diff --git a/test/src/unittests/extractor/test_tonalextractor.py b/test/src/unittests/extractor/test_tonalextractor.py new file mode 100644 index 000000000..1d8272c93 --- /dev/null +++ b/test/src/unittests/extractor/test_tonalextractor.py @@ -0,0 +1,154 @@ +#!/usr/bin/env python + +# Copyright (C) 2006-2021 Music Technology Group - Universitat Pompeu Fabra +# +# This file is part of Essentia +# +# Essentia is free software: you can redistribute it and/or modify it under +# the terms of the GNU Affero General Public License as published by the Free +# Software Foundation (FSF), either version 3 of the License, or (at your +# option) any later version. +# +# This program is distributed in the hope that it will be useful, but WITHOUT +# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS +# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more +# details. +# +# You should have received a copy of the Affero GNU General Public License +# version 3 along with this program. If not, see http://www.gnu.org/licenses/ +import os.path + +from essentia_test import * +import numpy as np + + +class TestTonalExtractor(TestCase): + + def testEmpty(self): + # Test if the algorithm handles an empty input signal correctly + with self.assertRaises(RuntimeError): + chords_changes_rate, _, _, chords_number_rate, _, _, _, _, _, _, _, key_strength = TonalExtractor()(np.array([], dtype=np.float32)) + + def testSilence(self): + # In this test we check three of the output parameters of type real + silence_vec = np.zeros(44100, dtype=np.single) + chords_changes_rate, _, _, chords_number_rate, _, _, _, _, _, _, _, key_strength = TonalExtractor()(silence_vec) + self.assertEqual(chords_changes_rate, 0.0) + self.assertGreaterEqual(chords_number_rate, 0.0) + self.assertEqual(key_strength, 0.0) + + def testInvalidParameters(self): + # Test if the algorithm handles invalid parameters correctly + extractor = TonalExtractor() + + # Test case 1: Negative frameSize + with self.subTest(msg="Negative frameSize"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=-1, hopSize=2048, tuningFrequency=440.0) + + # Test case 2: Negative hopSize + with self.subTest(msg="Negative hopSize"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=4096, hopSize=-1, tuningFrequency=440.0) + + # Test case 3: Negative tuningFrequency + with self.subTest(msg="Negative tuningFrequency"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=4096, hopSize=2048, tuningFrequency=-440.0) + + # Test case 4: Zero frameSize and hopSize + with self.subTest(msg="Zero frameSize and hopSize"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=0, hopSize=0, tuningFrequency=440.0) + + # Test case 5: Zero frameSize + with self.subTest(msg="Zero frameSize"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=0, hopSize=2048, tuningFrequency=440.0) + + # Test case 6: Zero hopSize + with self.subTest(msg="Zero hopSize"): + with self.assertRaises(RuntimeError): + extractor.configure(frameSize=4096, hopSize=0, tuningFrequency=440.0) + + # Test case 7: Non-negative parameters + with self.subTest(msg="Valid parameters"): + # This should not raise an exception + extractor.configure(frameSize=4096, hopSize=2048, tuningFrequency=440.0) + + def testRandomInput(self): + n = 10 + for _ in range(n): + rand_input = np.random.random(88200).astype(np.single) * 2 - 1 + result = TonalExtractor()(rand_input) + expected_result = np.sum(rand_input * rand_input) ** 0.67 + self.assertAlmostEqual(result[0], expected_result, 9.999e+02) + + def testRegression(self): + frameSizes = [256, 512, 1024, 2048, 4096, 8192] + hopSizes = [128, 256, 512, 1024, 2048, 4096] + + input_filename = join(testdata.audio_dir, "recorded", "dubstep.wav") # Replace 'testdata' with actual path + realAudio = MonoLoader(filename=input_filename)() + + # Iterate through pairs of frameSize and corresponding hopSize + # TODO: Extend loop to try different tuningFrequency values + for fs, hs in zip(frameSizes, hopSizes): + with self.subTest(frameSize=fs, hopSize=hs): + # Process the algorithm on real audio with the current frameSize and hopSize + te = TonalExtractor() + te.configure(frameSize=fs, hopSize=hs) + chords_changes_rate, _, _, chords_number_rate, _, _, _, _, _, _, _, key_strength= te(realAudio) + + # Perform assertions on one or more outputs + # Example: Assert that chords_changes_rate is a non-negative scalar + self.assertIsInstance(chords_changes_rate, (int, float)) + self.assertGreaterEqual(chords_changes_rate, 0) + self.assertIsInstance(chords_number_rate, (int, float)) + self.assertGreaterEqual(chords_number_rate, 0) + self.assertIsInstance(key_strength, (int, float)) + self.assertGreaterEqual(key_strength, 0) + # You can add more assertions on other outputs as needed + + def testRealAudio(self): + + # These reference values could also be compared with the results of tonal extractors of alternative + # audio libraries (e.g. MadMom, libs from Alexander Lerch etc.) + # ccr = chord changes rate ; cnr = chord number rate; ks = key strength + mozart_ccr = 0.03400309011340141 + mozart_cnr = 0.010819165036082268 + mozart_ks = 0.8412253260612488 + + vivaldi_ccr = 0.052405908703804016 + vivaldi_cnr = 0.004764173645526171 + vivaldi_ks = 0.7122617959976196 + + thresh = 0.5 + + def test_on_real_audio(path, ccr, cnr, ks): + realAudio = MonoLoader(filename=path)() + + # Use default configuration of algorothm + # This function could be extended to test for more outputs + # TODO: Extend to test non-scalar and string outputs: + # i.e. chords_histogram, chords_progression, chords_scale, chords_strength + # hpcp, hpcp_highres, key_key and key_scale + te = TonalExtractor() + chords_changes_rate, _, _, chords_number_rate, _, _, _, _, _, _, _, key_strength= te(realAudio) + self.assertIsInstance(chords_changes_rate, (int, float)) + self.assertGreaterEqual(chords_changes_rate, 0) + self.assertAlmostEqual(chords_changes_rate, ccr, thresh) + self.assertIsInstance(chords_number_rate, (int, float)) + self.assertGreaterEqual(chords_number_rate, 0) + self.assertAlmostEqual(chords_number_rate, cnr, thresh) + self.assertIsInstance(key_strength, (int, float)) + self.assertGreaterEqual(key_strength, 0) + self.assertAlmostEqual(key_strength, ks, thresh) + + test_on_real_audio(join(testdata.audio_dir, "recorded", "mozart_c_major_30sec.wav"), mozart_ccr, mozart_cnr, mozart_ks) + test_on_real_audio(join(testdata.audio_dir, "recorded", "Vivaldi_Sonata_5_II_Allegro.wav"), vivaldi_ccr, vivaldi_cnr, vivaldi_ks) + +suite = allTests(TestTonalExtractor) + +if __name__ == '__main__': + unittest.main() diff --git a/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence.npy b/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence.npy new file mode 100644 index 000000000..9ec2a122b Binary files /dev/null and b/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence.npy differ diff --git a/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence_eqloud.npy b/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence_eqloud.npy new file mode 100644 index 000000000..a476e4421 Binary files /dev/null and b/test/src/unittests/tonal/pitchmelodia/pitchmelodiaconfidence_eqloud.npy differ diff --git a/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch.npy b/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch.npy new file mode 100644 index 000000000..0359daa36 Binary files /dev/null and b/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch.npy differ diff --git a/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch_eqloud.npy b/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch_eqloud.npy new file mode 100644 index 000000000..5eb9613ea Binary files /dev/null and b/test/src/unittests/tonal/pitchmelodia/pitchmelodiapitch_eqloud.npy differ diff --git a/test/src/unittests/tonal/test_pitchmelodia.py b/test/src/unittests/tonal/test_pitchmelodia.py index e84af7e9f..f2d9e2a10 100644 --- a/test/src/unittests/tonal/test_pitchmelodia.py +++ b/test/src/unittests/tonal/test_pitchmelodia.py @@ -18,18 +18,187 @@ # version 3 along with this program. If not, see http://www.gnu.org/licenses/ - from essentia_test import * - +import math as math +import numpy as np class TestPitchMelodia(TestCase): - def testZero(self): - signal = zeros(256) + def testInvalidParam(self): + # Test for all the values above the boundary limits. + self.assertConfigureFails(PitchMelodia(), {'binResolution': -1}) + self.assertConfigureFails(PitchMelodia(), {'binResolution': 0}) + self.assertConfigureFails(PitchMelodia(), {'filterIterations': 0}) + self.assertConfigureFails(PitchMelodia(), {'filterIterations': -1}) + self.assertConfigureFails(PitchMelodia(), {'frameSize': -1}) + self.assertConfigureFails(PitchMelodia(), {'frameSize': 0}) + self.assertConfigureFails(PitchMelodia(), {'harmonicWeight': -1}) + self.assertConfigureFails(PitchMelodia(), {'hopSize': 0}) + self.assertConfigureFails(PitchMelodia(), {'hopSize': -1}) + self.assertConfigureFails(PitchMelodia(), {'magnitudeCompression': -1}) + self.assertConfigureFails(PitchMelodia(), {'magnitudeCompression': 0}) + self.assertConfigureFails(PitchMelodia(), {'magnitudeCompression': 2}) + self.assertConfigureFails(PitchMelodia(), {'magnitudeThreshold': -1}) + self.assertConfigureFails(PitchMelodia(), {'maxFrequency': -1}) + self.assertConfigureFails(PitchMelodia(), {'minDuration': 0}) + self.assertConfigureFails(PitchMelodia(), {'minDuration': -1}) + self.assertConfigureFails(PitchMelodia(), {'minFrequency': -1}) + self.assertConfigureFails(PitchMelodia(), {'numberHarmonics': -1}) + self.assertConfigureFails(PitchMelodia(), {'peakDistributionThreshold': -1}) + self.assertConfigureFails(PitchMelodia(), {'peakDistributionThreshold': 2.1}) + self.assertConfigureFails(PitchMelodia(), {'peakFrameThreshold': -1}) + self.assertConfigureFails(PitchMelodia(), {'peakFrameThreshold': 2}) + self.assertConfigureFails(PitchMelodia(), {'pitchContinuity': -1}) + self.assertConfigureFails(PitchMelodia(), {'referenceFrequency': 0}) + self.assertConfigureFails(PitchMelodia(), {'referenceFrequency': -1}) + self.assertConfigureFails(PitchMelodia(), {'sampleRate': 0}) + self.assertConfigureFails(PitchMelodia(), {'sampleRate': -1}) + self.assertConfigureFails(PitchMelodia(), {'timeContinuity': 0}) + self.assertConfigureFails(PitchMelodia(), {'timeContinuity': -1}) + + def testDefaultParameters(self): + signal = np.random.random(1024) pitch, confidence = PitchMelodia()(signal) - self.assertAlmostEqualVector(pitch, [0., 0., 0.]) - self.assertAlmostEqualVector(confidence, [0., 0., 0.]) + # Assert that default parameters produce valid outputs + self.assertIsNotNone(pitch) + self.assertIsNotNone(confidence) + + def testEmptyInput(self): + pitch, confidence = PitchMelodia()([]) + self.assertEqualVector(pitch, []) + self.assertEqualVector(confidence, []) + + def testZerosInput(self): + signal = zeros(1024) + pitch, confidence = PitchMelodia()(signal) + self.assertAlmostEqualVector(pitch, [0.] * 9) + self.assertAlmostEqualVector(confidence, [0.] * 9) + + def testOnesInput(self): + signal = ones(1024) + pitch, confidence = PitchMelodia()(signal) + self.assertAlmostEqualVector(pitch, [0.] * 9) + self.assertAlmostEqualVector(confidence, [0.] * 9) + + def testCustomParameters(self): + signal = np.random.random(2048) + # Use custom parameters + params = { + 'binResolution': 5, + 'filterIterations': 5, + 'frameSize': 1024, + 'guessUnvoiced': True, + 'harmonicWeight': 0.9, + 'hopSize': 256, + 'magnitudeCompression': 0.5, + 'magnitudeThreshold': 30, + 'maxFrequency': 15000, + 'minDuration': 50, + 'minFrequency': 60, + 'numberHarmonics': 15, + 'peakDistributionThreshold': 1.0, + 'peakFrameThreshold': 0.8, + 'pitchContinuity': 30.0, + 'referenceFrequency': 60, + 'sampleRate': 22050, + 'timeContinuity': 150 + } + pitch, confidence = PitchMelodia(**params)(signal) + # Assert that custom parameters produce valid outputs + self.assertIsNotNone(pitch) + self.assertIsNotNone(confidence) + + def testInputWithSilence(self): + rand_signal = np.random.random(512) + signal = np.concatenate([zeros(512), rand_signal, zeros(512)]) + pitch, confidence = PitchMelodia()(signal) + # Assert that silent portions don't have pitch information + self.assertTrue(all(p == 0.0 for p in pitch[:512])) + self.assertTrue(all(c == 0.0 for c in confidence[:512])) + + def testHighPitchResolution(self): + rand_signal = np.random.random(1024) + pitch, confidence = PitchMelodia(binResolution=1)(rand_signal) + # Assert that using high bin resolution produces valid outputs + self.assertIsNotNone(pitch) + self.assertIsNotNone(confidence) + self.assertEqual(len(pitch), 9) + self.assertEqual(len(confidence), 9) + + def testRealCase(self): + filename = join(testdata.audio_dir, 'recorded', 'vignesh.wav') + audio = MonoLoader(filename=filename)() + pm = PitchMelodia() + pitch, pitchConfidence = pm(audio) + + # np.save reference values for later np.loading + #np.save('pitchmelodiapitch.npy', pitch) + #np.save('pitchmelodiaconfidence.npy', pitchConfidence) + + np.loadedPitchMelodiaPitch = np.load(join(filedir(), 'pitchmelodia/pitchmelodiapitch.npy')) + self.assertAlmostEqualVectorFixedPrecision(pitch, np.loadedPitchMelodiaPitch.tolist(), 8) + + np.loadedPitchConfidence = np.load(join(filedir(), 'pitchmelodia/pitchmelodiaconfidence.npy')) + self.assertAlmostEqualVectorFixedPrecision(pitchConfidence, np.loadedPitchConfidence.tolist(), 8) + + def testRealCaseEqualLoudness(self): + filename = join(testdata.audio_dir, 'recorded', 'vignesh.wav') + audio = MonoLoader(filename=filename)() + pm = PitchMelodia() + eq = EqualLoudness() + eqAudio = eq(audio) + pitch, pitchConfidence = pm(eqAudio) + + # np.save reference values for later np.loading + #np.save('pitchmelodiapitch_eqloud.npy', pitch) + #np.save('pitchmelodiaconfidence_eqloud.npy', pitchConfidence) + + np.loadedPitchMelodiaPitch = np.load(join(filedir(), 'pitchmelodia/pitchmelodiapitch_eqloud.npy')) + self.assertAlmostEqualVectorFixedPrecision(pitch, np.loadedPitchMelodiaPitch.tolist(), 8) + + np.loadedPitchConfidence = np.load(join(filedir(), 'pitchmelodia/pitchmelodiaconfidence_eqloud.npy')) + self.assertAlmostEqualVectorFixedPrecision(pitchConfidence, np.loadedPitchConfidence.tolist(), 8) + + def test110Hz(self): + signal = 0.5 * numpy.sin((array(range(10 * 4096))/44100.) * 110 * 2*math.pi) + pm = PitchMelodia() + pitch, confidence = pm(signal) + self.assertAlmostEqual(pitch[50], 110.0, 10) + + def test110HzPeakThresholds(self): + signal = 0.5 * numpy.sin((array(range(10 * 4096))/44100.) * 110 * 2*math.pi) + pm_default = PitchMelodia() + pm_hw0 = PitchMelodia(peakFrameThreshold=0) + pm_hw1 = PitchMelodia(peakFrameThreshold=1) + + pitch_default, confidence_default = pm_default(signal) + pitch_hw0, confidence_hw0 = pm_hw0(signal) + pitch_hw1, confidence_hw1 = pm_hw1(signal) + + self.assertAlmostEqual(pitch_default[50], 110.0, 10) + self.assertAlmostEqual(pitch_hw0[50], 110.0, 10) + self.assertAlmostEqual(pitch_hw1[50], 110.0, 10) + + def testDifferentPeaks(self): + signal_55Hz = 0.5 * numpy.sin((array(range(10 * 4096))/44100.) * 55 * 2*math.pi) + signal_85Hz = 0.5 * numpy.sin((array(range(10 * 4096))/44100.) * 85 * 2*math.pi) + signal = signal_55Hz + signal_85Hz + pm = PitchMelodia() + pitch, confidence = pm(signal) + + for p in pitch[83:129]: # Adjusted the range to be more clear + self.assertGreater(p, 55) + self.assertLess(p, 85) + + def testBelowReferenceFrequency1(self): + signal_50Hz = 1.5 * numpy.sin((array(range(10 * 4096))/44100.) * 50 * 2*math.pi) + pitch, confidence = PitchMelodia()(signal_50Hz) + self.assertAlmostEqual(pitch[10], 100.0, 2) + def testBelowReferenceFrequency2(self): + signal_30Hz = 1.5 * numpy.sin((array(range(10 * 4096))/44100.) * 30 * 2*math.pi) + pitch, confidence = PitchMelodia(referenceFrequency=40)(signal_30Hz) + self.assertAlmostEqual(pitch[10], 60.0, 2) suite = allTests(TestPitchMelodia)