fix speaker dim inference

2026-09-09 11:16:00 +08:00 · 2021-06-01 15:15:26 +02:00
parent 5b89ef2c6e
commit c1eb9bdcca
1 changed files with 1 additions and 0 deletions
@@ -255,6 +255,7 @@ class Tacotron2(TacotronAbstract):
        if self.num_speakers > 1:
            if not self.embeddings_per_sample:
                speaker_embeddings = self.speaker_embedding(speaker_ids)[:, None]
+                speaker_embeddings = torch.unsqueeze(speaker_embeddings, 0).transpose(1, 2)
            encoder_outputs = self._concat_speaker_embedding(encoder_outputs, speaker_embeddings)

        decoder_outputs, alignments, stop_tokens = self.decoder.inference(encoder_outputs)