Spaces:

---

styletts2

/

styletts2

like 693

Running on L4

Hi, try cloning a voice with more emotion. For example, try using the Voice Cloning tab and record yourself speaking in a very sad voice, the resulting audio should sound sad (but not very much like your voice, because the LibriTTS dataset is pretty small). Let me know if this adds more emotion to your synthesized speech.

Hey @mrfakename ! Thanks so much for your response! I will definitely try that!

Hi, one last thing – personally, I feel that the LJSpeech model tends to have a little bit more emotion in the voices than the other voices. You can access the LJSpeech model using the LJSpeech tab.

I will try this too! Thanks

Okay I tried it but how can I choose male and female voices in LJ? I only got a female.

Unfortunately LJSpeech only supports a single voice. If you need multiple speakers, I'd recommend trying cloning emotive voices. However, since StyleTTS 2 is mostly trained on LibriVox, I'm not sure how well that will work. If you're fine with slower generations and non-commercial outputs, then I'd personally recommend XTTS for more emotion.

Thanks...and where is the XTTS? Sorry, kinda new to this.

Hi,
The model (run locally/offline) is here.
The demo (run online) is here.
Please keep in mind that they have a non-commercial license so it cannot be used commercially.
Hope this helps!

Thank you! I will keep that in mind.

Hi, were you able to find a solution to this?

Hey @mrfakename , I just re-rolled until I got really close to what I needed. lol. Thanks so much for this program!

Nice, glad you were able to resolve this!