Speech Generation

Generate speech from text using a saved voice (voice_id) or a one-off reference audio clip (ref_audio).

import base64
from pathlib import Path
from mistralai.client import Mistral

client = Mistral(api_key="your-api-key")

response = client.audio.speech.complete(
    model="voxtral-mini-tts-2603",
    input="Hello! This is Voxtral, Mistral's text-to-speech model.",
    voice_id="your-voice-id",
    response_format="mp3",
)

Path("output.mp3").write_bytes(base64.b64decode(response.audio_data))
print("Saved to output.mp3")

import base64
from pathlib import Path
from mistralai.client import Mistral

client = Mistral(api_key="your-api-key")

response = client.audio.speech.complete(
    model="voxtral-mini-tts-2603",
    input="Hello! This is Voxtral, Mistral's text-to-speech model.",
    voice_id="your-voice-id",
    response_format="mp3",
)

Path("output.mp3").write_bytes(base64.b64decode(response.audio_data))
print("Saved to output.mp3")

• Language match: the voice prompt should be in the same language as the text prompt for best results.
• Cross-lingual prompts: the model also supports cross-lingual voice transfer. For example, a French voice prompt with English text will produce French-accented English.
• Verbalizable form: convert numbers and symbols to their spoken equivalent to avoid ambiguity. For example, use one thousand two hundred thirty four instead of 1234, or twelve thirty four depending on context.
• No rich formatting: avoid markdown, emojis, or special characters in the text — they will not be rendered and may degrade output quality.
• Abbreviations: spell out abbreviations for better pronunciation. Use F-B-I or F.B.I. instead of FBI.
• Length: keep prompts under 300 words for best results.