Return the list of target instruments based on the configuration. If a specific target instrument is specified in the configuration,

Returns array with shape (target_channels, chunk_size)

Load and instantiate a model using a configuration file. Given a `model_type` and a path to a configuration, this function loads the con

Combines multiple audio files into one by overlaying them. Parameters: - files (List[str]): List of file paths to be combined. Return

Perform audio source separation with a given model. Supports both Demucs-specific and generic processing modes, including overlapping ch

Validate a separator model across multiple GPUs with a unified API. Runs validation either in Distributed Data Parallel (DDP) mode—detect

Load an initial checkpoint into `model`. For `type_ == "train"`, performs a tolerant load using `old_model` (a state dict or a checkpoin

Enhance source separation results using Test-Time Augmentation (TTA). Applies augmentations such as channel reversal and polarity inversion

r"""Get the adjust learning rate.

Print a log message and optionally append it to an in-memory list. In Distributed Data Parallel (DDP) contexts, the message is printed only

Initializes SDLayer with subband interval, input dimension, output dimension, downsample stride, number of convolutional modules, ker

Initializes SULayer with input dimension, output dimension, upsample stride, subband shape, and subband interval.

Create branch module after DPRNN model for predicting metric score. Args: d_model (int): The number of expected features in the input.

Precompute all chunks and save to cache with config

Compute average metrics across instruments (DDP-aware) and optionally log to file. For each metric, computes the mean value per instrumen

Create intervals based on splits provided. Args: - splits (List[Union[float, int]]): List of floats or integers representing splits.

Load a model configuration from a file. Based on `model_type`, returns either an OmegaConf (e.g., for 'htdemucs') or a YAML-parsed Confi

Initialize random seeds for reproducibility. Sets the seed across Python's `random`, NumPy, and PyTorch (CPU and CUDA) libraries, and up

Run source separation on a list of mixtures and collect evaluation metrics. Performs optional resampling and normalization, demixes each

Read an audio file and return transposed waveform data with channels first. Loads the audio file from `path`, converts mono signals to 2D fo