Is voice selection unsupported in JP TTS?
Hi! I'm running LFM2.5-Audio-1.5B-JP locally on Windows with Python (liquid-audio).
TTS works fine with "Perform TTS in japanese.", but the output is always a female voice. I tried the following, with no luck:
Perform TTS in japanese. Use the US male voice.→ female voicePerform TTS in japanese. Use the Japanese male voice.→ female voicePerform TTS. Use the US male voice.→ no audio tokens generatedPerform TTS in japanese. Use the US female voice.→ female voice (voice keyword seems ignored entirely)
The model card says TTS uses the fixed system prompt Perform TTS in japanese., and the voice keywords (US/UK × male/female) seem to apply only to the base model. So my understanding is:
- The JP model ships with a single (female) voice, and voice selection is not supported — is that correct?
- If so, is there any plan to add a male Japanese voice? Or is fine-tuning with liquid-audio's trainer the recommended path?
Thanks!
Hi, thanks for the question.
The JP model ships with a single (female) voice, and voice selection is not supported — is that correct?
Yes, voice selection is currently not supported. The only system prompt currently supported is Perform TTS in japanese..
If so, is there any plan to add a male Japanese voice? Or is fine-tuning with liquid-audio's trainer the recommended path?
Yes, we plan on working on this, however, if you need this feature immediately, we'd recommend using liquid-audio to finetune it on your own voice/character of choice.
Thank you for the clarification! One follow-up question: does the fine-tuning pipeline in examples/preprocess_jenny_tts.py also work with the JP model (LFM2.5-Audio-1.5B-JP)? Or is fine-tuning only supported for the base model?