def test_invalid_files():
Nonexist = ['temp.wav', 'temp.mp3']
target_dbfs = -13.5
process_files(Nonexist, target_dbfs)
Unsupported = ['audio1.ape', 'audio2.aac']
for u in Unsupported:
with open(u, 'w'):
pass
process_files(Unsupported, target_dbfs)
cleanup_on_finish(Unsupported, '/')
def convert_single_wav_to_input(self, audio_filename):
aus = []
audio_file = os.path.join(self.src_dir, 'demo_wav', audio_filename)
# Default param
TARGET_AUDIO_DBFS = -20.0
WAV_STEP = int(0.2 * 16000) # 0.2s = 5 frames
STFT_WINDOW_SIZE = {'25': 320, '29.97': 356}
STFT_WINDOW_STEP = {'25': 4, '29.97': 3}
FPS = 25
# Step 1 : Normalize the volume
target_dbfs = TARGET_AUDIO_DBFS
pynormalize.process_files(Files=[audio_file],
target_dbfs=target_dbfs,
directory=os.path.join(
self.src_dir, 'raw_wav'))
# Step 2 : load wav file
sample_rate, samples = wav.read(audio_file)
assert (sample_rate == 16000)
if (len(samples.shape) > 1):
samples = samples[:, 0] # pick mono
# Step 3 : STFT,
# 1 frame = 1/25 * 16k = 640 samples => windowsize=320, overlap=160
# 1 frame = 1/29.97 * 16k = 533.86 samples => windowsize=356, overlap=178, (mis-align = 4.2sample / 1s)
f, t, Zxx = stft(samples,
fs=sample_rate,
nperseg=STFT_WINDOW_SIZE[str(FPS)])
# stft_abs = np.abs(Zxx)
stft_abs = np.log(np.abs(Zxx)**2 + 1e-10)
stft_abs_max = np.max(stft_abs)
stft_abs /= stft_abs_max
# Step 4 : align AV (drop last 2 frames of V)
fl_length = stft_abs.shape[1] // STFT_WINDOW_STEP[str(FPS)]
audio_stft_length = (fl_length - 2) * STFT_WINDOW_STEP[str(FPS)]
stft_signal = Zxx[:, 0:audio_stft_length]
stft_abs = stft_abs[:, 0:audio_stft_length]
audio_wav_length = int((fl_length - 2) * sample_rate / FPS)
wav_signal = samples[0:audio_wav_length]
# # Step 6 : Save audio
# info_audio = (0, stft_signal, fl_length - 2, audio_stft_length, audio_wav_length)
# au_data = (stft_abs, wav_signal, info_audio)
aus.append((stft_abs.T, None, (0, audio_filename, 0)))
return aus
def test_process_files():
Files = ['temp.wav']
target_dbfs = -13.5
empty_audio = AudioSegment.silent(duration=3000)
empty_audio.export(Files[0], format='wav')
process_files(Files, target_dbfs)
if Files[0] in os.listdir(STORE) and \
get_modified_time_diff(os.path.join(STORE, Files[0])) < 100:
assert True
else:
assert False
cleanup_on_finish(Files, STORE)
def test_process_files_different_directory():
Files = ['temp.wav']
target_dbfs = -13.5
EDITED_STORE = '_TEMP'
empty_audio = AudioSegment.silent(duration=3000)
empty_audio.export(Files[0], format='wav')
process_files(Files=Files, target_dbfs=target_dbfs, directory=EDITED_STORE)
if Files[0] in os.listdir(EDITED_STORE) and \
get_modified_time_diff(os.path.join(EDITED_STORE, Files[0])) < 100:
assert True
else:
assert False
cleanup_on_finish(Files, EDITED_STORE)