Wav2Vec2FABundle.Aligner¶

class torchaudio.pipelines.Wav2Vec2FABundle.Aligner¶: Interface of the aligner

Methods¶

abstract Aligner.__call__(emission: Tensor, tokens: List[List[int]]) → List[List[TokenSpan]]¶

Generate list of time-stamped token sequences

Parameters:

emission (Tensor) – Sequence of token probability distributions in log-domain. Shape: (time, tokens).
tokens (list of python:integer sequence) – Tokenized transcript. Output from torchaudio.pipelines.Wav2Vec2FABundle.Tokenizer.

Returns:

Tokens with time stamps and scores.

Return type:

(list of TokenSpan sequence)