vocos/feature_extractors.py

from typing import List

import torch
import torchaudio
# from encodec import EncodecModel
from torch import nn
from omegaconf import OmegaConf
from vocos.modules import safe_log
# from vocos.xcodec.models.soundstream_hubert_new import SoundStream

class FeatureExtractor(nn.Module):
    """Base class for feature extractors."""

    def forward(self, audio: torch.Tensor, **kwargs) -> torch.Tensor:
        """
        Extract features from the given audio.

        Args:
            audio (Tensor): Input audio waveform.

        Returns:
            Tensor: Extracted features of shape (B, C, L), where B is the batch size,
                    C denotes output features, and L is the sequence length.
        """
        raise NotImplementedError("Subclasses must implement the forward method.")


class MelSpectrogramFeatures(FeatureExtractor):
    def __init__(self, sample_rate=24000, n_fft=1024, hop_length=256, n_mels=100, padding="center"):
        super().__init__()
        if padding not in ["center", "same"]:
            raise ValueError("Padding must be 'center' or 'same'.")
        self.padding = padding
        self.mel_spec = torchaudio.transforms.MelSpectrogram(
            sample_rate=sample_rate,
            n_fft=n_fft,
            hop_length=hop_length,
            n_mels=n_mels,
            center=padding == "center",
            power=1,
        )

    def forward(self, audio, **kwargs):
        if self.padding == "same":
            pad = self.mel_spec.win_length - self.mel_spec.hop_length
            audio = torch.nn.functional.pad(audio, (pad // 2, pad // 2), mode="reflect")
        mel = self.mel_spec(audio)
        features = safe_log(mel)
        return features


class EncodecFeatures(FeatureExtractor):
    def __init__(
        self,
        encodec_model: str = "encodec_24khz",
        bandwidths: List[float] = [1.5, 3.0, 6.0, 12.0],
        train_codebooks: bool = False,
    ):
        super().__init__()
        if encodec_model == "encodec_24khz":
            encodec = EncodecModel.encodec_model_24khz
        elif encodec_model == "encodec_48khz":
            encodec = EncodecModel.encodec_model_48khz
        else:
            raise ValueError(
                f"Unsupported encodec_model: {encodec_model}. Supported options are 'encodec_24khz' and 'encodec_48khz'."
            )
        self.encodec = encodec(pretrained=True)
        for param in self.encodec.parameters():
            param.requires_grad = False
        self.num_q = self.encodec.quantizer.get_num_quantizers_for_bandwidth(
            self.encodec.frame_rate, bandwidth=max(bandwidths)
        )
        codebook_weights = torch.cat([vq.codebook for vq in self.encodec.quantizer.vq.layers[: self.num_q]], dim=0)
        self.codebook_weights = torch.nn.Parameter(codebook_weights, requires_grad=train_codebooks)
        self.bandwidths = bandwidths

    @torch.no_grad()
    def get_encodec_codes(self, audio):
        audio = audio.unsqueeze(1)
        emb = self.encodec.encoder(audio)
        codes = self.encodec.quantizer.encode(emb, self.encodec.frame_rate, self.encodec.bandwidth)
        return codes

    def forward(self, audio: torch.Tensor, **kwargs):
        bandwidth_id = kwargs.get("bandwidth_id")
        if bandwidth_id is None:
            raise ValueError("The 'bandwidth_id' argument is required")
        self.encodec.eval()  # Force eval mode as Pytorch Lightning automatically sets child modules to training mode
        self.encodec.set_target_bandwidth(self.bandwidths[bandwidth_id])
        codes = self.get_encodec_codes(audio)
        # Instead of summing in the loop, it stores subsequent VQ dictionaries in a single `self.codebook_weights`
        # with offsets given by the number of bins, and finally summed in a vectorized operation.
        offsets = torch.arange(
            0, self.encodec.quantizer.bins * len(codes), self.encodec.quantizer.bins, device=audio.device
        )
        embeddings_idxs = codes + offsets.view(-1, 1, 1)
        features = torch.nn.functional.embedding(embeddings_idxs, self.codebook_weights).sum(dim=0)
        return features.transpose(1, 2)

class xCodecFeatures(FeatureExtractor):
    def __init__(
        self,
        config: str,
        ckpt: str,
    ):
        super().__init__()
        self.config = OmegaConf.load(config)
        self.model = eval(self.config.generator.name)(**self.config.generator.config)
        parameter_dict = torch.load(ckpt, map_location='cpu')
        self.model.load_state_dict(parameter_dict['codec_model'])
        self.resampler = torchaudio.transforms.Resample(orig_freq=44100, new_freq=16000).to('cuda')
        self.model.eval()
    
    def forward(self, audio: torch.Tensor):
        # resample audio from 44100 to 16000
        audio = self.resampler(audio)
        with torch.no_grad():
            codes = self.model.encode(audio, target_bw=6)
        return codes