Round with straight through gradients.

Perform TTS inference and save the generated audio.

Turn special token of task.

tokenize the input mel spectrogram

Args: dim: input dimension groups: number of groups upsample_scale: upsampling scale downsamp

Initializes the BiCodec model with the required components. Args: mel_params (dict): Parameters for the mel-spectrogram

Initializes the SparkTTS model with the provided configurations and device. Args: model_dir (Path): Directory containing

apply highpass fileter to audio Args: audio (np.ndarray): sample_rate (ind): highpass_cutoff_freq (int):

detokenize the input indices

Execute inference on the batched requests. Args: requests: List of inference requests Returns:

Execute inference on the batched requests. Args: requests: List of inference requests Returns:

Execute inference on the batched requests. Args: requests: List of inference requests Returns:

einstein notation b - batch n - sequence (or flattened spatial dimensions) d - feature dimension c - number o

Args: x (torch.Tensor): (batch_size, input_channels, length) Returns: x (torch.Tensor): (batch_size, encode_

encoder forward. Args: x (torch.Tensor): (batch_size, input_channels, length) Returns: x (torch.Tensor): (ba

einstein notation b - batch h - heads n, i, j - sequence length (base sequence length, source, target) d - fe

Args: mels: (B, D_mel, T1) Return: x_vector: (B, out_dim) d_vector: (B, out_dim)

x: a 3-dimensional tensor in tdnn-based architecture (B,F,T) or a 4-dimensional tensor in resnet architecture (B,C,F,T)

input: a 3-dimensional tensor in xvector architecture or a 4-dimensional tensor in resnet architecture 0-dim: batch-d

input: a 3-dimensional tensor in xvector architecture or a 4-dimensional tensor in resnet architecture 0-dim: batch-d

Quantized the input tensor using a fixed codebook and returns the corresponding codebook vectors Parameters ----------

Args: x (Tensor): Input tensor of shape (B, C, L), where B is the batch size, C denotes output features,

Performs a forward pass through the model. Args: batch (dict): A dictionary containing features, reference waveform, and