Hi, and thank you for your amazing work on LLaSA TTS!
I have two questions regarding multi-speaker usage and training:
-
I’m working with a multi-speaker version of the model, and I’d like to know how I can explicitly select a specific speaker at inference time. Is there a speaker_id parameter, speaker embedding, or reference audio I should provide to control the speaker output?
-
I’m also preparing to fine-tune LLaSA for multiple speakers, and I’d like to know if there’s anything special I need to consider during data preparation or training. For example, should the dataset include a speaker_id field or follow a specific format to enable speaker conditioning?
Hi, and thank you for your amazing work on LLaSA TTS!
I have two questions regarding multi-speaker usage and training:
I’m working with a multi-speaker version of the model, and I’d like to know how I can explicitly select a specific speaker at inference time. Is there a speaker_id parameter, speaker embedding, or reference audio I should provide to control the speaker output?
I’m also preparing to fine-tune LLaSA for multiple speakers, and I’d like to know if there’s anything special I need to consider during data preparation or training. For example, should the dataset include a speaker_id field or follow a specific format to enable speaker conditioning?