def JDC(self, wavPath, output, sr=SAMPLE_RATE):
"""<Joint Detection and Classification of Singing Voice Melody Using Convolutional Recurrent Neural Networks>"""
commands = ("python", "./melodyExtraction_JDC.py", wavPath, output)
ret = subprocess.call(commands, cwd=ALGO_BASE_DIRS["JDC"])
assert ret == 0, f"return value: {ret} != 0"
times, pitches = load_time_series(output, delimiter=r"\s+|,")
return {"times": times, "pitches": pitches}
def loadMel(melfile):
try:
times, freqs = load_time_series(melfile, delimiter=r'\s+|\t+|,')
except ValueError:
cols = load_delimited(melfile, [int, int, str, float, float])
times, freqs = np.array(cols[0]) / 100., np.array(cols[3])
return times, freqs
def callOnFile(wav, commands, wavPath, melPath, scKwargs={}):
'''call <commands> with <wav> write to <wavPath>, then collect results from <melPath>'''
if not VERBOSE_OUT:
scKwargs['stdout'] = subprocess.DEVNULL
if not VERBOSE_ERR:
scKwargs['stderr'] = subprocess.DEVNULL
write_wav(wavPath, wav[0], wav[1])
ret = subprocess.call(commands, **scKwargs)
assert ret == 0, f'return value: {ret} != 0'
times, pitches = load_time_series(melPath, delimiter=r'\s+|,')
return times, pitches
def SSL(self, wavPath, output, sr=SAMPLE_RATE):
"""<Semi-supervised learning using teacher-student models for vocal melody extraction>"""
dirname = os.path.dirname(output) + "/"
output = f"pitch_{os.path.basename(wavPath)}.txt"
output = os.path.join(dirname, output)
commands = ("python", "./melodyExtraction_NS.py", "-p", wavPath, "-o", dirname)
logger.debug(f"SSL commands={commands}")
ret = subprocess.call(commands, cwd=ALGO_BASE_DIRS["SSL"])
assert ret == 0, f"return value: {ret} != 0"
times, pitches = load_time_series(output, delimiter=r"\s+|,")
return {"times": times, "pitches": pitches}
def loadGT(self, GTPath):
gt = load_time_series(GTPath, delimiter=',')
return gt
def loadGT(self, GTPath):
gt = load_time_series(GTPath)
return gt