BuzzASR/kamba

1---2language: kam3license: mit4library_name: transformers5pipeline_tag: automatic-speech-recognition6base_model: openai/whisper-large-v37tags: [automatic-speech-recognition, whisper, kamba, buzzasr]8datasets: [google/fleurs]9metrics: [cer, wer]10---11 12# BuzzASR — Kamba13 14A monolingual automatic speech recognition model for **Kamba**, fine-tuned from15[openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3). Part of **BuzzASR**,16a suite of 102 language-specialized ASR models17([paper: arXiv:2609.09554](https://arxiv.org/abs/2609.09554), Findings of EMNLP 2026).18 19This model uses **simple fine-tuning (Whisper's tokenizer, ASR fine-tuning only)**.20 21## Results (normalized CER / WER, %)22 23| Test set | CER | WER | Whisper-large-v3 (zero-shot) CER |24|---|---|---|---|25| FLEURS | 15.92 | 52.08 | 41.55 |26| Combined | 15.92 | 52.08 | 41.55 |27 28~2.6x CER reduction over Whisper zero-shot on the combined test set.29 30## Usage31 32```python33import torch, torchaudio34from transformers import WhisperForConditionalGeneration, WhisperProcessor35 36model = WhisperForConditionalGeneration.from_pretrained("BuzzASR/kamba", torch_dtype=torch.float16).to("cuda").eval()37proc  = WhisperProcessor.from_pretrained("BuzzASR/kamba")38 39wav, sr = torchaudio.load("audio.wav")           # 16 kHz mono40feats = proc(wav[0], sampling_rate=16000, return_tensors="pt").input_features.to("cuda").half()41ids = model.generate(feats, num_beams=1, no_repeat_ngram_size=3, repetition_penalty=1.2)42print(proc.batch_decode(ids, skip_special_tokens=True)[0])43```44The language/task prompt is baked into the generation config, so no `language=` argument is needed.45 46## Training data47[FLEURS](https://huggingface.co/datasets/google/fleurs), capped per the paper. Text-only data from the **Goldfish** corpus (Chang et al., 2026).48 49## Limitations50Monolingual (Kamba only). Evaluated on FLEURS / Common Voice test splits; other domains or dialects may differ.51 52## Links & citation53- **Paper:** https://arxiv.org/abs/2609.09554 (Findings of EMNLP 2026)54- **Project page:** https://lemn-lab.github.io/buzz-asr/55- **All models:** https://huggingface.co/BuzzASR56 57```bibtex58@misc{buzzasr2026,59  title         = {BuzzASR: A Swarm of 100+ Monolingual Speech Recognition Models},60  author        = {Shivam Singh and Aditya Yadavalli and Catherine Arnett and Alex Warstadt},61  year          = {2026},62  eprint        = {2609.09554},63  archivePrefix = {arXiv},64  primaryClass  = {cs.CL},65  note          = {Findings of the Association for Computational Linguistics: EMNLP 2026},66  url           = {https://arxiv.org/abs/2609.09554}67}68```69