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 consisting of paired audio recordings and transcriptions.

The model converts spoken Telugu audio into Telugu text and is intended for speech transcription, accessibility, and language technology research.

Base Model

• Model: swechatelangana/swecha-gonthuka-asr
• Architecture: Wav2Vec2ForCTC
• Framework: Hugging Face Transformers
• Language: Telugu (te)

Fine-Tuning Dataset

The model was fine-tuned on a Telugu speech corpus containing:

• Telugu audio recordings
• Corresponding Telugu text transcriptions
• Audio sampled at 16 kHz

Intended Uses

This model can be used for:

• Telugu Speech-to-Text
• Automatic Speech Recognition (ASR)
• Voice-enabled applications
• Telugu transcription
• Language technology research

Usage

import librosa
import torch
from transformers import AutoProcessor, AutoModelForCTC

processor = AutoProcessor.from_pretrained(
    "Sujji77/swecha-gonthuka-finetuned"
)

model = AutoModelForCTC.from_pretrained(
    "Sujji77/swecha-gonthuka-finetuned"
)

audio, sr = librosa.load("sample.wav", sr=16000)

inputs = processor(
    audio,
    sampling_rate=16000,
    return_tensors="pt"
)

with torch.no_grad():
    logits = model(**inputs).logits

pred_ids = torch.argmax(logits, dim=-1)

transcription = processor.batch_decode(pred_ids)[0]

print(transcription)

Training Details

Limitations

• Performance depends on recording quality.
• Background noise may reduce transcription accuracy.
• Best suited for Telugu speech.

Author

Srujana Talari

Frontend Developer • AI Intern

Citation

If you use this model in your research or application, please cite this repository. v