vits

[wangchuang5419]

I want to use VITS to conduct transfer training for a TTS model of the Shaanxi dialect in China. How should I do it?

[wangchuang5419]

File "C:\coqui\TTS\TTS\tts\models\vits.py", line 1481, in format_batch
d_vectors = [d_vector_mapping[w]["embedding"] for w in batch["audio_unique_names"]]
File "C:\coqui\TTS\TTS\tts\models\vits.py", line 1481, in
d_vectors = [d_vector_mapping[w]["embedding"] for w in batch["audio_unique_names"]]
KeyError: 'shaanxi_dataset#wavs\newsh2'
how can i solve this problen ?

[eginhard]

It would be helpful to have a few more details about what model you're fine-tuning and your training script and config. Based on the error message you likely need to first compute the speaker embeddings with https://github.com/idiap/coqui-ai-TTS/blob/dev/TTS/bin/compute_embeddings.py