Voxtral

Voxtral is a new family of open-source audio models from Mistral AI. It is designed to provide advanced speech recognition and understanding capabilities. Voxtral is aimed at developers who want more control over how they use these models. It offers two versions: Voxtral Small for large-scale applications and Voxtral Mini for local and edge use. Both models are available for free on Hugging Face under the Apache 2.0 license. The API version, called Voxtral Mini Transcribe, is optimized for cost and speed.

Benefits

Voxtral supports up to 30 minutes of audio for transcription and 40 minutes for understanding. It can answer questions about audio content and create summaries without needing extra tools. Voxtral works in multiple languages, including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian. It can also trigger actions based on voice commands. Voxtral performs better than other open-source models and is competitive with top proprietary models. It is available for free download and can be deployed on-premise or in the cloud.

Use Cases

Voxtral can be used in various settings, such as customer service, healthcare, and education. It can transcribe and summarize meetings, interviews, and lectures. It can also be used in call centers to transcribe and analyze customer calls. Additionally, Voxtral can be used in voice-controlled applications and devices.

Pricing

Voxtral is available for free download on Hugging Face. API integration starts at $0.001 per minute.

Vibes

Voxtral has received positive reviews for its accuracy and multilingual capabilities. Users appreciate its flexibility and the ability to deploy it on-premise or in the cloud.

Additional Information

Mistral offers advanced features for enterprises, including private deployment, domain-specific fine-tuning, and dedicated support. The company is also inviting design partners to build support for speaker identification, emotion detection, advanced diarization, and longer context windows.