espnet.utils.training.batchfy.batchfy_by_bin

source

espnet.utils.training.batchfy.batchfy_by_bin(sorted_data, batch_bins, num_batches=0, min_batch_size=1, shortest_first=False, ikey='input', okey='output')

Make variably sized batch set, which maximizes.

the number of bins up to batch_bins.

• Parameters:
  • sorted_data (Dict *[*str , Dict *[*str , Any ] ]) – dictionary loaded from data.json
  • batch_bins (int) – Maximum frames of a batch
  • num_batches (int) – # number of batches to use (for debug)
  • min_batch_size (int) – minimum batch size (for multi-gpu)
  • test (int) – Return only every test batches
  • shortest_first (bool) – Sort from batch with shortest samples to longest if true, otherwise reverse
  • ikey (str) – key to access input (for ASR ikey=”input”, for TTS ikey=”output”.)
  • okey (str) – key to access output (for ASR okey=”output”. for TTS okey=”input”.)
• Returns: List[Tuple[str, Dict[str, List[Dict[str, Any]]]] list of batches