Telugu ASR Fine-Tuned Model
Overview
This repository contains a fine-tuned Automatic Speech Recognition (ASR) model for the Telugu language. The model is based on Swecha Gonthuka ASR and has been fine-tuned using a custom Telugu speech dataset.
The model is capable of converting spoken Telugu audio into Telugu text and can be used for speech transcription and related ASR applications.
Base Model
- Model:
swechatelangana/swecha-gonthuka-asr - Architecture: Wav2Vec2ForCTC
- Framework: Hugging Face Transformers
- Language: Telugu
Fine-Tuning Dataset
A custom Telugu speech dataset was used for fine-tuning.
| Dataset Split | Samples |
|---|---|
| Train | 712 |
| Validation | 89 |
| Test | 89 |
All audio files were resampled to 16 kHz before training.
Training Configuration
| Parameter | Value |
|---|---|
| Base Model | Swecha Gonthuka ASR |
| Framework | Hugging Face Transformers |
| Optimizer | AdamW |
| Epochs | 10 |
| Learning Rate | 1e-4 |
| Batch Size | 4 |
| Sampling Rate | 16000 Hz |
Evaluation Results
| Metric | Score |
|---|---|
| Word Error Rate (WER) | 0.5054 |
| Accuracy | 49.46% |
Note: The evaluation was performed on the held-out Telugu test dataset.
Model Files
This repository includes:
- ✅ Fine-tuned model weights (
model.safetensors) - ✅ Configuration (
config.json) - ✅ Processor
- ✅ Tokenizer
- ✅ Vocabulary
- ✅ Feature Extractor
Usage
Install the required libraries:
pip install transformers librosa torch
Load the model
from transformers import AutoProcessor, AutoModelForCTC
processor = AutoProcessor.from_pretrained(
"navyasree2006/telugu-asr-finetuned"
)
model = AutoModelForCTC.from_pretrained(
"navyasree2006/telugu-asr-finetuned"
)
Transcribe an audio file
import librosa
import torch
audio_path = "sample.mp3"
speech, sr = librosa.load(audio_path, sr=16000)
inputs = processor(
speech,
sampling_rate=16000,
return_tensors="pt",
padding=True
)
with torch.no_grad():
logits = model(**inputs).logits
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.batch_decode(predicted_ids)[0]
print(transcription)
Applications
This model can be used for:
- Telugu Speech-to-Text
- Voice Assistants
- Subtitle Generation
- Audio Transcription
- Educational Applications
- Accessibility Tools
- Voice-based Interfaces
Repository
- Base Model: https://huggingface.co/swechatelangana/swecha-gonthuka-asr
- Fine-Tuned Model: https://huggingface.co/navyasree2006/telugu-asr-finetuned
Author
Surya Navyasree
B.Tech – Computer Science and Engineering (Data Science)
Rajeev Gandhi Memorial College of Engineering and Technology (RGMCET)
GitHub: https://github.com/SuryaNavyasree
Hugging Face: https://huggingface.co/navyasree2006
Acknowledgements
This work was completed using:
- Swecha Telangana
- Hugging Face Transformers
- PyTorch
- Mozilla Common Voice
- Google Colab
Citation
If you use this model in your work, please cite this repository and the original Swecha Gonthuka ASR model.