espnet2.train.dataset.ESPnetSpeechLMDataset

source

class espnet2.train.dataset.ESPnetSpeechLMDataset(example_list: List, task: str, **kwargs)

Bases: ESPnetDataset

Dataset object that is specifically designed for SpeechLM. It will allows dataset-level operations (e.g., on-the-fly speaker prompt sampling). It is task-specific and can be queried by ESPnetMultiTaskDataset.

install_speaker_prompt(uid: str, data: Dict)

Assume the names are utt2spk and wav.scp. Hard code here.