Views
No views yet
| Model | # Params | Tigrinya WER (↓) |
|---|---|---|
| Ethio-ASR (afrihubert) | 94M | 42.42 |
| Ethio-ASR (mms-300) | 300M | 41.62 |
| Ethio-ASR (mms-1b) | 1B | 37.63 |
| Ethio-ASR (w2v-bert-2.0) | 600M | 35.22 |
| Monolingual SFT (w2v-bert-2.0) 📌 | 600M | 35.65 |
1from transformers import AutoModelForCTC, AutoProcessor
2import torchaudio, torch
3
4processor = AutoProcessor.from_pretrained("badrex/Ethio-ASR-tigrinya")
5model = AutoModelForCTC.from_pretrained("badrex/Ethio-ASR-tigrinya")
6
7audio, sr = torchaudio.load("audio.wav")
8inputs = processor(audio.squeeze(), sampling_rate=sr, return_tensors="pt")
9
10with torch.no_grad():
11 logits = model(**inputs).logits
12
13pred_ids = torch.argmax(logits, dim=-1)
14transcription = processor.batch_decode(pred_ids)[0]
15
16print(transcription)1@misc{ethio_asr_2026,
2 author = {
3 Abdullah, Badr M. and
4 Azime, Israel Abebe and
5 Tonja, Atnafu Lambebo and
6 Alabi, Jesujoba O. and
7 Alemu, Abel Mulat and
8 Hagos, Eyob G. and
9 Balcha, Bontu Fufa and
10 Nerea, Mulubrhan A. and
11 Yadeta, Debela Desalegn and
12 Marilign, Dagnachew Mekonnen and
13 Fentahun, Amanuel Temesgen and
14 Kebede, Tadesse and
15 Gebru, Israel D. and
16 Woldeyohannis, Michael Melese and
17 Sewunetie, Walelign Tewabe and
18 Möbius, Bernd and
19 Klakow, Dietrich
20 },
21 title = {Ethio-ASR: Joint Multilingual Speech Recognition and Language Identification for Ethiopian Languages},
22 year = {2026},
23 howpublished = {\url{https://huggingface.co/badrex/Ethio-ASR-multilingual-600M}}
24}