Returns keys1 e.g. train,eval.

Get Node Rank. Use for "multiprocessing distributed" mode. The initial RANK equals to the Node id in this case and the real Rank is set a

Calculate the relative transfer function (RTF). Args: psd_speech (torch.complex64): speech covariance matrix (..., F, C, C)

Handles the selection and initialization of ASR model. This method dynamically loads the selected ASR based on the provided option.

Handles the selection and initialization of a LLM. This method dynamically loads the selected LLM based on the provided option.

Handles the selection and initialization of a Text-to-Speech (TTS) model. This method dynamically loads the selected TTS model based

Handles the selection of the spoken dialogue model type (Cascaded or E2E) and dynamically updates the interface based on the selected option.

Infer the default artifact type from a Python value.

Generate the sequence of features given the sequences of characters. Args: text (LongTensor): Input sequence of characters (T_tex

Initialize tmp state.

Convert mel to Hz.

Check if value is a tuple or not.

Process invert stft.

Load lines from a text file and return as a list of strings. This function is used as a custom resolver in OmegaConf, allowing YAML files to

Load a model state and set it to the model. Args: init_param: <file_path>:<src_key>:<dst_key>:<exclude_Keys> Examples: >>> l

Log a component instance with class info, repr, and attributes. Description: Emits a structured log block for a single object. The block

Get number of parallel streams (SSL + codec). Returns: Total number of streams combining SSL and codec

This functions picks an item from the collection according to the associated probability distribution. The probability estimate of each item

Perform post-processing of beam search iterations. Args: i (int): The length of hypothesis tokens. maxlen (int): The

Get specified ID state from decoder hidden states. Args: states: Decoder hidden states. ((N, B, D_dec), (N, B, D_dec) or None)

Set config file for streaming decoding. Args: asr_config (str): The config file for asr training

Expand `signal` into several frames, with each frame of length `frame_length`. Args: signal : (..., T) frame_length: length of

This function smooths the probability distribution in the list

Sort hypotheses by score or score given sequence length. Args: hyps: Hypothesis. Return: hyps: Sorted hypoth

Sort hypotheses by score or score given sequence length. Args: hyps: Hypothesis. Return: hyps: Sorted hypoth

Create loss mask for special tokens (1 frame, multi-stream). Only first stream has the actual value, others are zero.

WPE Args: Y: Complex valued STFT signal with shape (F, C, T) taps: Number of filter taps delay: Delay as a guard interval

Write a single inference artifact to disk and return its path. This function implements the artifact serialization rules used by ``espnet3.sy

Compute and return speech quality metrics for the given synthesized audio output using the Versa library. Args: TTS_audio_ou

Grouped LSTM. Reference: Efficient Sequence Learning with Group Recurrent Networks; Gao et al., 2018 Args: h

Band-Split RNN (BSRNN). References: [1] J. Yu, H. Chen, Y. Luo, R. Gu, and C. Weng, “High fidelity speech enhancement

Perform async all_reduce on statistics for efficient multi-GPU sync. Args: stats: Dictionary of statistics tensors to reduce acro

Worker: sort (length, key) items and run Best Fit Decreasing.

Collate all corpus data. Collate all specified corpus data. This is a hack to call the Kaldi collation scripts, which are required b

Create embeddings from multimodal inputs. Handles both discrete tokens (encoded on-the-fly) and continuous features (projecte

Build KmeansModel without joblib, using manual buffer registration.

Return the test-set names to score for the measurement stage.

Run a git command and return stdout, or None on failure.

Make masks for self-attention. Args: ilens (LongTensor): Batch of lengths (B,). Returns: Tensor: Mask tensor

Convert data to JSON-friendly dict.

Assemble the hypotheses.

Average frame-level features into token-level according to durations Args: ds (Tensor): Batched token duration (B, T_text). xs (T

Decode a batch of audios (either short-form or long-form).

Build a batch index ordering for a given seed. Args: seed: Random seed for shuffling. Different seeds produce dif