espnet2.asr.frontend.huggingface.HuggingFaceFrontend

source

class espnet2.asr.frontend.huggingface.HuggingFaceFrontend(model, fs: int | str = 16000, download_dir: str | None = None, load_pretrained: bool = True)

Bases: AbsFrontend

Use pretrained models from Hugging Face Transformers for ASR

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(inputs: Tensor, input_lengths: Tensor) → Tuple[Tensor, Tensor]

Wrapper for the transformers forward pass. Inputs are converted to numpy and re-encoded with the transformers processor.

• Parameters:

  • input – Input (B, L) single channel waveform.
  • input_lengths – Input lengths within batch.

• Returns: Output with dimensions (B, T, D), T is the processed length, : D is the feature dimension.

  Tensor: Output lengths within batch.

• Return type: Tensor

output_size() → int

reload_pretrained_parameters()