def create_spectro(self, item:AudioItem):
if self.config.mfcc:
mel = MFCC(sample_rate=item.sr, n_mfcc=self.config.sg_cfg.n_mfcc, melkwargs=self.config.sg_cfg.mel_args())(item.sig)
else:
mel = MelSpectrogram(**(self.config.sg_cfg.mel_args()))(item.sig)
if self.config.sg_cfg.to_db_scale:
mel = AmplitudeToDB(top_db=self.config.sg_cfg.top_db)(mel)
mel = mel.detach()
if self.config.standardize:
mel = standardize(mel)
if self.config.delta:
mel = torch.cat([torch.stack([m,torchdelta(m),torchdelta(m, order=2)]) for m in mel])
return mel
def create_spectro(self, item:AudioItem):
if self.config.mfcc:
mel = MFCC(sample_rate=item.sr, n_mfcc=self.config.sg_cfg.n_mfcc, melkwargs=self.config.sg_cfg.mel_args())(item.sig)
else:
if self.config.sg_cfg.custom_spectro != None:
mel = self.config.sg_cfg.custom_spectro(item.sig)
else:
if self.config.sg_cfg.n_mels > 0:
c = self.config.sg_cfg
mel = librosa.feature.melspectrogram(y=np.array(item.sig[0,:]), sr=item.sr, fmax=c.f_max, fmin=c.f_min, **(self.config.sg_cfg.mel_args()))
mel = torch.from_numpy(mel)
mel.unsqueeze_(0)
else:
mel = Spectrogram(**(self.config.sg_cfg.spectro_args()))(item.sig)
if self.config.sg_cfg.to_db_scale:
mel = AmplitudeToDB(top_db=self.config.sg_cfg.top_db)(mel)
mel = mel.detach()
if self.config.standardize:
mel = standardize(mel)
if self.config.delta:
mel = torch.cat([torch.stack([m,torchdelta(m),torchdelta(m, order=2)]) for m in mel])
return mel