File size: 2,400 Bytes
0920c8b 9492b04 0920c8b 9492b04 0920c8b 9492b04 dfc5e05 9492b04 9242bc4 9492b04 c5e131d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 | ---
title: BuzzASR
emoji: 🐝
colorFrom: yellow
colorTo: red
sdk: static
pinned: false
license: mit
---
# 🐝 BuzzASR — A Swarm of 100+ Monolingual Speech Recognition Models
**One specialist model per language.** BuzzASR is a suite of **102 monolingual ASR models**, each a fine-tune of
[Whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) specialized to a single language.
The work has been accepted at **EMNLP 2026** · built by [Lemn Lab](https://lemn-lab.github.io/buzz-asr/) in collaboration with [EleutherAI](https://huggingface.co/EleutherAI).
## The idea
Big multilingual models spread themselves thin across hundreds of languages. We asked whether a single specialist,
the same size as Whisper, could beat the giant generalists. It can to a considerable extent :)
## Highlights
- ✅ Beats Whisper-large-v3 on **89 of 102** languages (FLEURS)
- 🏆 **Open-source state-of-the-art** (beats Whisper, Omnilingual 1B/7B, MMS, Qwen3-ASR, Cohere) on **27 languages**
- 📉 Cuts character error rate ~3x on average, and 3.45x across Whisper's 51 worst languages
- 🧩 Each model is the same 1.55B parameters as Whisper — about 4.5x smaller than the strongest baseline (Omni-7B)
## A few standouts (combined FLEURS + Common Voice, normalized CER %)
| Language | BuzzASR CER | Whisper zero-shot | Note |
|---|---|---|---|
| [Cantonese](https://huggingface.co/BuzzASR/cantonese) | **13.0** | 36.4 | SOTA — 2x better than the next-best system |
| [Punjabi](https://huggingface.co/BuzzASR/punjabi) | **8.8** | 42.3 | SOTA |
| [Mongolian](https://huggingface.co/BuzzASR/mongolian) | **5.2** | 38.2 | SOTA |
| [Korean](https://huggingface.co/BuzzASR/korean) | **4.6** | 5.7 | SOTA |
| [Amharic](https://huggingface.co/BuzzASR/amharic) | **9.1** | 191 | >20x reduction over Whisper |
## Find your language
Browse all 102 models in the **Models** tab above, or go to `huggingface.co/BuzzASR/<language>`.
## Usage
```python
from transformers import WhisperForConditionalGeneration, WhisperProcessor
model = WhisperForConditionalGeneration.from_pretrained("BuzzASR/mongolian")
proc = WhisperProcessor.from_pretrained("BuzzASR/mongolian")
# the language/task prompt is baked in — just call model.generate(input_features)
```
## Links
- 📄 Paper — Findings of EMNLP 2026 (https://arxiv.org/abs/2609.09554)
- 🌐 Project & docs — https://lemn-lab.github.io/buzz-asr/
|