def test_get_vad_samples_stft_consistency():
sr = 48000
win_size_ms = 50
percent_overlap = 0
snr = 20
speech, sr = sp.loadsound(test_python, sr=sr)
noise = sp.generate_noise(len(speech), random_seed=40)
speech_snr, snr_measured = sp.dsp.add_backgroundsound(
speech,
noise,
sr=sr,
snr=snr,
delay_mainsound_sec=1,
total_len_sec=3,
random_seed=40)
vad_samples, vad_matrix1 = sp.feats.get_vad_samples(
speech_snr,
sr=sr,
win_size_ms=win_size_ms,
percent_overlap=percent_overlap)
vad_stft, vad_matrix2 = sp.feats.get_vad_stft(
speech_snr,
sr=sr,
win_size_ms=win_size_ms,
percent_overlap=percent_overlap)
vad_samples_ms = len(vad_samples) / sr / 0.001
vad_stft_ms = len(vad_stft) * win_size_ms
assert vad_samples_ms == vad_stft_ms
def test_get_samples_clipped_40SNR_50percentVAD_length_threshold():
sr = 48000
win_size_ms = 50
percent_overlap = 0
snr = 40
speech, sr = sp.loadsound(test_python, sr=sr, remove_dc=False)
# get just speech segment, no surrounding silence:
# goal VAD
speech_only, sr = sp.loadsound('{}python_speech_only.wav'.format(test_dir),
sr=sr,
remove_dc=False)
noise = sp.generate_noise(len(speech), random_seed=40)
speech_snr, snr_measured = sp.dsp.add_backgroundsound(
speech,
noise,
sr=sr,
snr=snr,
delay_mainsound_sec=1,
total_len_sec=3,
random_seed=snr,
remove_dc=False)
vad_samples, vad_vector = sp.feats.get_samples_clipped(
speech_snr,
sr=sr,
win_size_ms=win_size_ms,
percent_overlap=percent_overlap)
vad_samples_ms = len(vad_samples) / sr / 0.001
example_speech_ms = len(speech_only) / sr / 0.001
if example_speech_ms >= vad_samples_ms:
assert (example_speech_ms / vad_samples_ms) < 2
elif example_speech_ms <= vad_samples_ms:
assert (vad_samples_ms / example_speech_ms) < 2
#########################################################################
# Add them together and see what they look like:
sig3 = sig1 + sig2
sp.plotsound(sig3,
sr=sr,
feature_type='signal',
title='Mixed Signals: {} Hz + {} Hz'.format(sig1_hz, sig2_hz))
##########################################################################
# Generate Pseudo-Random Noise
# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#########################################################################
# Create noise to add to the signal:
noise = sp.generate_noise(len(sig3), amplitude=0.025, random_seed=40)
sp.plotsound(noise, sr=sr, feature_type='signal', title='Random Noise')
###########################################################################
# Control SNR: Adding a Background Sound
# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#########################################################################
# Add noise at signal-to-noise ratio of 40
sig_noisy, snr = sp.dsp.add_backgroundsound(audio_main=sig3,
audio_background=noise,
sr=sr,
snr=40,
clip_at_zero=False)
# keep energy between 1 and -1