Skip to content
← Projects

IndiVoice-DeepASR

Bridging the accent gap in ASR with Whisper + LoRA

Domain
Speech
Status
Active
Year
2026
Stars
★ 6

IndiVoice-DeepASR fine-tunes OpenAI's Whisper with LoRA to improve recognition of Indian-accented English, including Hindi, Tamil, Kannada, Bengali and Punjabi regional accents. It trains on the Svarah dataset with fault-tolerant checkpointing for Colab and Kaggle.

The problem

The README states commercial ASR systems show a 20-30% performance degradation on Indian English accents.

How it works

  1. 01

    Standardize audio to 16 kHz mono

  2. 02

    Fine-tune Whisper with LoRA adapters (under 2% of parameters)

  3. 03

    Checkpoint every 100 steps with auto-resume

  4. 04

    Evaluate with WER/CER; Gradio deployment

Figures

IndiVoice-DeepASR figure