Voxtral Transcribe 2 (mistral.ai)

Article summary

Mistral AI has released Voxtral Transcribe 2, a next-generation speech-to-text model with state-of-the-art transcription quality, diarization, and ultra-low latency. The model comes in two versions: Voxtral Mini Transcribe V2 for batch transcription and Voxtral Realtime for live applications. Voxtral Realtime is open-source and available under the Apache 2.0 license. The models support 13 languages and offer industry-leading accuracy at a fraction of the cost of competitors.

Main themes

• Speech-to-text technology
• Transcription models
• Diarization
• Low-latency transcription
• Language support
• Pricing and cost-effectiveness

What commenters say

• The new Voxtral Transcribe 2 model offers impressive transcription quality and diarization capabilities, but its real-time capabilities are limited.
• The model's low word error rate and cost-effectiveness make it a competitive option in the market.
• Some commenters question the comparison to other transcription models, such as Whisper, and argue that the evaluation metrics may be misleading.
• The model's support for multiple languages, including Italian, is notable, and some commenters discuss the language's phonetic characteristics and information density.
• There is disagreement about the significance of the model's performance on Italian, with some arguing that it is due to the language's unique properties and others citing the need for more evidence.
• Some commenters suggest that the model's pricing is competitive, but others point out that other options, such as Whisper, may be cheaper.
• The open-source release of Voxtral Realtime under the Apache 2.0 license is seen as a positive development, allowing for wider adoption and customization.