Convert applicable model parameters to fp16

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). This is the same as the DropConnect impl I created for Ef

Adapted from https://github.com/XinhaoMei/audio-text_retrieval/blob/main/tools/utils.py. 1. for text-to-audio retrieval, do 5 times and avera

r"""Convert feature maps to waveform. Args: input_tensor: (batch_size, target_sources_num * output_channels * self.K, time_steps,

r"""Get active frames.

r"""Create data_module. Mini-batch data can be obtained by: code-block:: python data_module.setup() for batch_data_dict in data

r"""Get directories and paths. Args: workspace (str): directory of workspace filename (str): filename of current .py file.

r"""Get learning scheduler. Args: lr_lambda_type (str), e.g., "constant_warm_up" | "linear_warm_up" Returns: lr_lambda_func

r"""Initialize weights.

r"""Initialize weights.

Returns the names of available CLIP models

return all models having the specified pretrain tag

Load a CLIP model, preserve its text pretrained part, and set in the CLAP model Parameters ---------- name : str A model name lis

Convert a NumPy array to a PyTorch tensor.

Sample a proportion of the data.

Convert a PyTorch tensor to a NumPy array.

r"""Train, evaluate, and save checkpoints. Args: workspace: str, directory of workspace gpus: int, number of GPUs to train

Record the top-k performance of the current epoch. current_top_k_metrics is a dictionary of the form: {1: top_1_ckpt_measure, 2: top_2_ckpt_m

Record the top-k performance of the current epoch. current_top_k_metrics is a dictionary of the form: {1: top_1_ckpt_measure, 2: top_2_ckpt_m

Waveform to spectrogram. Args: input: (batch_size, segment_samples, channels_num) Outputs: output: (batch_size,

Args: windows: (num_windows*B, window_size, window_size, C) window_size (int): Window size H (int): Height of image

r"""Evalute.

r"""Evalute.

r"""Evalute.

r"""Evalute.

r"""Evalute.

r"""Evalute.

Load waveform, text, and target of an audio clip Parameters ---------- index: int the index number

r"""Save a checkpoint every N steps. Args: checkpoints_dir (str): directory to save checkpoints save_step_frequency (

r"""Data module. To get one batch of data: code-block:: python data_module.setup() for batch_data_dict in data_modu

r"""Pytorch Lightning wrapper of PyTorch model, including forward, optimization of model, etc. Args: ss_model: nn.Module

r"""Residual block.

r"""Encoder block, contains 8 convolutional layers.

r"""Decoder block, contains 1 transposed convolutional and 8 convolutional layers.