Supertonic - GPU support not working

[jborza]

Played around with the webui branch after the merge -
tried conversion for a sample book (The Yellow Wallpaper By Charlotte Perkins Gilman).

On my machine (Ryzen 7 5800x, RTX 4060) it's doing conversion only on CPU - took 01:25

I checked settings, and did another conversion with "Use GPU" explicitly turned off, it took the same amount of time.

For comparison, conversion using Kokoro, bf_isabella, speed 1.2x took 00:46 (on GPU), so about half.

I looked at the code, this function https://github.com/denizsafak/abogen/blob/webui/abogen/tts_supertonic.py#L132 __configure_supertonic_gpu() is supposed to log a warning if it failed to do configuration, my log (standard output, not super if there is another log) doesn't show the warning

127.0.0.1 - - [12/Jan/2026 08:21:33] "GET /static/wizard.js HTTP/1.1" 304 -
2026-01-12 08:21:33 [INFO] [job 4617e51016ad4051bae0121a026b99ce] Chapter 1 title · 64/30657: 978ddfddbd _ pg-charlotte-perkins-gilman _ the-yellow-wallpaper.
2026-01-12 08:21:33 [DEBUG] [job 4617e51016ad4051bae0121a026b99ce] Emitting 269 paragraph chunks for chapter 1.
2026-01-12 08:21:33 [INFO] [job 4617e51016ad4051bae0121a026b99ce] Chunk 0000_p0001 · 92/30657: By Charlotte Perkins Gilman.

So - overall - supertonic is fast, really fast, but I want it to go faster with the GPU acceleration :)

FYI @jeremiahsb

[paramp07]

i thought abogen used kokoro. How do you switch to supertonic?

[jborza]

@paramp07 the branch webui had a recent update with supertonic models.