Rumble-2-Flash: Open Source Speech Model for 40+ Languages Released

Hugging Face has released Rumble-2-Flash, an open-source speech recognition model with 0.5 billion parameters, capable of recognizing speech in over 40 languages. This model is designed to be highly efficient, achieving performance comparable to much larger models while maintaining rapid processing speeds, transcribing an hour of audio in approximately half a second.

Lyra VoxleyModels & Research

1 min read1 source

Key takeaways

30-second read
  1. 1Developers can integrate Rumble-2-Flash into their applications without needing custom code, thanks to its compatibility with existing frameworks like 🤗 Transformers and vLLM.
  2. 2The model's automatic language detection and output control features allow for versatile applications in multilingual environments, enhancing user experience.
  3. 3With its competitive accuracy and speed, product teams can leverage Rumble-2-Flash for real-time transcription services, improving efficiency in customer support and communication tools.
  • Confirmed

    Official source: HuggingFace Blog

  • Reported

    No secondary reporting linked

  • Verify yourself

    Eligibility, local prices and dates for your car, state or utility

Rumble-2-Flash: Open Source Speech Model for 40+ Languages Released

Hugging Face has announced the open-source release of Rumble-2-Flash, a speech recognition model that boasts 0.5 billion parameters and supports over 40 languages. This model is notable for its ability to perform on par with larger models, achieving a 6.96% error rate across multiple languages, including English, Spanish, and Chinese, which is better than some models that are up to 15 times its size.

Key Features

Rumble-2-Flash is designed for speed and efficiency, capable of transcribing an hour of audio in about half a second, reaching over 7,000 real-time factors (RTFx) on full batches. It also excels in real-world applications, demonstrating strong performance on noisy phone calls, with a word error rate (WER) of 7.86% on a benchmark that includes various English accents, comparable to NVIDIA's larger Nemotron 3.5 model.

The model is built on the Qwen3-ASR architecture and uses an innovative on-policy distillation technique, allowing it to learn from its own outputs for improved accuracy. Developers can easily implement Rumble-2-Flash in their projects, as it requires no custom code and integrates seamlessly with existing AI frameworks.

This release opens up new possibilities for developers and product teams looking to enhance their applications with robust speech recognition capabilities, particularly in multilingual contexts. The model's performance and ease of use make it an attractive option for building real-time transcription services and improving customer interactions.

0 people like this

Share this story

speech recognitionopen sourceHugging FacemultilingualAI models

Buying guide: Best Microphone for Voice AI and Speech-to-Text (2026): How to Choose

Sources

1 official
  1. [1]
    Rumble-2-Flash: an open 0.5B speech model for 40+ languages

    HuggingFace Blog · huggingface.co · published Nov 17

    Official
This brief was drafted with AI and published automatically after checks on its sources and content; an editor has not reviewed it. Articles are expected to link back to source material. Read our editorial standards or contact us with corrections at [email protected].

Newsletter

Weekly AI signal. No spam.

Comments

Sign in to join the discussion

Loading comments…