awasthiabhijeet/Error-Driven-ASR-Personalization

Code for "Error-driven Fixed-Budget ASR Personalization for Accented Speakers" in ICASSP 2021

27
/ 100
Experimental

This project helps speech scientists and researchers improve Automatic Speech Recognition (ASR) systems for speakers with diverse accents. It takes existing accented speech audio and its human-verified transcripts to identify common recognition errors. The output is a highly targeted, smaller subset of data that can be used to fine-tune an ASR model, leading to better accuracy for accented speech without needing a massive amount of new data.

No commits in the last 6 months.

Use this if you need to personalize an ASR system to better understand specific accented speakers using a limited budget of additional training data.

Not ideal if you are looking for an off-the-shelf ASR system or if you have an abundance of new, fully transcribed accented speech data for fine-tuning.

speech-recognition accented-speech model-personalization speech-tech-research data-efficient-learning
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 5 / 25
Maturity 8 / 25
Community 14 / 25

How are scores calculated?

Stars

11

Forks

3

Language

Python

License

Last pushed

Jun 13, 2021

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/voice-ai/awasthiabhijeet/Error-Driven-ASR-Personalization"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.