Whisper Small Ro 80mel
Fine-tuned openai/whisper-small for Romanian automatic speech recognition. Trained as part of the Dorele project — a fully Romanian-language, low-latency smart-home voice assistant — where Whisper models serve as...
Modellquelle
- Art
- Text model
- Dateigröße
- 922 MB
- Basismodell
- openai/whisper-small
- Version
- vmodel
- Aktualisiert
- 24. Okt. 2026
- Quelle
- Hugging Face
Quellenbeschreibung
Whisper Small — Romanian (whisper-small_ro-80mel)
Fine-tuned openai/whisper-small for Romanian automatic speech recognition. Trained as part of the Dorele project — a fully Romanian-language, low-latency smart-home voice assistant — where Whisper models serve as on-premise STT engines.
Results
Evaluated on a held-out validation split of 25,348 samples, beam size 5:
| Metric | Value |
|---|---|
| WER | 9.21 % |
| WER (normalized: lowercase, punctuation stripped) | 8.8 % |
| CER | 3.03 % |
The validation set is a deliberately hard multi-source mix (read speech, parliamentary/spontaneous speech, command-domain utterances), so numbers are not directly comparable to single-corpus benchmarks like Common Voice test.
Training data
~185 K training samples drawn from a combined Romanian ASR corpus (~290 K samples / ~619 h total across splits), assembled from:
- Common Voice 25 (Romanian)
- Google FLEURS (ro_ro)
- VoxPopuli (Romanian)
- Echo read-speech corpus
- RODigits (spoken digits)
- USPDATRO
- CRoWL
- RO-Smart synthetic TTS data (smart-home command domain)
Audio pre-processed to 80-bin log-mel features (30 s windows, Whisper standard). Splits: train 185,175 / validation 25,348 / test 78,167.
Training procedure
Full fine-tune (no LoRA) on a single RTX 5070 Ti 16 GB, BF16 autocast, SDPA attention.
| Hyperparameter | Value |
|---|---|
| Parameters | 244 M |
| Effective batch size | 192 (12 × 16 grad accum) |
| Learning rate | 2e-5, cosine schedule |
| Warmup steps | 750 |
| Max epochs | 10 (early stopping, patience 5 on eval WER) |
| Optimizer | AdamW (fused) |
| Label smoothing | 0.0 |
| SpecAugment | 2 freq masks (≤27 bins), 2 time masks (≤100 frames) |
| Selection metric | raw WER, best checkpoint reloaded at end |
Labels use the canonical Whisper target sequence
(<|startoftranscript|><|ro|><|transcribe|><|notimestamps|> ... <|endoftext|>)
with teacher-forcing shift, and decoder_start_token_id set to
<|startoftranscript|> so training and generation are exactly aligned.
Usage
from transformers import pipeline
asr = pipeline(
"automatic-speech-recognition",
model="IonGrozea/whisper-small_ro-80mel",
device="cuda",
)
print(asr("audio.wav", generate_kwargs={"language": "romanian", "task": "transcribe"}))
For low-latency deployment, the model converts cleanly to CTranslate2 / faster-whisper and whisper.cpp (GGUF).
Intended use & limitations
- Romanian speech only; language token is forced to
<|ro|>. - Strongest on read speech and short command-style utterances matching the training distribution; expect degraded accuracy on noisy far-field audio, heavy dialects, and code-switching.
Citation
If you use this model, please also credit the underlying datasets (Common Voice, FLEURS, VoxPopuli, etc.) and OpenAI Whisper:
Radford et al., "Robust Speech Recognition via Large-Scale Weak Supervision", 2022.
Quellen
1 Quelle
Quellen
1 QuelleVerifiziert 7. Aug.
Modellartefakte
2 Artefakte
Modellartefakte
2 Artefaktefinal_best/model.safetensors
safetensors · 922 MB · SHA-256 ac152ad10fb9…0dae · Hugging Face
HerunterladenQuellenauszüge
2 Auszüge
Quellenauszüge
2 Auszüge--- language: - ro license: apache-2.0 library_name: transformers pipeline_tag: automatic-speech-recognition base_model: openai/whisper-small tags: - whisper - romanian - asr - speech-recognition - generated_from_trainer datasets: - mozilla-foundation/common_voice_25_0 - google/fleurs - facebook/voxpopuli metrics: - wer - cer model-index: - name: whisper-small_ro-80mel results: - task: type: automatic-speech-recognition name: Automatic Speech Recognition dataset: type: mixed name: Romanian ASR corpus (multi-source, held-out validation) metrics: - type: wer value: 9.21 name: WER (beam=5) - type: wer value: 8.8 name: WER normalized (lowercase, no punctuation) - type: cer value: 3.03 name: CER (beam=5) --- # Whisper Small — Romanian (whisper-small_ro-80mel) Fine-tuned [openai/whisper-small](https://huggingface.co/openai/whisper-small) for Romanian automatic speech recognition. Trained as part of the **Dorele** project — a fully Romanian-language, low-latency smart-home voice assistant — where Whisper models serve as on-premise STT engines. ## Results Evaluated on a held-out validation split of 25,348 samples, beam size 5: | Metric | Value | |---|---| | WER | **9.21 %** | | WER (normalized: lowercase, punctuation stripped) | 8.8 % | | CER | 3.03 % | The validation set is a deliberately hard multi-source mix (read speech, parliamentary/spontaneous speech, command-domain utterances), so numbers are not directly comparable to single-corpus benchmarks like Common Voice test. ## Training data ~185 K training samples drawn from a combined Romanian ASR corpus (~290 K samples / ~619 h total across splits), assembled from: - Common Voice 25 (Romanian) - Google FLEURS (ro_ro) - VoxPopuli (Romanian) - Echo read-speech corpus - RODigits (spoken digits) - USPDATRO - CRoWL - RO-Smart synthetic TTS data (smart-home command domain) Audio pre-processed to 80-bin log-mel features (30 s windows, Whisper standard). Splits: train 185,175 / validation 25,348 / test 78,167. ## Training procedure Full fine-tune (no LoRA) on a single RTX 5070 Ti 16 GB, BF16 autocast, SDPA attention. | Hyperparameter | Value | |---|---| | Parameters | 244 M | | Effective batch size | 192 (12 × 16 grad accum) | | Learning rate | 2e-5, cosine schedule | | Warmup steps | 750 | | Max epochs | 10 (early stopping, patience 5 on eval WER) | | Optimizer | AdamW (fused) | | Label...
Source context: 127 downloads · 0 likes · Pipeline automatic-speech-recognition · Library transformers · Repo IonGrozea/whisper-small_ro-80mel
