Telugu ASR Fine-Tuned Model
Overview
This repository contains a fine-tuned Automatic Speech Recognition (ASR) model for the Telugu language.
The model is built upon Swecha Gonthuka ASR and further fine-tuned using a custom Telugu speech dataset to improve transcription quality on domain-specific Telugu speech.
The model converts spoken Telugu audio into Telugu Unicode text and can be used for:
- Automatic Speech Recognition (ASR)
- Speech-to-Text applications
- Voice transcription
- Voice assistants
- Educational applications
- Telugu speech processing research
Model Details
| Property | Value |
|---|---|
| Model Type | Wav2Vec2ForCTC |
| Language | Telugu |
| Framework | Hugging Face Transformers |
| Base Model | swechatelangana/swecha-gonthuka-asr |
| Task | Automatic Speech Recognition |
| Fine-tuned By | Dinesh |
Base Model
This model has been fine-tuned from:
swechatelangana/swecha-gonthuka-asr
Original architecture:
Harveenchadha/wav2vec2-pretrained-clsril-23-10k
↓
swechatelangana/swecha-gonthuka-asr
↓
dinesh480/telugu-asr-finetuned
Fine-Tuning Dataset
The model was fine-tuned on a custom Telugu speech dataset.
Dataset Characteristics:
- Language: Telugu
- Audio Format: WAV
- Sampling Rate: 16 kHz
- Mono Audio
- UTF-8 Telugu Transcriptions
Training Configuration
| Hyperparameter | Value |
|---|---|
| Framework | Transformers |
| Architecture | Wav2Vec2ForCTC |
| Optimizer | AdamW |
| Learning Rate | 1e-4 |
| Epochs | 10 |
| Batch Size | 8 |
| Mixed Precision | FP16 |
| Loss | CTC Loss |
Usage
Install dependencies
pip install transformers torchaudio librosa
Load the model
from transformers import AutoProcessor
from transformers import AutoModelForCTC
processor = AutoProcessor.from_pretrained(
"dinesh480/telugu-asr-finetuned"
)
model = AutoModelForCTC.from_pretrained(
"dinesh480/telugu-asr-finetuned"
)
Or use the pipeline
from transformers import pipeline
pipe = pipeline(
"automatic-speech-recognition",
model="dinesh480/telugu-asr-finetuned"
)
result = pipe("audio.wav")
print(result["text"])
Example
Input Audio
audio.wav
Predicted Output
ఈ రోజు వాతావరణం చాలా బాగుంది.
Performance
The model has been fine-tuned to improve Telugu speech transcription quality over the base model on the custom training dataset.
Evaluation metrics may include:
- Word Error Rate (WER)
- Character Error Rate (CER)
Applications
This model can be used for:
- Telugu Speech Recognition
- Audio Transcription
- Educational Projects
- Research
- Voice Assistants
- Speech Analytics
- Subtitle Generation
Limitations
Performance may decrease for:
- Heavy background noise
- Multiple speakers
- Very low-quality recordings
- Strong regional accents not present in the training data
Citation
If you use this model in your research, please cite:
@misc{telugu_asr_finetuned,
title={Telugu ASR Fine-Tuned Model},
author={K. Dinesh},
year={2026},
publisher={Hugging Face}
}
Acknowledgements
This project is based on:
- Swecha Telangana
- Hugging Face Transformers
- PyTorch
- Wav2Vec2
- Hugging Face Hub
Special thanks to the developers of Swecha Gonthuka ASR for releasing the original Telugu ASR model.
License
This model follows the Apache-2.0 License inherited from the base model.