espnet.nets.pytorch_backend.fastspeech.duration_calculator.DurationCalculator

source

class espnet.nets.pytorch_backend.fastspeech.duration_calculator.DurationCalculator(teacher_model)

Bases: Module

Duration calculator module for FastSpeech.

Initialize duration calculator module.

• Parameters:teacher_model (e2e_tts_transformer.Transformer) – Pretrained auto-regressive Transformer.

forward(xs, ilens, ys, olens, spembs=None)

Calculate forward propagation.

• Parameters:
  • xs (Tensor) – Batch of the padded sequences of character ids (B, Tmax).
  • ilens (Tensor) – Batch of lengths of each input sequence (B,).
  • ys (Tensor) – Batch of the padded sequence of target features (B, Lmax, odim).
  • olens (Tensor) – Batch of lengths of each output sequence (B,).
  • spembs (Tensor , optional) – Batch of speaker embedding vectors (B, spk_embed_dim).
• Returns: Batch of durations (B, Tmax).
• Return type: Tensor