Initialize device based on the provided arguments. Args: args (argparse.Namespace): Parsed arguments. Returns: str: Dev

Initialize runtime environment settings. Sets random seeds for reproducibility, adjusts PyTorch cuDNN behavior, configures multiprocessi

Move a model to the correct computation device and wrap with DataParallel if needed. Selects GPU(s) if CUDA is available; otherwise defaults

Normalize SDR value using bounded logarithmic scaling.

L1-SNR loss with dB-scale level regularization. Extends L1-SNR with adaptive level-matching regularization in dB scale. From torch-l1-sn

L1-SNR loss in time domain. L1-based signal-to-noise ratio loss (without additional regularization). From torch-l1-snr package. Arg

1) Randomly choose instrument (class) 2) Randomly choose track containing this instrument 3) Load aligned chunk from this tra

Load LoRA weights into a model. This function updates the given model with LoRA-specific weights from the specified checkpoint file. It d

Load the model and configuration from the given arguments. Args: args (argparse.Namespace): Parsed arguments. device (str):

Log comprehensive model information

Updates configuration files in the `configs` directory for use with test data. This function processes configuration files defined in the gl

Combined time + STFT domain L1-SNR loss. Balances time-domain and spectral-domain L1-SNR with optional regularization. This is the recom

Compute a (multi-resolution) STFT-based loss on waveforms. Reshapes inputs to (B, C*T, L) when needed and delegates to a provided multi-

Apply mean-variance normalization to a pair of tensors. Computes the mean and standard deviation from `x` and normalizes both `x` and `y

Stream the output audio to the speakers. Args: waveforms_orig (Dict[str, np.ndarray]): Dictionary containing the separat

Parse command-line arguments for configuring the model, dataset, and training parameters. Args: dict_args: Dict of command-line argu

Parse command-line arguments. Args: dict_args (Union[Dict[str, Any], None]): Optional dictionary of arguments

Parse command-line arguments for configuring the model, dataset, and training parameters. Args: dict_args: Dict of command-line argu

Parse command-line arguments for configuring the model, dataset, and training parameters. Args: dict_args: Dict of command-line argu

Parse command-line arguments for configuring the model, dataset, and training parameters. Args: dict_args: Dict of command-line argu

Parse command-line arguments for inference configuration. Builds the CLI for model selection, configuration path, input/output handling,

Process the recorded audio using the model and save the output. Args: audio_data (bytes): The recorded audio data.

Processes a folder containing audio tracks, copying and combining necessary files into the target structure. Parameters: - src_folder (st

Continuously record audio from the input stream, process it using the model, and stream the output to the speakers in real-time.

Record audio from the input stream until the user presses 'S' to stop. Args: stream_input (pyaudio.Stream): The input au

Process a folder of audio files for source separation. Parameters: ---------- model : torch.nn.Module Pre-trained model for

Run parallel validation using multiple processes. Each process handles a subset of the mixture files and computes the metrics. The results

Executes validation tests for all configurations. This function updates configurations, generates random dummy data for testing, and run

Save a configuration object (ConfigDict or OmegaConf) to a file. Parameters: ---------- config : Union[ConfigDict, OmegaConf]

Save the latest training checkpoint for continuation or recovery. The checkpoint is always written to: {args.results_path}/last_{arg

Initialize a Distributed Data Parallel (DDP) process group. Configures environment variables for the DDP master node, attempts to initia

Compute Scale-Invariant Signal-to-Distortion Ratio (SI-SDR) for one or more audio tracks. SI-SDR is a variant of the SDR metric that is inva

Quantile-masked MSE in the complex STFT domain. Computes a complex STFT for prediction and target, forms an elementwise MSE in the spect

RMSE in the complex STFT domain. Computes STFT for prediction and target, represents complex values as real+imag pairs, and applies RMSE

Build :class:`torchaudio.prototype.models.SquimObjective` model with default arguments.

Build a custome :class:`torchaudio.prototype.models.SquimObjective` model. Args: feat_dim (int, optional): The feature dimension after En

L1-SNR loss in multi-resolution STFT domain. Applies L1-SNR to complex STFT (real/imaginary) across multiple resolutions. From torch-l1-

Train the model for one epoch. Args: world_size: scheduler: model: The model to train. config: Configuration

The metric defined by ITU-T P.862 is often called 'PESQ score', which is defined for narrow-band signals and has a value range of [-0.5, 4.5] exac

Main function for type 1 script, which records audio, processes it, and saves the output. Args: args (argparse.Namespace): Parsed ar

Main function for type 2 script, which continuously records audio, processes it, and streams the output in real-time. Args: args