Qipchoq-STT
Qipchoq-STT is a fine-tuned automatic speech recognition (ASR) model based on xmullaboyeva/mullaboyeva-stt. The model converts speech audio into text and is intended for research and practical speech recognition applications.
Model description
This model was fine-tuned from xmullaboyeva/mullaboyeva-stt using a custom speech dataset. It is designed for speech-to-text transcription and can be integrated into Python applications using the Hugging Face Transformers library.
Usage
from transformers import pipeline
pipe = pipeline(
"automatic-speech-recognition",
model="xmullaboyeva/Qipchoq-STT"
)
result = pipe("audio.wav")
print(result["text"])
Intended uses
The model can be used for:
- Automatic speech recognition (ASR)
- Speech-to-text transcription
- Speech technology research
- Voice-enabled applications
- Dataset transcription and annotation
Limitations
Recognition accuracy may decrease for:
- Noisy audio recordings
- Speech with accents or dialects not represented in the training data
- Overlapping speakers
- Extremely short audio clips
Training and evaluation data
The model was fine-tuned on a custom speech dataset.
Training procedure
Training hyperparameters
| Hyperparameter | Value |
|---|---|
| Learning rate | 1e-5 |
| Train batch size | 8 |
| Evaluation batch size | 2 |
| Gradient accumulation steps | 4 |
| Total train batch size | 32 |
| Optimizer | Adam (β₁=0.9, β₂=0.999, ε=1e-8) |
| Learning rate scheduler | Cosine |
| Warmup ratio | 0.05 |
| Number of epochs | 5 |
| Random seed | 42 |
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.2231 | 1.97 | 200 | 0.1962 |
| 0.1199 | 3.93 | 400 | 0.2005 |
Final validation loss: 0.2005
Framework versions
- Transformers 4.38.0
- PyTorch 2.9.1+cu126
- Datasets 2.18.0
- Tokenizers 0.15.2