Voxtral TTS

Voxtral 4B TTS 2603 runs locally on Apple Silicon through the VoxtralTTS module: 20 preset voices, nine languages, and 24 kHz mono speech.

Command line

speech speak "Hello, welcome to Voxtral." --engine voxtral --voxtral-voice casual_male --stream --voxtral-chunk-frames 3 -o hello.wav
speech speak --engine voxtral --list-speakers
speech speak "Bonjour." --engine voxtral --voxtral-voice fr_female --voxtral-max-frames 500 --voxtral-seed 0 -o bonjour.wav

Swift API

import VoxtralTTS

let model = try await VoxtralTTSModel.fromPretrained()
for try await chunk in model.generateStream(text: "Hello, welcome to Voxtral.", voice: "casual_male") {
    print(chunk.frameIndex, chunk.samples.count, chunk.isFinal)
}

Server API

curl http://localhost:8080/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -d '{"model":"voxtral-tts","input":"Hello, welcome to Voxtral.","voice":"casual_male","response_format":"wav"}' \
  -o hello.wav

/v1/realtime: session.update → {"model":"voxtral-tts","voice":"casual_male"}; response.create → {"instructions":"Hello, welcome to Voxtral."}.

curl http://localhost:8080/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -d '{"model":"voxtral-tts","input":"Hello, welcome to Voxtral.","voice":"casual_male","stream":true,"response_format":"pcm"}' \
  -o hello.pcm

Preset voices

`en`	`casual_male, casual_female, cheerful_female, neutral_male, neutral_female`
`fr`	`fr_male, fr_female`
`es`	`es_male, es_female`
`de`	`de_male, de_female`
`it`	`it_male, it_female`
`pt`	`pt_male, pt_female`
`nl`	`nl_male, nl_female`
`ar`	`ar_male`
`hi`	`hi_male, hi_female`

Limits and ownership

Preset voices only; reference-audio cloning is unavailable. Audio output streams through the SDK, CLI, WebSocket and HTTP PCM. Supply the complete input text up front. Model weights use CC BY-NC 4.0.

Each request owns its language cache, codec history and random seed. Callback delivery waits for the consumer. The default chunk is 3 frames (240 ms of audio), and the limit is 500 frames (40 seconds). A streaming failure can follow earlier audio chunks; only normal completion emits a final marker.

VoxtralTTSOptions(streamingChunkFrames: 3) · generateStreaming(onChunk:) · AudioChunk.isFinal.

MLX 4-bit · Mistral · VoxtralTTS