Is there a way to bump up the audio quality created via the webui?

[Taluen79]

I used the webui to generate some audio. But once I downloaded the files, I noticed that the bit rate was only at 83 kbps? Is there a way to bump this up? Perhaps via the code? If someone can point me to the right direction. Seems like it was set to 83 kbps for fast generation? But sometimes we really need HD voices too. Thanks in advance!

[Taluen79]

So the reason for my first post is that I found this on HF:

https://huggingface.co/spaces/ysharma/Make_Custom_Voices_With_KokoroTTS

And the sound is SO crisp and clear, compared to what Kokoro-FastAPI delivers. But I have no idea why. Some sort of upsampling? That code on the HF link however, does not allow for very long text as Kokoro-FastAPI.

[RBEmerson970]

While I'm satisfied with the clarity of the speech through both Fast Koko and Open WebUI, I do find the volume varies from voice to voice. For example, am_adam is louder the af_kore.. A consistent volume across voices would be welcome.

[fireblade2534]

I think thats just the model itself

[remsky]

The WebUI is set to streaming by default I believe, but the download link at the end should be full. I'll run some checks and see what's up

[Taluen79]

Please do take a a listen to that HF link. The sound there is so crisp. The code for that page is also available. Perhaps you can see if he's doing some additional tricks.

[remsky]

Should be bumped up in the release (just pushing it in a few minutes). Sample output stats are below. I'm not sure if they have any tricks going, though it looks like that is the v0_19 model as well. Will try and take a closer look shortly

Sample Rate: 24000 Hz
Channels: 1
Duration: 2.86 seconds
Raw Bitrate: 1536.0 kbps

Audio Metrics:
RMS Level: -23.7 dB
Peak Level: -5.6 dB
Dynamic Range: 18.1 dB