search

kimjammer / Neuro

A recreation of Neuro-Sama originally created in 7 days.
MIT License
204 stars 26 forks source link

This isnt really much of an issue but #5

Closed KaisuzanAI closed 5 months ago

KaisuzanAI commented 6 months ago

I was just curious if you had any suggestions how to make the tts voice output better as it for me outputs rather choppy.

Thanks

kimjammer commented 6 months ago

Sure, it sounds like your computer isn't generating the TTS fast enough to be real time. What CPU and GPU do you have? You need to run the TTS on the GPU for real time synthesis. Check RealtimeTTS's GPU installation instructions for guidance. If you've disabled deepspeed, that could also be a factor.

Finally, sometimes this happens when you just run out of GPU compute/VRAM. See if your GPU is at 100% usage or 100% VRAM while generating TTS. This has happened to me when I was trying to run the whole project and stream and record.

In any case, adding full_sentences=True to the TTS configs in tts.py will force the TTS to wait until a full sentence is generated before playing the audio. This will solve the stuttering at the cost of having to wait.

engine = CoquiEngine(
            use_deepspeed=True,
            voice="./voices/" + VOICE_REFERENCE,
            speed=1.1,
            full_sentences=True
        )
KaisuzanAI commented 6 months ago

Awesome thanks, my cpu is an i9 and my gpu is the 4090. Ill try your suggestion