pocket-tts fails to find custom voice

[Expro]

As per documentation:

name: pocket-tts
backend: pocket-tts
tts:
voice: "azelma" # Or use local file path: "path/to/voice.wav"

I have provided custom voice:

backend: pocket-tts
description: |
Pocket TTS is a lightweight text-to-speech model designed to run efficiently on CPUs.
This model supports voice cloning through HuggingFace voice URLs or local audio files.
known_usecases:
- tts
name: pocket-tts
options:
- default_voice:"/models/maciej.wav"
parameters:
model: ""
tts:
voice: "/models/maciej.wav"

But it fails with: error during TTS: Voice not found or failed to load: /models/maciej.wav. Please provide a valid voice URL or file path

[rampa3]

@Expro Voice cloners don't accept voice sample through voice parameter. That parameter is for selecting voice on classic TTS. You have to use parameter audio_path, like here:

...

overrides:
    tts:
      audio_path: /var/lib/local-ai/voice_samples/me.wav

...

[Expro]

Still doesn't work: error during TTS: Voice not found or failed to load: None. Please provide a valid voice URL or file path.

[rampa3]

OK, seems to behave differently than other voice cloning TTSes. Usually voice cloning TTS does not use voice parameter when cloning if I remember. I have to check the source and docs...

[rampa3]

Are you on Docker or binary? Where exactly is the sample on your machine? That path is a bit weird...

[Expro]

Container - I have volumes mounted for /backends, /models and /configuration, if I want to have persistent voice sample it has to be put into one of those (sub)paths.

[rampa3]

OK. For that, that path would be valid. What happens if you use file:// URL? That is probably the last idea I have right now. Try voice parameter and file:// URL.

[Expro]

Still not loading it (same error, but with "file://..." content).