Helper function to instantiate BatchSampler. Args: type: mini-batch type. "unsorted", "sorted", "folded", "numel", "length",

Build a factory object of mini-batch iterator. This object is invoked at every epochs to build the iterator for each epoch as followi

Perform on collect_stats mode. Running for deriving the shape information from data and gathering statistics. This method is used before

Create decoder hidden states. Args: states: Decoder hidden states. ((N, B, D_dec), (N, B, D_dec)) new_states: Decoder

Frontend + Encoder. Note that this method is used by asr_inference.py Args: speech: (Batch, Length, ...) speech_lengt

Encode pre-processed batch data for GPU-based batch processing. This method handles data that has already been processed into proper

Util to format and print the results. Format results in tabular format. :param results: the dictionary output by the get_metric() method

Find free port using bind(). There are some interval between finding this port and using it and the other process might catch the port by tha

Executes a command and returns its stdout output as a string. The command is executed with shell=True, so it may contain pipes and o

This static method allows to build a Distance object. :param distance: The distance to be returned. :return: A `Distance` ob

Computes the metrics starting from TPs, FPs, and FNs. :return: A Dict per label containing a tuple for\ precision, r

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

Calculate the relative transfer function (RTF) Algorithm of power method: 1) rtf = reference_vector 2) for i in range(iterations)

Run inference. Args: text (Tensor): Input text index tensor (B, T_text,). text_lengths (Tensor): Text length tensor (

Initialize decoder states. Args: batch_size: Batch size. Returns: : Initial decoder hidden states. ((N, B, D

Initialize decoder states. Args: batch_size: Batch size. Returns: : Initial decoder hidden states. ((N, B, D

Instantiate a single attention module. :param str atype: attention type :param int eprojs: # projection-units of encoder :param int dunit

remove redundant whitespaces

Invoke stage methods on ``system`` in order with logging and timing. Args: system: Object providing stage methods named in ``stages_to_ru

Select state with relative ids in the main beam search. Args: state: Decoder state for prefix tokens i (int): Index t

Set up the worker by calling the user-defined setup function.

Return the epoch which resulted the best value. Example: >>> val = reporter.sort_epochs_and_values('eval', 'loss', 'min')

read lines from text or write lines to txt :param textpath: filepath of text :param lines_content: list of lines or None, None means read

read lines from text or write lines to txt :param textpath: filepath of text :param lines_content: list of lines or None, None means read

Initialize HiFiGANGenerator module. Args: in_channels (int): Number of input channels. out_channels (int): Number of

Apply delay interleaving to multi-stream tokens. Each stream is delayed by its index number of frames: - Stream 0: no delay -

Log a message using the logger if provided; otherwise print.

Parses the string option to --trainer.dropout-schedule and returns a list of dropout schedules for different component name patterns. Calls _p

Run a single forward step through the model and produce logits. Accepts either token IDs or precomputed embeddings (exactly one must

Generate averaged model from n-best models Args: output_dir: The directory contains the model file for each epoch reporter: Repor

Obtain initialization values for the mel scale. Args: channels: Number of channels. fs: Sample rate. Returns

Return "collate_fn", which is a callable object and given to DataLoader. >>> from torch.utils.data import DataLoader >>> loader = Dat

Build the data collation function for SpeechLM. Returns: A callable function for collating SpeechLM batch data.

Check if the dataset satisfy the requirement of current Task

Create LoRA adapter for the base model. See: https://arxiv.org/pdf/2106.09685.pdf Args: model (torch.nn.Module): Base model to be ad

3x3 convolution with DDPM initialization.

Generate a list of long utterances from a list of short utterances.