 6, 21] >>> audioSearch.quarterLengthEstimation(durationList) 20.625 Example 2: suppose these are the inputted durations for a score where most of the notes are half notes. Show how long a quarter note should be: >>> audioSearch.quarterLengthEstimation(durationList, mostRepeatedQuarterLength=2.0) 10.3125 ''' dl = copy.copy(durationList) dl.append(0) pdf, bins = histogram(dl, 8.0) # environLocal.printDebug(f' HISTOGRAM {pdf} {bins}') i = len(pdf) - 1 # backwards! it has more sense while pdf[i] != max(pdf): i = i - 1 qle = (bins[i] + bins[i + 1]) / 2.0 if mostRepeatedQuarterLength == 0: mostRepeatedQuarterLength = 1.0 binPosition = 0 - math.log2(mostRepeatedQuarterLength) qle = qle * math.pow(2, binPosition) # it normalizes the length to a quarter note # environLocal.printDebug('QUARTER ESTIMATION') # environLocal.printDebug(f' bins {bins} ') # environLocal.printDebug(f' pdf {pdf}') # environLocal.printDebug(f' quarterLengthEstimate {qle}') return qle
[docs] def notesAndDurationsToStream( notesList, durationList, scNotes=None, removeRestsAtBeginning=True, qle=None ): # noinspection PyShadowingNames ''' take a list of :class:`~music21.note.Note` objects or rests and an equally long list of how long each one lasts in terms of samples and returns a Stream using the information from quarterLengthEstimation and quantizeDurations. returns a :class:`~music21.stream.Score` object, containing a metadata object and a single :class:`~music21.stream.Part` object, which in turn contains the notes, etc. Does not run :meth:`~music21.stream.base.Stream.makeNotation` on the Score. >>> durationList = [20, 19, 10, 30, 6, 21] >>> n = note.Note >>> noteList = [n('C#4'), n('D5'), n('B4'), n('F#5'), n('C5'), note.Rest()] >>> s,lengthPart = audioSearch.notesAndDurationsToStream(noteList, durationList) >>> s.show('text') {0.0} <music21.metadata.Metadata object at ...> {0.0} <music21.stream.Part ...> {0.0} <music21.note.Note C#> {1.0} <music21.note.Note D> {2.0} <music21.note.Note B> {2.5} <music21.note.Note F#> {4.0} <music21.note.Note C> {4.25} <music21.note.Rest quarter> ''' # rounding lengths p2 = stream.Part() # If the score is available, the quarter estimation is better: # It could take into account the changes of tempo during the song, but it # would take more processing time if scNotes is not None: fe = features.native.MostCommonNoteQuarterLength(scNotes) mostCommon = fe.extract().vector[0] qle = quarterLengthEstimation(durationList, mostCommon) elif scNotes is None: # this is for the transcriber qle = quarterLengthEstimation(durationList) for i in range(len(durationList)): actualDuration = quantizeDuration(durationList[i] / qle) notesList[i].quarterLength = actualDuration if not (removeRestsAtBeginning and (notesList[i].name == 'rest')): p2.append(notesList[i]) removeRestsAtBeginning = False sc = stream.Score() sc.metadata = metadata.Metadata() sc.metadata.title = 'Automatic Music21 Transcription' sc.insert(0, p2) if scNotes is None: # Case transcriber return sc, len(p2) else: # case follower return sc, qle
[docs] def decisionProcess( partsList, notePrediction, beginningData, lastNotePosition, countdown, firstNotePage=None, lastNotePage=None ): # noinspection PyShadowingNames ''' Decides which of the given parts of the score has the best match with the recorded part of the song. If there is not a part of the score with a high probability to be the correct part, it starts a "countdown" in order stop the score following if the bad matching persists. In this case, it does not match the recorded part of the song with any part of the score. Inputs: partsList, contains all the possible parts of the score, sorted from the higher probability to be the best matching at the beginning to the lowest probability. notePrediction is the position of the score in which the next note should start. beginningData is a list with all the beginnings of the used fragments of the score to find the best matching. lastNotePosition is the position of the score in which the last matched fragment of the score finishes. Countdown is a counter of consecutive errors in the matching process. Outputs: Returns the beginning of the best matching fragment of score and the countdown. >>> scNotes = corpus.parse('luca/gloria').parts[0].flatten().notes.stream() >>> scoreStream = scNotes >>> sfp = common.getSourceFilePath() #_DOCS_HIDE >>> readPath = sfp / 'audioSearch' / 'test_audio.wav' #_DOCS_HIDE >>> freqFromAQList = audioSearch.getFrequenciesFromAudioFile(waveFilename=readPath) #_DOCS_HIDE >>> tf = 'test_audio.wav' >>> #_DOCS_SHOW freqFromAQList = audioSearch.getFrequenciesFromAudioFile(waveFilename=tf) >>> chrome = scale.ChromaticScale('C4') >>> detectedPitchesFreq = audioSearch.detectPitchFrequencies(freqFromAQList, useScale=chrome) >>> detectedPitchesFreq = audioSearch.smoothFrequencies(detectedPitchesFreq) >>> (detectedPitches, listPlot) = audioSearch.pitchFrequenciesToObjects( ... detectedPitchesFreq, useScale=chrome) >>> (notesList, durationList) = audioSearch.joinConsecutiveIdenticalPitches(detectedPitches) >>> transcribedScore, qle = audioSearch.notesAndDurationsToStream(notesList, durationList, ... scNotes=scNotes, qle=None) >>> hop = 6 >>> tn_recording = 24 >>> totScores = [] >>> beginningData = [] >>> lengthData = [] >>> for i in range(4): ... excerpt = scoreStream[i * hop + 1:i * hop + tn_recording + 1] ... scNotes = stream.Part(excerpt) ... name = str(i) ... beginningData.append(i * hop + 1) ... lengthData.append(tn_recording) ... scNotes.id = name ... totScores.append(scNotes) >>> listOfParts = search.approximateNoteSearch(transcribedScore.flatten().notes.stream(), ... totScores) >>> notePrediction = 0 >>> lastNotePosition = 0 >>> countdown = 0 >>> positionInList, countdown = audioSearch.decisionProcess( ... listOfParts, notePrediction, beginningData, lastNotePosition, countdown) >>> print(positionInList) 0 The countdown result is 1 because the song used is completely different from the score!! >>> print(countdown) 1 ''' i = 0 position = 0 while i < len(partsList) and beginningData[int(partsList[i].id)] < notePrediction: i = i + 1 position = i if len(partsList) == 1: # it happens when you don't play anything during a recording period position = 0 dist = math.fabs(beginningData[0] - notePrediction) for i in range(len(partsList)): positionBeginningData = beginningData[int(partsList[i].id)] if ((partsList[i].matchProbability >= 0.9 * partsList[0].matchProbability) and (positionBeginningData > lastNotePosition)): # let's take a 90% if math.fabs(positionBeginningData - notePrediction) < dist: dist = math.fabs(positionBeginningData - notePrediction) position = i positionBeginningData = beginningData[int(partsList[position].id)] # print('ERRORS', position, len(partsList), lastNotePosition, # partsList[position].matchProbability , positionBeginningData) if position < len(partsList) and positionBeginningData <= lastNotePosition: environLocal.printDebug(f' error ? {positionBeginningData}, {lastNotePosition}') if partsList[position].matchProbability < 0.6 or len(partsList) == 1: # the latter for the all-rest case environLocal.printDebug('Are you sure you are playing the right song?') countdown = countdown + 1 elif dist > 20 and countdown == 0: countdown += 1 environLocal.printDebug(f'Excessive distance? {dist=}') elif dist > 30 and countdown == 1: countdown += 1 environLocal.printDebug(f'Excessive distance? {dist=}') elif ((firstNotePage is not None and lastNotePage is not None) and ((positionBeginningData < firstNotePage or positionBeginningData > lastNotePage) and countdown < 2)): countdown += 1 environLocal.printDebug('playing in a not shown part') else: countdown = 0 environLocal.printDebug(['****????**** DECISION PROCESS: dist from expected:', dist, 'beginning data:', positionBeginningData, 'lastNotePos', lastNotePosition]) return position, countdown
class AudioSearchException(exceptions21.Music21Exception): pass # ----------------------------------------- class Test(unittest.TestCase): pass # ------------------------------------------------------------------------------ # define presented order in documentation _DOC_ORDER: list[type] = [] if __name__ == '__main__': import music21 music21.mainTest(Test)