feat: Integrate IndexTTS2 model and update related schemas and frontend components

qwen3-tts-backend/indextts/s2mel/dac/model/__init__.py

@@ -0,0 +1,4 @@
+from .base import CodecMixin
+from .base import DACFile
+from .dac import DAC
+from .discriminator import Discriminator

qwen3-tts-backend/indextts/s2mel/dac/model/base.py

@@ -0,0 +1,294 @@
+import math
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Union
+
+import numpy as np
+import torch
+import tqdm
+from audiotools import AudioSignal
+from torch import nn
+
+SUPPORTED_VERSIONS = ["1.0.0"]
+
+
+@dataclass
+class DACFile:
+    codes: torch.Tensor
+
+    # Metadata
+    chunk_length: int
+    original_length: int
+    input_db: float
+    channels: int
+    sample_rate: int
+    padding: bool
+    dac_version: str
+
+    def save(self, path):
+        artifacts = {
+            "codes": self.codes.numpy().astype(np.uint16),
+            "metadata": {
+                "input_db": self.input_db.numpy().astype(np.float32),
+                "original_length": self.original_length,
+                "sample_rate": self.sample_rate,
+                "chunk_length": self.chunk_length,
+                "channels": self.channels,
+                "padding": self.padding,
+                "dac_version": SUPPORTED_VERSIONS[-1],
+            },
+        }
+        path = Path(path).with_suffix(".dac")
+        with open(path, "wb") as f:
+            np.save(f, artifacts)
+        return path
+
+    @classmethod
+    def load(cls, path):
+        artifacts = np.load(path, allow_pickle=True)[()]
+        codes = torch.from_numpy(artifacts["codes"].astype(int))
+        if artifacts["metadata"].get("dac_version", None) not in SUPPORTED_VERSIONS:
+            raise RuntimeError(
+                f"Given file {path} can't be loaded with this version of descript-audio-codec."
+            )
+        return cls(codes=codes, **artifacts["metadata"])
+
+
+class CodecMixin:
+    @property
+    def padding(self):
+        if not hasattr(self, "_padding"):
+            self._padding = True
+        return self._padding
+
+    @padding.setter
+    def padding(self, value):
+        assert isinstance(value, bool)
+
+        layers = [
+            l for l in self.modules() if isinstance(l, (nn.Conv1d, nn.ConvTranspose1d))
+        ]
+
+        for layer in layers:
+            if value:
+                if hasattr(layer, "original_padding"):
+                    layer.padding = layer.original_padding
+            else:
+                layer.original_padding = layer.padding
+                layer.padding = tuple(0 for _ in range(len(layer.padding)))
+
+        self._padding = value
+
+    def get_delay(self):
+        # Any number works here, delay is invariant to input length
+        l_out = self.get_output_length(0)
+        L = l_out
+
+        layers = []
+        for layer in self.modules():
+            if isinstance(layer, (nn.Conv1d, nn.ConvTranspose1d)):
+                layers.append(layer)
+
+        for layer in reversed(layers):
+            d = layer.dilation[0]
+            k = layer.kernel_size[0]
+            s = layer.stride[0]
+
+            if isinstance(layer, nn.ConvTranspose1d):
+                L = ((L - d * (k - 1) - 1) / s) + 1
+            elif isinstance(layer, nn.Conv1d):
+                L = (L - 1) * s + d * (k - 1) + 1
+
+            L = math.ceil(L)
+
+        l_in = L
+
+        return (l_in - l_out) // 2
+
+    def get_output_length(self, input_length):
+        L = input_length
+        # Calculate output length
+        for layer in self.modules():
+            if isinstance(layer, (nn.Conv1d, nn.ConvTranspose1d)):
+                d = layer.dilation[0]
+                k = layer.kernel_size[0]
+                s = layer.stride[0]
+
+                if isinstance(layer, nn.Conv1d):
+                    L = ((L - d * (k - 1) - 1) / s) + 1
+                elif isinstance(layer, nn.ConvTranspose1d):
+                    L = (L - 1) * s + d * (k - 1) + 1
+
+                L = math.floor(L)
+        return L
+
+    @torch.no_grad()
+    def compress(
+        self,
+        audio_path_or_signal: Union[str, Path, AudioSignal],
+        win_duration: float = 1.0,
+        verbose: bool = False,
+        normalize_db: float = -16,
+        n_quantizers: int = None,
+    ) -> DACFile:
+        """Processes an audio signal from a file or AudioSignal object into
+        discrete codes. This function processes the signal in short windows,
+        using constant GPU memory.
+
+        Parameters
+        ----------
+        audio_path_or_signal : Union[str, Path, AudioSignal]
+            audio signal to reconstruct
+        win_duration : float, optional
+            window duration in seconds, by default 5.0
+        verbose : bool, optional
+            by default False
+        normalize_db : float, optional
+            normalize db, by default -16
+
+        Returns
+        -------
+        DACFile
+            Object containing compressed codes and metadata
+            required for decompression
+        """
+        audio_signal = audio_path_or_signal
+        if isinstance(audio_signal, (str, Path)):
+            audio_signal = AudioSignal.load_from_file_with_ffmpeg(str(audio_signal))
+
+        self.eval()
+        original_padding = self.padding
+        original_device = audio_signal.device
+
+        audio_signal = audio_signal.clone()
+        original_sr = audio_signal.sample_rate
+
+        resample_fn = audio_signal.resample
+        loudness_fn = audio_signal.loudness
+
+        # If audio is > 10 minutes long, use the ffmpeg versions
+        if audio_signal.signal_duration >= 10 * 60 * 60:
+            resample_fn = audio_signal.ffmpeg_resample
+            loudness_fn = audio_signal.ffmpeg_loudness
+
+        original_length = audio_signal.signal_length
+        resample_fn(self.sample_rate)
+        input_db = loudness_fn()
+
+        if normalize_db is not None:
+            audio_signal.normalize(normalize_db)
+        audio_signal.ensure_max_of_audio()
+
+        nb, nac, nt = audio_signal.audio_data.shape
+        audio_signal.audio_data = audio_signal.audio_data.reshape(nb * nac, 1, nt)
+        win_duration = (
+            audio_signal.signal_duration if win_duration is None else win_duration
+        )
+
+        if audio_signal.signal_duration <= win_duration:
+            # Unchunked compression (used if signal length < win duration)
+            self.padding = True
+            n_samples = nt
+            hop = nt
+        else:
+            # Chunked inference
+            self.padding = False
+            # Zero-pad signal on either side by the delay
+            audio_signal.zero_pad(self.delay, self.delay)
+            n_samples = int(win_duration * self.sample_rate)
+            # Round n_samples to nearest hop length multiple
+            n_samples = int(math.ceil(n_samples / self.hop_length) * self.hop_length)
+            hop = self.get_output_length(n_samples)
+
+        codes = []
+        range_fn = range if not verbose else tqdm.trange
+
+        for i in range_fn(0, nt, hop):
+            x = audio_signal[..., i : i + n_samples]
+            x = x.zero_pad(0, max(0, n_samples - x.shape[-1]))
+
+            audio_data = x.audio_data.to(self.device)
+            audio_data = self.preprocess(audio_data, self.sample_rate)
+            _, c, _, _, _ = self.encode(audio_data, n_quantizers)
+            codes.append(c.to(original_device))
+            chunk_length = c.shape[-1]
+
+        codes = torch.cat(codes, dim=-1)
+
+        dac_file = DACFile(
+            codes=codes,
+            chunk_length=chunk_length,
+            original_length=original_length,
+            input_db=input_db,
+            channels=nac,
+            sample_rate=original_sr,
+            padding=self.padding,
+            dac_version=SUPPORTED_VERSIONS[-1],
+        )
+
+        if n_quantizers is not None:
+            codes = codes[:, :n_quantizers, :]
+
+        self.padding = original_padding
+        return dac_file
+
+    @torch.no_grad()
+    def decompress(
+        self,
+        obj: Union[str, Path, DACFile],
+        verbose: bool = False,
+    ) -> AudioSignal:
+        """Reconstruct audio from a given .dac file
+
+        Parameters
+        ----------
+        obj : Union[str, Path, DACFile]
+            .dac file location or corresponding DACFile object.
+        verbose : bool, optional
+            Prints progress if True, by default False
+
+        Returns
+        -------
+        AudioSignal
+            Object with the reconstructed audio
+        """
+        self.eval()
+        if isinstance(obj, (str, Path)):
+            obj = DACFile.load(obj)
+
+        original_padding = self.padding
+        self.padding = obj.padding
+
+        range_fn = range if not verbose else tqdm.trange
+        codes = obj.codes
+        original_device = codes.device
+        chunk_length = obj.chunk_length
+        recons = []
+
+        for i in range_fn(0, codes.shape[-1], chunk_length):
+            c = codes[..., i : i + chunk_length].to(self.device)
+            z = self.quantizer.from_codes(c)[0]
+            r = self.decode(z)
+            recons.append(r.to(original_device))
+
+        recons = torch.cat(recons, dim=-1)
+        recons = AudioSignal(recons, self.sample_rate)
+
+        resample_fn = recons.resample
+        loudness_fn = recons.loudness
+
+        # If audio is > 10 minutes long, use the ffmpeg versions
+        if recons.signal_duration >= 10 * 60 * 60:
+            resample_fn = recons.ffmpeg_resample
+            loudness_fn = recons.ffmpeg_loudness
+
+        recons.normalize(obj.input_db)
+        resample_fn(obj.sample_rate)
+        recons = recons[..., : obj.original_length]
+        loudness_fn()
+        recons.audio_data = recons.audio_data.reshape(
+            -1, obj.channels, obj.original_length
+        )
+
+        self.padding = original_padding
+        return recons

qwen3-tts-backend/indextts/s2mel/dac/model/dac.py

@@ -0,0 +1,400 @@
+import math
+from typing import List
+from typing import Union
+
+import numpy as np
+import torch
+from audiotools import AudioSignal
+from audiotools.ml import BaseModel
+from torch import nn
+
+from .base import CodecMixin
+from indextts.s2mel.dac.nn.layers import Snake1d
+from indextts.s2mel.dac.nn.layers import WNConv1d
+from indextts.s2mel.dac.nn.layers import WNConvTranspose1d
+from indextts.s2mel.dac.nn.quantize import ResidualVectorQuantize
+from .encodec import SConv1d, SConvTranspose1d, SLSTM
+
+
+def init_weights(m):
+    if isinstance(m, nn.Conv1d):
+        nn.init.trunc_normal_(m.weight, std=0.02)
+        nn.init.constant_(m.bias, 0)
+
+
+class ResidualUnit(nn.Module):
+    def __init__(self, dim: int = 16, dilation: int = 1, causal: bool = False):
+        super().__init__()
+        conv1d_type = SConv1d# if causal else WNConv1d
+        pad = ((7 - 1) * dilation) // 2
+        self.block = nn.Sequential(
+            Snake1d(dim),
+            conv1d_type(dim, dim, kernel_size=7, dilation=dilation, padding=pad, causal=causal, norm='weight_norm'),
+            Snake1d(dim),
+            conv1d_type(dim, dim, kernel_size=1, causal=causal, norm='weight_norm'),
+        )
+
+    def forward(self, x):
+        y = self.block(x)
+        pad = (x.shape[-1] - y.shape[-1]) // 2
+        if pad > 0:
+            x = x[..., pad:-pad]
+        return x + y
+
+
+class EncoderBlock(nn.Module):
+    def __init__(self, dim: int = 16, stride: int = 1, causal: bool = False):
+        super().__init__()
+        conv1d_type = SConv1d# if causal else WNConv1d
+        self.block = nn.Sequential(
+            ResidualUnit(dim // 2, dilation=1, causal=causal),
+            ResidualUnit(dim // 2, dilation=3, causal=causal),
+            ResidualUnit(dim // 2, dilation=9, causal=causal),
+            Snake1d(dim // 2),
+            conv1d_type(
+                dim // 2,
+                dim,
+                kernel_size=2 * stride,
+                stride=stride,
+                padding=math.ceil(stride / 2),
+                causal=causal,
+                norm='weight_norm',
+            ),
+        )
+
+    def forward(self, x):
+        return self.block(x)
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        d_model: int = 64,
+        strides: list = [2, 4, 8, 8],
+        d_latent: int = 64,
+        causal: bool = False,
+        lstm: int = 2,
+    ):
+        super().__init__()
+        conv1d_type = SConv1d# if causal else WNConv1d
+        # Create first convolution
+        self.block = [conv1d_type(1, d_model, kernel_size=7, padding=3, causal=causal, norm='weight_norm')]
+
+        # Create EncoderBlocks that double channels as they downsample by `stride`
+        for stride in strides:
+            d_model *= 2
+            self.block += [EncoderBlock(d_model, stride=stride, causal=causal)]
+
+        # Add LSTM if needed
+        self.use_lstm = lstm
+        if lstm:
+            self.block += [SLSTM(d_model, lstm)]
+
+        # Create last convolution
+        self.block += [
+            Snake1d(d_model),
+            conv1d_type(d_model, d_latent, kernel_size=3, padding=1, causal=causal, norm='weight_norm'),
+        ]
+
+        # Wrap black into nn.Sequential
+        self.block = nn.Sequential(*self.block)
+        self.enc_dim = d_model
+
+    def forward(self, x):
+        return self.block(x)
+
+    def reset_cache(self):
+        # recursively find all submodules named SConv1d in self.block and use their reset_cache method
+        def reset_cache(m):
+            if isinstance(m, SConv1d) or isinstance(m, SLSTM):
+                m.reset_cache()
+                return
+            for child in m.children():
+                reset_cache(child)
+
+        reset_cache(self.block)
+
+
+class DecoderBlock(nn.Module):
+    def __init__(self, input_dim: int = 16, output_dim: int = 8, stride: int = 1, causal: bool = False):
+        super().__init__()
+        conv1d_type = SConvTranspose1d #if causal else WNConvTranspose1d
+        self.block = nn.Sequential(
+            Snake1d(input_dim),
+            conv1d_type(
+                input_dim,
+                output_dim,
+                kernel_size=2 * stride,
+                stride=stride,
+                padding=math.ceil(stride / 2),
+                causal=causal,
+                norm='weight_norm'
+            ),
+            ResidualUnit(output_dim, dilation=1, causal=causal),
+            ResidualUnit(output_dim, dilation=3, causal=causal),
+            ResidualUnit(output_dim, dilation=9, causal=causal),
+        )
+
+    def forward(self, x):
+        return self.block(x)
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        input_channel,
+        channels,
+        rates,
+        d_out: int = 1,
+        causal: bool = False,
+        lstm: int = 2,
+    ):
+        super().__init__()
+        conv1d_type = SConv1d# if causal else WNConv1d
+        # Add first conv layer
+        layers = [conv1d_type(input_channel, channels, kernel_size=7, padding=3, causal=causal, norm='weight_norm')]
+
+        if lstm:
+            layers += [SLSTM(channels, num_layers=lstm)]
+
+        # Add upsampling + MRF blocks
+        for i, stride in enumerate(rates):
+            input_dim = channels // 2**i
+            output_dim = channels // 2 ** (i + 1)
+            layers += [DecoderBlock(input_dim, output_dim, stride, causal=causal)]
+
+        # Add final conv layer
+        layers += [
+            Snake1d(output_dim),
+            conv1d_type(output_dim, d_out, kernel_size=7, padding=3, causal=causal, norm='weight_norm'),
+            nn.Tanh(),
+        ]
+
+        self.model = nn.Sequential(*layers)
+
+    def forward(self, x):
+        return self.model(x)
+
+
+class DAC(BaseModel, CodecMixin):
+    def __init__(
+        self,
+        encoder_dim: int = 64,
+        encoder_rates: List[int] = [2, 4, 8, 8],
+        latent_dim: int = None,
+        decoder_dim: int = 1536,
+        decoder_rates: List[int] = [8, 8, 4, 2],
+        n_codebooks: int = 9,
+        codebook_size: int = 1024,
+        codebook_dim: Union[int, list] = 8,
+        quantizer_dropout: bool = False,
+        sample_rate: int = 44100,
+        lstm: int = 2,
+        causal: bool = False,
+    ):
+        super().__init__()
+
+        self.encoder_dim = encoder_dim
+        self.encoder_rates = encoder_rates
+        self.decoder_dim = decoder_dim
+        self.decoder_rates = decoder_rates
+        self.sample_rate = sample_rate
+
+        if latent_dim is None:
+            latent_dim = encoder_dim * (2 ** len(encoder_rates))
+
+        self.latent_dim = latent_dim
+
+        self.hop_length = np.prod(encoder_rates)
+        self.encoder = Encoder(encoder_dim, encoder_rates, latent_dim, causal=causal, lstm=lstm)
+
+        self.n_codebooks = n_codebooks
+        self.codebook_size = codebook_size
+        self.codebook_dim = codebook_dim
+        self.quantizer = ResidualVectorQuantize(
+            input_dim=latent_dim,
+            n_codebooks=n_codebooks,
+            codebook_size=codebook_size,
+            codebook_dim=codebook_dim,
+            quantizer_dropout=quantizer_dropout,
+        )
+
+        self.decoder = Decoder(
+            latent_dim,
+            decoder_dim,
+            decoder_rates,
+            lstm=lstm,
+            causal=causal,
+        )
+        self.sample_rate = sample_rate
+        self.apply(init_weights)
+
+        self.delay = self.get_delay()
+
+    def preprocess(self, audio_data, sample_rate):
+        if sample_rate is None:
+            sample_rate = self.sample_rate
+        assert sample_rate == self.sample_rate
+
+        length = audio_data.shape[-1]
+        right_pad = math.ceil(length / self.hop_length) * self.hop_length - length
+        audio_data = nn.functional.pad(audio_data, (0, right_pad))
+
+        return audio_data
+
+    def encode(
+        self,
+        audio_data: torch.Tensor,
+        n_quantizers: int = None,
+    ):
+        """Encode given audio data and return quantized latent codes
+
+        Parameters
+        ----------
+        audio_data : Tensor[B x 1 x T]
+            Audio data to encode
+        n_quantizers : int, optional
+            Number of quantizers to use, by default None
+            If None, all quantizers are used.
+
+        Returns
+        -------
+        dict
+            A dictionary with the following keys:
+            "z" : Tensor[B x D x T]
+                Quantized continuous representation of input
+            "codes" : Tensor[B x N x T]
+                Codebook indices for each codebook
+                (quantized discrete representation of input)
+            "latents" : Tensor[B x N*D x T]
+                Projected latents (continuous representation of input before quantization)
+            "vq/commitment_loss" : Tensor[1]
+                Commitment loss to train encoder to predict vectors closer to codebook
+                entries
+            "vq/codebook_loss" : Tensor[1]
+                Codebook loss to update the codebook
+            "length" : int
+                Number of samples in input audio
+        """
+        z = self.encoder(audio_data)
+        z, codes, latents, commitment_loss, codebook_loss = self.quantizer(
+            z, n_quantizers
+        )
+        return z, codes, latents, commitment_loss, codebook_loss
+
+    def decode(self, z: torch.Tensor):
+        """Decode given latent codes and return audio data
+
+        Parameters
+        ----------
+        z : Tensor[B x D x T]
+            Quantized continuous representation of input
+        length : int, optional
+            Number of samples in output audio, by default None
+
+        Returns
+        -------
+        dict
+            A dictionary with the following keys:
+            "audio" : Tensor[B x 1 x length]
+                Decoded audio data.
+        """
+        return self.decoder(z)
+
+    def forward(
+        self,
+        audio_data: torch.Tensor,
+        sample_rate: int = None,
+        n_quantizers: int = None,
+    ):
+        """Model forward pass
+
+        Parameters
+        ----------
+        audio_data : Tensor[B x 1 x T]
+            Audio data to encode
+        sample_rate : int, optional
+            Sample rate of audio data in Hz, by default None
+            If None, defaults to `self.sample_rate`
+        n_quantizers : int, optional
+            Number of quantizers to use, by default None.
+            If None, all quantizers are used.
+
+        Returns
+        -------
+        dict
+            A dictionary with the following keys:
+            "z" : Tensor[B x D x T]
+                Quantized continuous representation of input
+            "codes" : Tensor[B x N x T]
+                Codebook indices for each codebook
+                (quantized discrete representation of input)
+            "latents" : Tensor[B x N*D x T]
+                Projected latents (continuous representation of input before quantization)
+            "vq/commitment_loss" : Tensor[1]
+                Commitment loss to train encoder to predict vectors closer to codebook
+                entries
+            "vq/codebook_loss" : Tensor[1]
+                Codebook loss to update the codebook
+            "length" : int
+                Number of samples in input audio
+            "audio" : Tensor[B x 1 x length]
+                Decoded audio data.
+        """
+        length = audio_data.shape[-1]
+        audio_data = self.preprocess(audio_data, sample_rate)
+        z, codes, latents, commitment_loss, codebook_loss = self.encode(
+            audio_data, n_quantizers
+        )
+
+        x = self.decode(z)
+        return {
+            "audio": x[..., :length],
+            "z": z,
+            "codes": codes,
+            "latents": latents,
+            "vq/commitment_loss": commitment_loss,
+            "vq/codebook_loss": codebook_loss,
+        }
+
+
+if __name__ == "__main__":
+    import numpy as np
+    from functools import partial
+
+    model = DAC().to("cpu")
+
+    for n, m in model.named_modules():
+        o = m.extra_repr()
+        p = sum([np.prod(p.size()) for p in m.parameters()])
+        fn = lambda o, p: o + f" {p/1e6:<.3f}M params."
+        setattr(m, "extra_repr", partial(fn, o=o, p=p))
+    print(model)
+    print("Total # of params: ", sum([np.prod(p.size()) for p in model.parameters()]))
+
+    length = 88200 * 2
+    x = torch.randn(1, 1, length).to(model.device)
+    x.requires_grad_(True)
+    x.retain_grad()
+
+    # Make a forward pass
+    out = model(x)["audio"]
+    print("Input shape:", x.shape)
+    print("Output shape:", out.shape)
+
+    # Create gradient variable
+    grad = torch.zeros_like(out)
+    grad[:, :, grad.shape[-1] // 2] = 1
+
+    # Make a backward pass
+    out.backward(grad)
+
+    # Check non-zero values
+    gradmap = x.grad.squeeze(0)
+    gradmap = (gradmap != 0).sum(0)  # sum across features
+    rf = (gradmap != 0).sum()
+
+    print(f"Receptive field: {rf.item()}")
+
+    x = AudioSignal(torch.randn(1, 1, 44100 * 60), 44100)
+    model.decompress(model.compress(x, verbose=True), verbose=True)

qwen3-tts-backend/indextts/s2mel/dac/model/discriminator.py

@@ -0,0 +1,228 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from audiotools import AudioSignal
+from audiotools import ml
+from audiotools import STFTParams
+from einops import rearrange
+from torch.nn.utils import weight_norm
+
+
+def WNConv1d(*args, **kwargs):
+    act = kwargs.pop("act", True)
+    conv = weight_norm(nn.Conv1d(*args, **kwargs))
+    if not act:
+        return conv
+    return nn.Sequential(conv, nn.LeakyReLU(0.1))
+
+
+def WNConv2d(*args, **kwargs):
+    act = kwargs.pop("act", True)
+    conv = weight_norm(nn.Conv2d(*args, **kwargs))
+    if not act:
+        return conv
+    return nn.Sequential(conv, nn.LeakyReLU(0.1))
+
+
+class MPD(nn.Module):
+    def __init__(self, period):
+        super().__init__()
+        self.period = period
+        self.convs = nn.ModuleList(
+            [
+                WNConv2d(1, 32, (5, 1), (3, 1), padding=(2, 0)),
+                WNConv2d(32, 128, (5, 1), (3, 1), padding=(2, 0)),
+                WNConv2d(128, 512, (5, 1), (3, 1), padding=(2, 0)),
+                WNConv2d(512, 1024, (5, 1), (3, 1), padding=(2, 0)),
+                WNConv2d(1024, 1024, (5, 1), 1, padding=(2, 0)),
+            ]
+        )
+        self.conv_post = WNConv2d(
+            1024, 1, kernel_size=(3, 1), padding=(1, 0), act=False
+        )
+
+    def pad_to_period(self, x):
+        t = x.shape[-1]
+        x = F.pad(x, (0, self.period - t % self.period), mode="reflect")
+        return x
+
+    def forward(self, x):
+        fmap = []
+
+        x = self.pad_to_period(x)
+        x = rearrange(x, "b c (l p) -> b c l p", p=self.period)
+
+        for layer in self.convs:
+            x = layer(x)
+            fmap.append(x)
+
+        x = self.conv_post(x)
+        fmap.append(x)
+
+        return fmap
+
+
+class MSD(nn.Module):
+    def __init__(self, rate: int = 1, sample_rate: int = 44100):
+        super().__init__()
+        self.convs = nn.ModuleList(
+            [
+                WNConv1d(1, 16, 15, 1, padding=7),
+                WNConv1d(16, 64, 41, 4, groups=4, padding=20),
+                WNConv1d(64, 256, 41, 4, groups=16, padding=20),
+                WNConv1d(256, 1024, 41, 4, groups=64, padding=20),
+                WNConv1d(1024, 1024, 41, 4, groups=256, padding=20),
+                WNConv1d(1024, 1024, 5, 1, padding=2),
+            ]
+        )
+        self.conv_post = WNConv1d(1024, 1, 3, 1, padding=1, act=False)
+        self.sample_rate = sample_rate
+        self.rate = rate
+
+    def forward(self, x):
+        x = AudioSignal(x, self.sample_rate)
+        x.resample(self.sample_rate // self.rate)
+        x = x.audio_data
+
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+
+        return fmap
+
+
+BANDS = [(0.0, 0.1), (0.1, 0.25), (0.25, 0.5), (0.5, 0.75), (0.75, 1.0)]
+
+
+class MRD(nn.Module):
+    def __init__(
+        self,
+        window_length: int,
+        hop_factor: float = 0.25,
+        sample_rate: int = 44100,
+        bands: list = BANDS,
+    ):
+        """Complex multi-band spectrogram discriminator.
+        Parameters
+        ----------
+        window_length : int
+            Window length of STFT.
+        hop_factor : float, optional
+            Hop factor of the STFT, defaults to ``0.25 * window_length``.
+        sample_rate : int, optional
+            Sampling rate of audio in Hz, by default 44100
+        bands : list, optional
+            Bands to run discriminator over.
+        """
+        super().__init__()
+
+        self.window_length = window_length
+        self.hop_factor = hop_factor
+        self.sample_rate = sample_rate
+        self.stft_params = STFTParams(
+            window_length=window_length,
+            hop_length=int(window_length * hop_factor),
+            match_stride=True,
+        )
+
+        n_fft = window_length // 2 + 1
+        bands = [(int(b[0] * n_fft), int(b[1] * n_fft)) for b in bands]
+        self.bands = bands
+
+        ch = 32
+        convs = lambda: nn.ModuleList(
+            [
+                WNConv2d(2, ch, (3, 9), (1, 1), padding=(1, 4)),
+                WNConv2d(ch, ch, (3, 9), (1, 2), padding=(1, 4)),
+                WNConv2d(ch, ch, (3, 9), (1, 2), padding=(1, 4)),
+                WNConv2d(ch, ch, (3, 9), (1, 2), padding=(1, 4)),
+                WNConv2d(ch, ch, (3, 3), (1, 1), padding=(1, 1)),
+            ]
+        )
+        self.band_convs = nn.ModuleList([convs() for _ in range(len(self.bands))])
+        self.conv_post = WNConv2d(ch, 1, (3, 3), (1, 1), padding=(1, 1), act=False)
+
+    def spectrogram(self, x):
+        x = AudioSignal(x, self.sample_rate, stft_params=self.stft_params)
+        x = torch.view_as_real(x.stft())
+        x = rearrange(x, "b 1 f t c -> (b 1) c t f")
+        # Split into bands
+        x_bands = [x[..., b[0] : b[1]] for b in self.bands]
+        return x_bands
+
+    def forward(self, x):
+        x_bands = self.spectrogram(x)
+        fmap = []
+
+        x = []
+        for band, stack in zip(x_bands, self.band_convs):
+            for layer in stack:
+                band = layer(band)
+                fmap.append(band)
+            x.append(band)
+
+        x = torch.cat(x, dim=-1)
+        x = self.conv_post(x)
+        fmap.append(x)
+
+        return fmap
+
+
+class Discriminator(nn.Module):
+    def __init__(
+        self,
+        rates: list = [],
+        periods: list = [2, 3, 5, 7, 11],
+        fft_sizes: list = [2048, 1024, 512],
+        sample_rate: int = 44100,
+        bands: list = BANDS,
+    ):
+        """Discriminator that combines multiple discriminators.
+
+        Parameters
+        ----------
+        rates : list, optional
+            sampling rates (in Hz) to run MSD at, by default []
+            If empty, MSD is not used.
+        periods : list, optional
+            periods (of samples) to run MPD at, by default [2, 3, 5, 7, 11]
+        fft_sizes : list, optional
+            Window sizes of the FFT to run MRD at, by default [2048, 1024, 512]
+        sample_rate : int, optional
+            Sampling rate of audio in Hz, by default 44100
+        bands : list, optional
+            Bands to run MRD at, by default `BANDS`
+        """
+        super().__init__()
+        discs = []
+        discs += [MPD(p) for p in periods]
+        discs += [MSD(r, sample_rate=sample_rate) for r in rates]
+        discs += [MRD(f, sample_rate=sample_rate, bands=bands) for f in fft_sizes]
+        self.discriminators = nn.ModuleList(discs)
+
+    def preprocess(self, y):
+        # Remove DC offset
+        y = y - y.mean(dim=-1, keepdims=True)
+        # Peak normalize the volume of input audio
+        y = 0.8 * y / (y.abs().max(dim=-1, keepdim=True)[0] + 1e-9)
+        return y
+
+    def forward(self, x):
+        x = self.preprocess(x)
+        fmaps = [d(x) for d in self.discriminators]
+        return fmaps
+
+
+if __name__ == "__main__":
+    disc = Discriminator()
+    x = torch.zeros(1, 1, 44100)
+    results = disc(x)
+    for i, result in enumerate(results):
+        print(f"disc{i}")
+        for i, r in enumerate(result):
+            print(r.shape, r.mean(), r.min(), r.max())
+        print()

qwen3-tts-backend/indextts/s2mel/dac/model/encodec.py

@@ -0,0 +1,320 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""Convolutional layers wrappers and utilities."""
+
+import math
+import typing as tp
+import warnings
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from torch.nn.utils import spectral_norm, weight_norm
+
+import typing as tp
+
+import einops
+
+
+class ConvLayerNorm(nn.LayerNorm):
+    """
+    Convolution-friendly LayerNorm that moves channels to last dimensions
+    before running the normalization and moves them back to original position right after.
+    """
+    def __init__(self, normalized_shape: tp.Union[int, tp.List[int], torch.Size], **kwargs):
+        super().__init__(normalized_shape, **kwargs)
+
+    def forward(self, x):
+        x = einops.rearrange(x, 'b ... t -> b t ...')
+        x = super().forward(x)
+        x = einops.rearrange(x, 'b t ... -> b ... t')
+        return
+
+
+CONV_NORMALIZATIONS = frozenset(['none', 'weight_norm', 'spectral_norm',
+                                 'time_layer_norm', 'layer_norm', 'time_group_norm'])
+
+
+def apply_parametrization_norm(module: nn.Module, norm: str = 'none') -> nn.Module:
+    assert norm in CONV_NORMALIZATIONS
+    if norm == 'weight_norm':
+        return weight_norm(module)
+    elif norm == 'spectral_norm':
+        return spectral_norm(module)
+    else:
+        # We already check was in CONV_NORMALIZATION, so any other choice
+        # doesn't need reparametrization.
+        return module
+
+
+def get_norm_module(module: nn.Module, causal: bool = False, norm: str = 'none', **norm_kwargs) -> nn.Module:
+    """Return the proper normalization module. If causal is True, this will ensure the returned
+    module is causal, or return an error if the normalization doesn't support causal evaluation.
+    """
+    assert norm in CONV_NORMALIZATIONS
+    if norm == 'layer_norm':
+        assert isinstance(module, nn.modules.conv._ConvNd)
+        return ConvLayerNorm(module.out_channels, **norm_kwargs)
+    elif norm == 'time_group_norm':
+        if causal:
+            raise ValueError("GroupNorm doesn't support causal evaluation.")
+        assert isinstance(module, nn.modules.conv._ConvNd)
+        return nn.GroupNorm(1, module.out_channels, **norm_kwargs)
+    else:
+        return nn.Identity()
+
+
+def get_extra_padding_for_conv1d(x: torch.Tensor, kernel_size: int, stride: int,
+                                 padding_total: int = 0) -> int:
+    """See `pad_for_conv1d`.
+    """
+    length = x.shape[-1]
+    n_frames = (length - kernel_size + padding_total) / stride + 1
+    ideal_length = (math.ceil(n_frames) - 1) * stride + (kernel_size - padding_total)
+    return ideal_length - length
+
+
+def pad_for_conv1d(x: torch.Tensor, kernel_size: int, stride: int, padding_total: int = 0):
+    """Pad for a convolution to make sure that the last window is full.
+    Extra padding is added at the end. This is required to ensure that we can rebuild
+    an output of the same length, as otherwise, even with padding, some time steps
+    might get removed.
+    For instance, with total padding = 4, kernel size = 4, stride = 2:
+        0 0 1 2 3 4 5 0 0   # (0s are padding)
+        1   2   3           # (output frames of a convolution, last 0 is never used)
+        0 0 1 2 3 4 5 0     # (output of tr. conv., but pos. 5 is going to get removed as padding)
+            1 2 3 4         # once you removed padding, we are missing one time step !
+    """
+    extra_padding = get_extra_padding_for_conv1d(x, kernel_size, stride, padding_total)
+    return F.pad(x, (0, extra_padding))
+
+
+def pad1d(x: torch.Tensor, paddings: tp.Tuple[int, int], mode: str = 'zero', value: float = 0.):
+    """Tiny wrapper around F.pad, just to allow for reflect padding on small input.
+    If this is the case, we insert extra 0 padding to the right before the reflection happen.
+    """
+    length = x.shape[-1]
+    padding_left, padding_right = paddings
+    assert padding_left >= 0 and padding_right >= 0, (padding_left, padding_right)
+    if mode == 'reflect':
+        max_pad = max(padding_left, padding_right)
+        extra_pad = 0
+        if length <= max_pad:
+            extra_pad = max_pad - length + 1
+            x = F.pad(x, (0, extra_pad))
+        padded = F.pad(x, paddings, mode, value)
+        end = padded.shape[-1] - extra_pad
+        return padded[..., :end]
+    else:
+        return F.pad(x, paddings, mode, value)
+
+
+def unpad1d(x: torch.Tensor, paddings: tp.Tuple[int, int]):
+    """Remove padding from x, handling properly zero padding. Only for 1d!"""
+    padding_left, padding_right = paddings
+    assert padding_left >= 0 and padding_right >= 0, (padding_left, padding_right)
+    assert (padding_left + padding_right) <= x.shape[-1]
+    end = x.shape[-1] - padding_right
+    return x[..., padding_left: end]
+
+
+class NormConv1d(nn.Module):
+    """Wrapper around Conv1d and normalization applied to this conv
+    to provide a uniform interface across normalization approaches.
+    """
+    def __init__(self, *args, causal: bool = False, norm: str = 'none',
+                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
+        super().__init__()
+        self.conv = apply_parametrization_norm(nn.Conv1d(*args, **kwargs), norm)
+        self.norm = get_norm_module(self.conv, causal, norm, **norm_kwargs)
+        self.norm_type = norm
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.norm(x)
+        return x
+
+
+class NormConv2d(nn.Module):
+    """Wrapper around Conv2d and normalization applied to this conv
+    to provide a uniform interface across normalization approaches.
+    """
+    def __init__(self, *args, norm: str = 'none',
+                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
+        super().__init__()
+        self.conv = apply_parametrization_norm(nn.Conv2d(*args, **kwargs), norm)
+        self.norm = get_norm_module(self.conv, causal=False, norm=norm, **norm_kwargs)
+        self.norm_type = norm
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.norm(x)
+        return x
+
+
+class NormConvTranspose1d(nn.Module):
+    """Wrapper around ConvTranspose1d and normalization applied to this conv
+    to provide a uniform interface across normalization approaches.
+    """
+    def __init__(self, *args, causal: bool = False, norm: str = 'none',
+                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
+        super().__init__()
+        self.convtr = apply_parametrization_norm(nn.ConvTranspose1d(*args, **kwargs), norm)
+        self.norm = get_norm_module(self.convtr, causal, norm, **norm_kwargs)
+        self.norm_type = norm
+
+    def forward(self, x):
+        x = self.convtr(x)
+        x = self.norm(x)
+        return x
+
+
+class NormConvTranspose2d(nn.Module):
+    """Wrapper around ConvTranspose2d and normalization applied to this conv
+    to provide a uniform interface across normalization approaches.
+    """
+    def __init__(self, *args, norm: str = 'none',
+                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
+        super().__init__()
+        self.convtr = apply_parametrization_norm(nn.ConvTranspose2d(*args, **kwargs), norm)
+        self.norm = get_norm_module(self.convtr, causal=False, norm=norm, **norm_kwargs)
+
+    def forward(self, x):
+        x = self.convtr(x)
+        x = self.norm(x)
+        return x
+
+
+class SConv1d(nn.Module):
+    """Conv1d with some builtin handling of asymmetric or causal padding
+    and normalization.
+    """
+    def __init__(self, in_channels: int, out_channels: int,
+                 kernel_size: int, stride: int = 1, dilation: int = 1,
+                 groups: int = 1, bias: bool = True, causal: bool = False,
+                 norm: str = 'none', norm_kwargs: tp.Dict[str, tp.Any] = {},
+                 pad_mode: str = 'reflect', **kwargs):
+        super().__init__()
+        # warn user on unusual setup between dilation and stride
+        if stride > 1 and dilation > 1:
+            warnings.warn('SConv1d has been initialized with stride > 1 and dilation > 1'
+                          f' (kernel_size={kernel_size} stride={stride}, dilation={dilation}).')
+        self.conv = NormConv1d(in_channels, out_channels, kernel_size, stride,
+                               dilation=dilation, groups=groups, bias=bias, causal=causal,
+                               norm=norm, norm_kwargs=norm_kwargs)
+        self.causal = causal
+        self.pad_mode = pad_mode
+
+        self.cache_enabled = False
+
+    def reset_cache(self):
+        """Reset the cache when starting a new stream."""
+        self.cache = None
+        self.cache_enabled = True
+
+    def forward(self, x):
+        B, C, T = x.shape
+        kernel_size = self.conv.conv.kernel_size[0]
+        stride = self.conv.conv.stride[0]
+        dilation = self.conv.conv.dilation[0]
+        kernel_size = (kernel_size - 1) * dilation + 1  # effective kernel size with dilations
+        padding_total = kernel_size - stride
+        extra_padding = get_extra_padding_for_conv1d(x, kernel_size, stride, padding_total)
+
+        if self.causal:
+            # Left padding for causal
+            if self.cache_enabled and self.cache is not None:
+                # Concatenate the cache (previous inputs) with the new input for streaming
+                x = torch.cat([self.cache, x], dim=2)
+            else:
+                x = pad1d(x, (padding_total, extra_padding), mode=self.pad_mode)
+        else:
+            # Asymmetric padding required for odd strides
+            padding_right = padding_total // 2
+            padding_left = padding_total - padding_right
+            x = pad1d(x, (padding_left, padding_right + extra_padding), mode=self.pad_mode)
+
+        # Store the most recent input frames for future cache use
+        if self.cache_enabled:
+            if self.cache is None:
+                # Initialize cache with zeros (at the start of streaming)
+                self.cache = torch.zeros(B, C, kernel_size - 1, device=x.device)
+            # Update the cache by storing the latest input frames
+            if kernel_size > 1:
+                self.cache = x[:, :, -kernel_size + 1:].detach()  # Only store the necessary frames
+
+        return self.conv(x)
+
+
+
+class SConvTranspose1d(nn.Module):
+    """ConvTranspose1d with some builtin handling of asymmetric or causal padding
+    and normalization.
+    """
+    def __init__(self, in_channels: int, out_channels: int,
+                 kernel_size: int, stride: int = 1, causal: bool = False,
+                 norm: str = 'none', trim_right_ratio: float = 1.,
+                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
+        super().__init__()
+        self.convtr = NormConvTranspose1d(in_channels, out_channels, kernel_size, stride,
+                                          causal=causal, norm=norm, norm_kwargs=norm_kwargs)
+        self.causal = causal
+        self.trim_right_ratio = trim_right_ratio
+        assert self.causal or self.trim_right_ratio == 1., \
+            "`trim_right_ratio` != 1.0 only makes sense for causal convolutions"
+        assert self.trim_right_ratio >= 0. and self.trim_right_ratio <= 1.
+
+    def forward(self, x):
+        kernel_size = self.convtr.convtr.kernel_size[0]
+        stride = self.convtr.convtr.stride[0]
+        padding_total = kernel_size - stride
+
+        y = self.convtr(x)
+
+        # We will only trim fixed padding. Extra padding from `pad_for_conv1d` would be
+        # removed at the very end, when keeping only the right length for the output,
+        # as removing it here would require also passing the length at the matching layer
+        # in the encoder.
+        if self.causal:
+            # Trim the padding on the right according to the specified ratio
+            # if trim_right_ratio = 1.0, trim everything from right
+            padding_right = math.ceil(padding_total * self.trim_right_ratio)
+            padding_left = padding_total - padding_right
+            y = unpad1d(y, (padding_left, padding_right))
+        else:
+            # Asymmetric padding required for odd strides
+            padding_right = padding_total // 2
+            padding_left = padding_total - padding_right
+            y = unpad1d(y, (padding_left, padding_right))
+        return y
+
+class SLSTM(nn.Module):
+    """
+    LSTM without worrying about the hidden state, nor the layout of the data.
+    Expects input as convolutional layout.
+    """
+    def __init__(self, dimension: int, num_layers: int = 2, skip: bool = True):
+        super().__init__()
+        self.skip = skip
+        self.lstm = nn.LSTM(dimension, dimension, num_layers)
+        self.hidden = None
+        self.cache_enabled = False
+
+    def forward(self, x):
+        x = x.permute(2, 0, 1)
+        if self.training or not self.cache_enabled:
+            y, _ = self.lstm(x)
+        else:
+            y, self.hidden = self.lstm(x, self.hidden)
+        if self.skip:
+            y = y + x
+        y = y.permute(1, 2, 0)
+        return y
+
+    def reset_cache(self):
+        self.hidden = None
+        self.cache_enabled = True