omniASR-CTC-300M-v2-Zulu-All-v2
Wav2Vec2 CTC ASR model (v2) converted from the OmniLingual fairseq2 checkpoint omniASR_CTC_300M_v2.
Fine-tuned from facebook/omniASR-CTC-300M for isiZulu speech recognition on all isiZulu ASR training data in our mix, then converted to Hugging Face transformers format.
Best validation WER: 17.75 (step 685,000).
The model outputs CTC logits over a SentencePiece vocabulary.
Model details
| Property | Value |
|---|---|
| HF class | Wav2Vec2ForCTC |
| Encoder layers | 24 |
| Hidden size | 1024 |
| Attention heads | 16 |
| FFN intermediate | 4096 |
| Vocabulary size | 10288 |
| Source framework | fairseq2 |
| Source card | omniASR_CTC_300M_v2 |
| Parity verification | ✅ Verified |
Numerical parity against the original fairseq2 checkpoint has been confirmed: outputs match to within atol=1e-4 on a held-out audio sample.
Usage
from transformers import Wav2Vec2ForCTC, AutoProcessor
import torch, torchaudio
processor = AutoProcessor.from_pretrained("uctnlp/omniASR-CTC-300M-v2-Zulu-All-v2")
model = Wav2Vec2ForCTC.from_pretrained("uctnlp/omniASR-CTC-300M-v2-Zulu-All-v2")
model.eval()
waveform, sr = torchaudio.load("audio.wav")
if sr != 16_000:
waveform = torchaudio.functional.resample(waveform, sr, 16_000)
inputs = processor(
waveform.squeeze().numpy(), sampling_rate=16_000, return_tensors="pt"
)
with torch.no_grad():
logits = model(**inputs).logits # (1, T, vocab)
pred_ids = torch.argmax(logits, dim=-1)
transcript = processor.decode(pred_ids[0])
print(transcript)