Downloads · 30 days
11
9% of all-time downloads
KSE-RESEARCH-Group/whisper-small-dido-yvanchyk-v2
whisper-small-dido-yvanchyk-v2 is a automatic speech recognition model from KSE-RESEARCH-Group. Use it when you need speech turned into text. The card lists the license as mit.
This model is a fine-tuned version of openai/whisper-small on the Dido Yvanchyk Audio Dataset v2 for Ukrainian speech recognition.
Downloads · 30 days
11
9% of all-time downloads
All-time downloads
119
Public
Parameters
242M
967 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors967 MB · 99%
From the Hugging Face model README
This model is a fine-tuned version of openai/whisper-small on the Dido Yvanchyk Audio Dataset v2 for Ukrainian speech recognition.
Fine-tune openai/whisper-small on Dido-Yvanchyk dataset
| Property | Value |
|---|---|
| Dataset | KSE-RESEARCH-Group/Dido-Yvanchyk-Audio-Dataset-v2 |
| Training samples | 6811 |
| Validation samples | 757 |
| Test samples | 842 |
| Language | Ukrainian |
| Max token length | 448 |
| Parameter | Value |
|---|---|
| Base model | openai/whisper-small |
| Learning rate | 1e-05 |
| Warmup steps | 500 |
| Max steps | 8000 |
| Batch size (per device) | 4 |
| Gradient accumulation steps | 4 |
| Effective batch size | 16 |
| FP16 | True |
| Gradient checkpointing | False |
| Eval strategy | steps |
| Eval/Save steps | 500 |
| Metric for best model | cer |
The model was trained for 8000 steps with evaluation every 500 steps. The best checkpoint was selected based on the lowest CER.
| Step | Train Loss | Eval Loss | Eval CER (%) | Eval WER (%) |
|---|---|---|---|---|
| 500 | 0.239 | 0.315 | 8.39 | 24.72 |
| 1000 | 0.132 | 0.252 | 6.33 | 19.48 |
| 1500 | 0.079 | 0.251 | 5.73 | 19.37 |
| 2000 | 0.06 | 0.263 | 5.79 | 19.34 |
| 2500 | 0.027 | 0.28 | 5.1 | 18.42 |
| 3000 | 0.015 | 0.298 | 5.32 | 18.31 |
| 3500 | 0.007 | 0.305 | 4.88 | 18.02 |
| 4000 | 0.006 | 0.316 | 4.79 | 17.42 |
| 4500 | 0.002 | 0.326 | 4.89 | 17.56 |
| 5000 | 0.002 | 0.335 | 4.75 | 17.36 |
| 5500 | 0.001 | 0.339 | 4.81 | 17.58 |
| 6000 | 0.001 | 0.345 | 4.75 | 17.43 |
| 6500 | 0.001 | 0.347 | 4.68 | 17.45 |
| 7000 | 0.001 | 0.352 | 4.77 | 17.49 |
| 7500 | 0.0 | 0.355 | 4.71 | 17.59 |
| 8000 | 0.001 | 0.357 | 4.71 | 17.6 |
Best Model Checkpoint: Step 6500
| Metric | Value |
|---|---|
| CER | 4.71% |
| WER | 17.6% |
| Eval Loss | 0.357 |
| Metric | Value |
|---|---|
| CER | 4.72% |
| WER | 17.84% |
from transformers import pipeline
import torch
device = "cuda:0" if torch.cuda.is_available() else "cpu"
pipe = pipeline(
"automatic-speech-recognition",
model="artem-orlovskyi/whisper-small-whisper-small-dido-yvanchyk-v2",
device=device,
)
result = pipe(
"path/to/audio.wav",
generate_kwargs={
"task": "transcribe",
"language": "ukrainian",
},
chunk_length_s=30,
)
print(result["text"])
from transformers import WhisperForConditionalGeneration, WhisperProcessor
import torch
model_id = "artem-orlovskyi/whisper-small-whisper-small-dido-yvanchyk-v2"
processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)
# Move to GPU if available
device = "cuda:0" if torch.cuda.is_available() else "cpu"
model = model.to(device)
# Process audio (audio_array should be a numpy array at 16kHz)
input_features = processor(
audio_array,
sampling_rate=16000,
return_tensors="pt"
).input_features.to(device)
# Generate transcription
predicted_ids = model.generate(input_features)
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
print(transcription)
| Component | Specification |
|---|---|
| GPU | NVIDIA GeForce RTX 4090 |
| GPU Memory | 47.4 GB |
| GPU Count | 1 |
| CUDA Compute Capability | 8.9 |
| Package | Version |
|---|---|
| Python | 3.12.12 |
| PyTorch | 2.8.0+cu128 |
| CUDA | 12.8 |
| Transformers | 4.57.3 |
| Datasets | 2.21.0 |
| Evaluate | 0.4.6 |
| Metric | Value |
|---|---|
| Total training time | 1 day, 14:19:54.145862 |
| Training started | 2025-12-27 16:03:58 |
| Training completed | 2025-12-29 06:23:52 |
| Property | Value |
|---|---|
| Experiment ID | whisper-small-005 |
| WandB Project | dido-yvanchik-stt |
| WandB Run | whisper-small-005 |
If you use this model, please cite:
@misc{artem-orlovskyi-whisper-small-whisper-small-dido-yvanchyk-v2,
author = {artem-orlovskyi},
title = {whisper-small - Fine-tuned for Ukrainian ASR},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/artem-orlovskyi/whisper-small-whisper-small-dido-yvanchyk-v2}
}
This model is released under the MIT license.