def extract_features(song_path, args, opt):
y_wav, sr = librosa.load(song_path, sr=opt.sampling_rate)
# useful quantities
feature_name = opt.feature_name
feature_size = opt.feature_size
sampling_rate = opt.sampling_rate
beat_subdivision = opt.beat_subdivision
try:
step_size = opt.step_size
using_bpm_time_division = opt.using_bpm_time_division
except: # older model
using_bpm_time_division = True
sr = sampling_rate
hop = int(step_size * sr)
#get feature
if feature_name == "chroma":
features = extract_features_hybrid(y_wav, sr, hop)
elif feature_name == "mel":
features = extract_features_mel(y_wav, sr, hop, mel_dim=feature_size)
elif feature_name == "multi_mel":
features = extract_features_multi_mel(y_wav,
sr=sampling_rate,
hop=hop,
nffts=[1024, 2048, 4096],
mel_dim=feature_size)
return hop, features
def extract_features(song_path, args, opt):
y_wav, sr = librosa.load(song_path, sr=opt.sampling_rate)
# useful quantities
bpm = args.bpm
feature_name = opt.feature_name
feature_size = opt.feature_size
sampling_rate = opt.sampling_rate
beat_subdivision = opt.beat_subdivision
try:
step_size = opt.step_size
using_bpm_time_division = opt.using_bpm_time_division
except: # older model
using_bpm_time_division = True
sr = sampling_rate
beat_duration = 60/bpm #beat duration in seconds
beat_duration_samples = int(60*sr/bpm) #beat duration in samples
if using_bpm_time_division:
# duration of one time step in samples:
hop = int(beat_duration_samples * 1/beat_subdivision)
step_size = beat_duration/beat_subdivision # in seconds
else:
beat_subdivision = 1/(step_size*bpm/60)
hop = int(step_size*sr)
#get feature
if feature_name == "chroma":
if using_bpm_time_division:
state_times = np.arange(0,y_wav.shape[0]/sr,step=step_size)
features = extract_features_hybrid_beat_synced(y_wav,sr,state_times,bpm,beat_discretization=1/beat_subdivision)
else:
features = extract_features_hybrid(y_wav,sr,hop)
elif feature_name == "mel":
if using_bpm_time_division:
raise NotImplementedError("Mel features with beat synced times not implemented, but trivial TODO")
else:
features = extract_features_mel(y_wav,sr,hop,mel_dim=feature_size)
elif feature_name == "multi_mel":
if using_bpm_time_division:
raise NotImplementedError("Mel features with beat synced times not implemented, but trivial TODO")
else:
features = extract_features_multi_mel(y_wav, sr=sampling_rate, hop=hop, nffts=[1024,2048,4096], mel_dim=feature_size)
return hop, features
bpm = info['_beatsPerMinute']
sr = sampling_rate
hop = int(sr * step_size)
#get feature
if feature_name == "chroma":
if using_bpm_time_division:
state_times = np.arange(0, y_wav.shape[0] / sr, step=step_size)
features = extract_features_hybrid_beat_synced(
y_wav,
sr,
state_times,
bpm,
beat_discretization=1 / beat_subdivision)
else:
features = extract_features_hybrid(y_wav, sr, hop)
elif feature_name == "mel":
if using_bpm_time_division:
raise NotImplementedError(
"Mel features with beat synced times not implemented, but trivial TODO"
)
else:
features = extract_features_mel(y_wav,
sr,
hop,
mel_dim=feature_size)
elif feature_name == "multi_mel":
if using_bpm_time_division:
raise NotImplementedError(
"Mel features with beat synced times not implemented, but trivial TODO"
)