Rename TTS_part_sample_rate to encoder_sample_rate

This commit is contained in:
Edresson Casanova 2022-04-22 07:57:27 -03:00
parent 3f3efe88bb
commit b3e2c58398
3 changed files with 18 additions and 18 deletions

View File

@ -455,14 +455,14 @@ class VitsArgs(Coqpit):
freeze_waveform_decoder (bool):
Freeze the waveform decoder weigths during training. Defaults to False.
TTS_part_sample_rate (int):
encoder_sample_rate (int):
If not None this sample rate will be used for training the Posterior Encoder,
flow, text_encoder and duration predictor. The decoder part (vocoder) will be
trained with the `config.audio.sample_rate`. Defaults to None.
interpolate_z (bool):
If `TTS_part_sample_rate` not None and this parameter True the nearest interpolation
will be used to upsampling the latent variable z with the sampling rate `TTS_part_sample_rate`
If `encoder_sample_rate` not None and this parameter True the nearest interpolation
will be used to upsampling the latent variable z with the sampling rate `encoder_sample_rate`
to the `config.audio.sample_rate`. If it is False you will need to add extra
`upsample_rates_decoder` to match the shape. Defaults to True.
@ -521,7 +521,7 @@ class VitsArgs(Coqpit):
freeze_PE: bool = False
freeze_flow_decoder: bool = False
freeze_waveform_decoder: bool = False
TTS_part_sample_rate: int = None
encoder_sample_rate: int = None
interpolate_z: bool = True
@ -648,10 +648,10 @@ class Vits(BaseTTS):
use_spectral_norm=self.args.use_spectral_norm_disriminator,
)
if self.args.TTS_part_sample_rate:
self.interpolate_factor = self.config.audio["sample_rate"] / self.args.TTS_part_sample_rate
if self.args.encoder_sample_rate:
self.interpolate_factor = self.config.audio["sample_rate"] / self.args.encoder_sample_rate
self.audio_resampler = torchaudio.transforms.Resample(
orig_freq=self.config.audio["sample_rate"], new_freq=self.args.TTS_part_sample_rate
orig_freq=self.config.audio["sample_rate"], new_freq=self.args.encoder_sample_rate
)
def init_multispeaker(self, config: Coqpit):
@ -906,7 +906,7 @@ class Vits(BaseTTS):
# select a random feature segment for the waveform decoder
z_slice, slice_ids = rand_segments(z, y_lengths, self.spec_segment_size, let_short_samples=True, pad_short=True)
if self.args.TTS_part_sample_rate:
if self.args.encoder_sample_rate:
slice_ids = slice_ids * int(self.interpolate_factor)
spec_segment_size = self.spec_segment_size * int(self.interpolate_factor)
if self.args.interpolate_z:
@ -1029,7 +1029,7 @@ class Vits(BaseTTS):
z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * self.inference_noise_scale
z = self.flow(z_p, y_mask, g=g, reverse=True)
if self.args.TTS_part_sample_rate and self.args.interpolate_z:
if self.args.encoder_sample_rate and self.args.interpolate_z:
z = z.unsqueeze(0) # pylint: disable=not-callable
z = torch.nn.functional.interpolate(z, scale_factor=[1, self.interpolate_factor], mode="nearest").squeeze(0)
y_mask = (
@ -1155,7 +1155,7 @@ class Vits(BaseTTS):
# compute melspec segment
with autocast(enabled=False):
if self.args.TTS_part_sample_rate:
if self.args.encoder_sample_rate:
spec_segment_size = self.spec_segment_size * int(self.interpolate_factor)
else:
spec_segment_size = self.spec_segment_size
@ -1370,7 +1370,7 @@ class Vits(BaseTTS):
"""Compute spectrograms on the device."""
ac = self.config.audio
if self.args.TTS_part_sample_rate:
if self.args.encoder_sample_rate:
wav = self.audio_resampler(batch["waveform"])
else:
wav = batch["waveform"]
@ -1378,7 +1378,7 @@ class Vits(BaseTTS):
# compute spectrograms
batch["spec"] = wav_to_spec(wav, ac.fft_size, ac.hop_length, ac.win_length, center=False)
if self.args.TTS_part_sample_rate:
if self.args.encoder_sample_rate:
# recompute spec with high sampling rate to the loss
spec_mel = wav_to_spec(batch["waveform"], ac.fft_size, ac.hop_length, ac.win_length, center=False)
else:
@ -1393,14 +1393,14 @@ class Vits(BaseTTS):
fmax=ac.mel_fmax,
)
if not self.args.TTS_part_sample_rate:
if not self.args.encoder_sample_rate:
assert batch["spec"].shape[2] == batch["mel"].shape[2], f"{batch['spec'].shape[2]}, {batch['mel'].shape[2]}"
# compute spectrogram frame lengths
batch["spec_lens"] = (batch["spec"].shape[2] * batch["waveform_rel_lens"]).int()
batch["mel_lens"] = (batch["mel"].shape[2] * batch["waveform_rel_lens"]).int()
if not self.args.TTS_part_sample_rate:
if not self.args.encoder_sample_rate:
assert (batch["spec_lens"] - batch["mel_lens"]).sum() == 0
# zero the padding frames
@ -1518,7 +1518,7 @@ class Vits(BaseTTS):
# as it is probably easier for model distribution.
state["model"] = {k: v for k, v in state["model"].items() if "speaker_encoder" not in k}
if self.args.TTS_part_sample_rate is not None and eval:
if self.args.encoder_sample_rate is not None and eval:
# audio resampler is not used in inference time
self.audio_resampler = None
@ -1549,7 +1549,7 @@ class Vits(BaseTTS):
from TTS.utils.audio import AudioProcessor
upsample_rate = torch.prod(torch.as_tensor(config.model_args.upsample_rates_decoder)).item()
if not config.model_args.TTS_part_sample_rate:
if not config.model_args.encoder_sample_rate:
assert (
upsample_rate == config.audio.hop_length
), f" [!] Product of upsample rates must be equal to the hop length - {upsample_rate} vs {config.audio.hop_length}"

View File

@ -42,7 +42,7 @@ config.model_args.d_vector_dim = 256
# test upsample interpolation approach
config.model_args.TTS_part_sample_rate = 11025
config.model_args.encoder_sample_rate = 11025
config.model_args.interpolate_z = True
config.model_args.upsample_rates_decoder = [8, 8, 2, 2]
config.model_args.periods_multi_period_discriminator = [2, 3, 5, 7]

View File

@ -42,7 +42,7 @@ config.model_args.d_vector_dim = 256
# test upsample
config.model_args.TTS_part_sample_rate = 11025
config.model_args.encoder_sample_rate = 11025
config.model_args.interpolate_z = False
config.model_args.upsample_rates_decoder = [8, 8, 4, 2]
config.model_args.periods_multi_period_discriminator = [2, 3, 5, 7]