API to list voices ?

[pinage404]

Hello,

Is there an API to list available voices associated with language ?

For example pocket-tts should output this list (a more structured output in JSON could help) :

alba (en)
giovanni (it)
lola (es)
juergen (de)
rafael (pt)
estelle (fr)
anna (en)
azelma (en)
bill_boerst (en)
caro_davy (en)
charles (en)
cosette (en)
eponine (en)
eve (en)
fantine (en)
george (en)
jane (en)
jean (en)
javert (en)
marius (en)
mary (en)
michael (en)
paul (en)
peter_yearsley (en)
stuart_bell (en)
vera (en)

and kokoro-multi-lang-v1.0-sherpa should output this list

Maybe a map of language to list of voices could be great

[0xshubhs]

No, there isn't one today.

The TTS surface is write-only: /v1/audio/speech, /tts, and the ElevenLabs-compatible POST /v1/text-to-speech/:voice-id (core/http/routes/elevenlabs.go:21). Note that last one takes a voice id but has no GET /v1/voices counterpart, unlike the real ElevenLabs API it mirrors — so there's nothing to enumerate against.

It also can't be added at the HTTP layer alone. backend/backend.proto has TTS, TTSStream and SoundGeneration, but no ListVoices-style RPC, so LocalAI has no channel to ask a loaded backend what it supports. Implementing this means a new proto method plus an implementation in every TTS backend (piper, kokoro, sherpa-onnx, vibevoice, fish-speech, qwen-tts…), each of which exposes voices differently — which is probably why it doesn't exist yet.

Your language→voices map is the right shape for it. Worth filing as a feature request rather than a question, since the proto change is the part that needs a maintainer decision.

In the meantime the voice lists are static per model, so pulling them from the upstream model card (the two you linked) and hardcoding is the only option.

[pinage404]

I created a feature request #11604