language: af
license: mit
library_name: transformers
pipeline_tag: automatic-speech-recognition
base_model: openai/whisper-large-v3
tags:
- automatic-speech-recognition
- whisper
- afrikaans
- buzzasr
datasets:
- google/fleurs
metrics:
- cer
- wer
BuzzASR — Afrikaans
A monolingual automatic speech recognition model for Afrikaans, fine-tuned from openai/whisper-large-v3. Part of BuzzASR, a suite of 102 language-specialized ASR models (Findings of EMNLP 2026).
This model uses simple fine-tuning (Whisper's tokenizer, ASR fine-tuning only).
Results (normalized CER / WER, %)
| Test set | CER | WER | Whisper-large-v3 (zero-shot) CER |
|---|---|---|---|
| FLEURS | 7.67 | 22.36 | 11.74 |
| Common Voice 25 | 1.03 | 4.21 | 9.6 |
| Combined | 5.35 | 16.01 | 10.88 |
~2.0x CER reduction over Whisper zero-shot on the combined test set.
Usage
import torch, torchaudio
from transformers import WhisperForConditionalGeneration, WhisperProcessor
model = WhisperForConditionalGeneration.from_pretrained("BuzzASR/afrikaans", torch_dtype=torch.float16).to("cuda").eval()
proc = WhisperProcessor.from_pretrained("BuzzASR/afrikaans")
wav, sr = torchaudio.load("audio.wav") # 16 kHz mono
feats = proc(wav[0], sampling_rate=16000, return_tensors="pt").input_features.to("cuda").half()
ids = model.generate(feats, num_beams=1, no_repeat_ngram_size=3, repetition_penalty=1.2)
print(proc.batch_decode(ids, skip_special_tokens=True)[0])
The language/task prompt is baked into the generation config, so no language= argument is needed.
Training data
FLEURS + Common Voice Corpus 25.0 (Mozilla, March 2025; https://commonvoice.mozilla.org/en/datasets), capped per the paper. Text-only data from the Goldfish corpus (Chang et al., 2026).
Limitations
Monolingual (Afrikaans only). Evaluated on FLEURS / Common Voice test splits; other domains or dialects may differ.
Citation
Project page: https://lemn-lab.github.io/buzzasr-docs/