iCreatAI News
Skip to content
Hugging Face Blog·· 3 days agoFeaturedAI score62

Rumble-2-Flash: an open 0.5B speech model for 40+ languages

Rumble-2-Flash: an open 0.5B speech model for 40+ languages

AI summary

Reson8 has open-sourced Rumble-2-Flash under Apache 2.0, a 0.5B-parameter speech recognition model for 40+ languages with automatic language detection, output controls, and native support in Transformers and vLLM. It reports a 6.96% average error rate on the Open ASR Leaderboard multilingual track and 7.86% WER on AppTek's 302-hour call-centre benchmark, on par with NVIDIA's Nemotron 3.5 Transcribe (30B) at 7.78%. The model was distilled from the larger Rumble-2 using on-policy distillation with GOLD, which merges tokens across differing tokenizers.

Why it matters

It offers a concrete look at how far on-policy distillation can shrink a speech model while staying competitive with much larger systems on noisy real-world audio.

Source: Hugging Face Blog · huggingface.co