Mistral AI has launched its latest innovation, the Voxtral Transcribe 2, which aims to enhance automatic speech recognition (ASR) for various applications. This new family of models is designed to cater to both batch and real-time transcription needs, ensuring efficiency and accuracy.
The Voxtral Transcribe 2 includes two main models: the Voxtral Mini Transcribe V2, which focuses on batch transcription and speaker diarization, and the Voxtral Realtime, optimized for low-latency streaming transcription. Both models support 13 languages, including English, Spanish, and Chinese, making them versatile tools for global communication.
Mistral emphasizes that the Voxtral Mini Transcribe V2 is tailored for high-quality transcription across different domains. It features advanced speaker diarization, which allows for accurate identification of speakers during conversations. This is particularly useful for meetings and interviews.
On the other hand, the Voxtral Realtime model is built for speed. It can deliver transcription with a delay of less than 500 milliseconds, making it suitable for live applications. Users can adjust the latency based on their needs, ranging from 80 milliseconds for interactive tasks to 2.4 seconds for scenarios where accuracy is paramount.
Mistral has made these models accessible through its API, with pricing set at $0.003 per minute for batch processing and $0.006 per minute for real-time services. The Realtime model is also available as open weights, allowing developers to integrate it into their own applications easily.
The launch of Voxtral Transcribe 2 marks a significant step for Mistral in the competitive field of speech recognition technology, aiming to meet the growing demand for efficient and reliable transcription solutions in various industries.