Make mask tensor containing indices of padded part. See description of make_non_pad_mask. Args: lengths (torch.Tensor): Batch of len

Flushes the remaining partial uint8, call this at the end of the stream to encode.

Construct dataset from arguments We have two shuffle stage in the Dataset. The first is global shuffle at shards tar/raw file level.

Instantiates a class with the given args and init. Args: args: Positional arguments required for instantiation. init: Dict of the

Instantiates a class with the given args and init. Args: args: Positional arguments required for instantiation. init: Dict of the

Computes the element-wise logarithm of the input tensor with clipping to avoid near-zero values. Args: x (Tensor): Input tensor.

Construct an PositionalEncoding object.

Class method to create a new Vocos model instance from a pre-trained model stored in the Hugging Face model hub.

Log a scalar variable.

Log a scalar variable.

Class method to create a new Vocos model instance from a pre-trained model stored in the Hugging Face model hub.

Return the proper normalization module. If causal is True, this will ensure the returned module is causal, or return an error if the normalization

Plot a spectrogram and convert it to a numpy array. Args: spectrogram (ndarray): Spectrogram data. Returns: ndarray: Nu

Forward one step. This is only used for decoding. Args: memory: encoded memory, float32 (batch, maxlen_in, feat)

Return bandwidth per quantizer for a given input frame rate. Each quantizer encodes a frame with lg(bins) bits.

Push the given symbol on the stream, flushing out bits if possible. Args: symbol (int): symbol to encode with the AC.

Static batch the data by `batch_size` Args: data: Iterable[{key, feat, label}] batch_size: batch size Retur

Args: input_size (int): input dim output_size (int): dimension of attention attention_heads (int): the nu

Simple const lr policy

Turn the given PDF into a quantized CDF that splits [0, 2 ** self.total_range_bits - 1] into chunks of size roughly proportional to the PDF.

Transforms an input sequence of discrete tokens (codes) into feature embeddings using the feature extractor's codebook weights.

Cross validation on

Decode the given frames into a waveform. Note that the output might be a bit bigger than the input. In that case, any extra steps at t

Given a tensor `x`, returns a list of frames containing the discrete encoded codes for `x`, along with rescaling factors for each segm

Reset the positional encodings.

Given a list of files, compute their embeddings in batches. If channels == 1: stereo audio is downmixed to mono. Mono embeddings are of dim=

Given a list of files, compute their embeddings one by one. If channels == 1: stereo audio is downmixed to mono. Mono embeddings are of dim=

Apply a linear fade-out effect to the given audio waveform. Parameters: audio (torch.Tensor): The audio waveform tensor. sample_rate

Compute attention context vector. Args: value (torch.Tensor): Transformed value, size (#batch, n_head, time2, d_k

Transform query, key and value. Args: query (torch.Tensor): Query tensor (#batch, time1, size). key (torch.Tensor): K

Class method to create a new Vocos model instance from hyperparameters stored in a yaml configuration file.

Class method to create a new Vocos model instance from hyperparameters stored in a yaml configuration file.