Whisper Base el-en
A 72.6-million-parameter Whisper Base fine-tune that keeps Greek orthography as written: tonos, dialytika, final sigma, capitalization, and punctuation.
Specifications
- Status
- Released
- Parameters
- 72.6 million
- Base model
- OpenAI Whisper Base
- Languages
- Greek, English
- Training data
- 179.2 h Greek, 90.0 h English
- Training cost
- 65 minutes on one RTX 4090
- Formats
- Transformers (fp32), whisper.cpp (q8_0)
- License
- Apache-2.0
Greek word error rate
| Model | Parameters | WER |
|---|---|---|
| Whisper Base el-en | 72.6 M | 28.65% |
| Whisper Base | 72.6 M | 62.50% |
| Whisper Small | 241.7 M | 35.11% |
| Whisper Medium | 763.9 M | 20.77% |
| Whisper large-v3 | 1,543.5 M | 11.45% |
Lexical WER on the held-out FLEURS Greek test split. Lower is better. Accents count.
Limits
- Larger stock Whisper models are more accurate on Greek.
- With the language forced to English, it scores 4.55 points worse than stock Whisper Base on FLEURS English.
- It deletes the final word of 7.4% of Greek FLEURS validation clips.
- Without voice activity detection, silence can produce short invented phrases.
- We did not evaluate timestamps, telephone audio, children’s speech, or mixed-language speech.
- Every result comes from one training run with one seed.
Run it
from transformers import pipeline
asr = pipeline(
"automatic-speech-recognition",
model="AliacAI/whisper-base-el-en",
)
result = asr(
"speech.wav",
generate_kwargs={"language": "el", "task": "transcribe"},
)
print(result["text"])Need a model for your own task?
Bring the data and the constraints. We train, evaluate, and deploy it with you.
Discuss your project