Introduction to Mistral's Audio Capabilities

Update November 2, 2026 — Le Chat is now Vibe. Since May 28, 2026, Le Chat has been rebranded as Vibe (still at chat.mistral.ai): a single agent for work and code, carrying over your conversations, settings, and subscriptions. The legacy Le Chat experience remains accessible in Chat mode; in Work mode, agents have become skills (invoked with @), memories are now the knowledge base, and deep research is the Deep Research skill. The plans are now called Mistral Free, Pro, Team, and Enterprise. This lesson uses the name Vibe; note that some menus described here may have changed names or locations.

When AI Learns to Listen and Speak

After vision, audio represents Mistral AI’s second major multimodal revolution. With the Voxtral model family, Vibe no longer just reads and writes text—it can now transcribe speech, participate in voice conversations, and generate speech from text. In short, the loop is complete: the AI listens to you, understands you, and answers you out loud. This lesson introduces all these capabilities before subsequent lessons detail them one by one.

The Voxtral Family

Mistral has developed a comprehensive suite of audio models under the Voxtral brand, with each fulfilling a specific role in the audio pipeline. Voxtral Mini Transcribe is the batch transcription model (Speech-to-Text): you give it an audio file, and it returns written text with remarkable accuracy—this is the model at work when you upload a meeting recording. Voxtral Realtime is its real-time counterpart: designed for voice conversations, it transcribes speech as it happens with ultra-low latency, configurable under 200 milliseconds—this is what makes Vibe’s fluid voice mode possible. Finally, Voxtral TTS goes the other way: text-to-speech synthesis transforms text into natural voice, with the ability to clone a voice from just a few seconds of recording.

Speech-to-Text: From Voice to Text

Audio transcription is the ability to convert a voice recording into written text, and Voxtral Mini Transcribe stands out on several fronts. First, its accuracy: the word error rate remains low, even on medium-quality recordings—the kind of file captured by a phone sitting in the middle of a conference table. It also performs diarization, meaning it automatically identifies different speakers in a conversation: the transcription distinguishes “Speaker 1” and “Speaker 2,” which changes everything for meeting minutes. Its noise robustness allows it to maintain good accuracy in difficult environments—cafes, streets, multi-voice meetings. Finally, it handles long recordings, up to 3 hours of audio in a single request, enough to cover an entire conference.

On the language front, 13 are supported: English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Hindi, Arabic, Russian, Japanese, and Korean. For common European use cases, you are well covered.

Text-to-Speech: From Text to Voice

Text-to-speech works in the opposite direction: you provide text, and Voxtral TTS reads it aloud with a natural voice. Its most spectacular feature is voice cloning: from just 2 to 3 seconds of your voice recording, the model can generate speech that sounds like you. This cloning relies on the voice-as-an-instruction principle: the model reproduces the intonation, rhythm, and emotion of the reference voice sample without needing to annotate the text with prosody tags—the sample itself serves as the instruction.

The model is multilingual—English, French, Spanish, Portuguese, Italian, Dutch, German, Hindi, Arabic—and can even mix languages within a single passage. Finally, it operates in streaming mode: voice generation begins in about 90 milliseconds, enabling near-instant interactions rather than a wait between your question and the spoken response.

What Is Audio Used for in Vibe?

Concretely, as a Vibe user, these capabilities translate into four everyday actions. You can transcribe a meeting: upload the recording and get a text summary in seconds. You can talk by voice by activating voice mode, for a natural conversation without a keyboard. You can listen to responses by asking Vibe to read its text aloud—handy while walking or cooking. And you can dictate your messages, speaking instead of typing.

In the following lessons, we will detail each of these uses step by step: file transcription first, then real-time voice mode, and finally text-to-speech synthesis.

Key Takeaways

• Voxtral is Mistral’s family of audio models, covering transcription, real-time processing, and text-to-speech
• Transcription supports 13 languages, diarization, and up to 3 hours of audio
• Text-to-speech can clone a voice in seconds and speak in 9+ languages
• In Vibe, everything is integrated: you can transcribe, speak, and listen without external tools