Description
Voxtral Mini 4B Realtime Arabic is a sophisticated streaming speech-to-text model tailored for Arabic dialects and Modern Standard Arabic. Developed under the strategic partnership between Mistral and Morocco's Ministère de la Transition Numérique et de la Réforme Administrative, this model is designed to handle challenging real-world audio scenarios, including code-switching where speakers alternate between languages within the same conversation.
Read More
The model is fine-tuned from Voxtral-Mini-4B-Realtime-2602 and boasts approximately 4.4 billion parameters, with weights in BF16 precision. It processes 16 kHz audio through a causal audio encoder and language decoder, producing text as audio arrives. This capability makes it suitable for applications such as live captions and voice interfaces, with a configurable transcription delay to optimize performance.
At a configured transcription delay of 480 ms, Voxtral Mini 4B Realtime Arabic achieves an average Character Error Rate (CER) of 8.82% across seven Arabic benchmarks, outperforming several open-source and proprietary offline models. It comes within 0.91 percentage points of the benchmark set by Voxtral Transcribe Arabic's 7.91% CER.
The model is available for integration with vLLM and can run natively in Transformers starting from version 5.2.0. It is licensed under the Apache 2.0 License, ensuring open-source accessibility while protecting intellectual property rights. Users are advised to adhere to the licensing terms to avoid infringement.
Voxtral Mini 4B Realtime Arabic is particularly beneficial for developers and organizations focusing on Arabic language applications, offering a robust solution for real-time transcription needs. However, it is primarily targeted at Arabic transcription, and users requiring support for other languages should consider the original Voxtral Mini 4B Realtime model.
- CategoryAI Transcription Tools
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
Voxtral Mini 4B Realtime Arabic Highlights
Streaming speech-to-text model
Handles Arabic dialects and Modern Standard Arabic
Supports real-world audio and code-switching
Fine-tuned from Voxtral-Mini-4B-Realtime-2602
Approximately 4.4 billion parameters
Configurable transcription delay
Average CER of 8.82% across benchmarks
Licensed under Apache 2.0
Getting Started with Voxtral Mini 4B Realtime Arabic
Access model: Visit Hugging Face page
Authenticate: Ensure proper credentials
Set up environment: Install necessary dependencies
Integrate via API: Use vLLM or Transformers
Optimize: Configure transcription delay
Voxtral Mini 4B Realtime Arabic's Use Cases
- Live Captions
- Voice Interfaces
- Code-Switching Transcription
- Broadcast Transcription
- Educational Tools
