Vividh-ASR: A Complexity-Tiered Benchmark and Optimization Dynamics for Robust Indic Speech Recognition

TL;DR

Vividh-ASR introduces a complexity-stratified benchmark for Indian languages and proposes R-MFT, a training method that enhances multilingual ASR performance while maintaining model efficiency.

Contribution

The paper presents Vividh-ASR benchmark and R-MFT training recipe, improving low-resource language speech recognition and understanding model adaptation dynamics.

Findings

Early large parameter updates improve global WER by 12 points.

A hard-to-easy curriculum adds gains for spontaneous speech.

R-MFT enables smaller models to match larger fine-tuned models.

Abstract

Fine-tuning multilingual ASR models like Whisper for low-resource languages often improves read speech but degrades spontaneous audio performance, a phenomenon we term studio-bias. To diagnose this mismatch, we introduce Vividh-ASR, a complexity-stratified benchmark for Hindi and Malayalam across four tiers: studio, broadcast, spontaneous, and synthetic noise. Through a controlled study of learning-rate timing and curriculum ordering, we find that early large parameter updates improve global WER by 12 absolute points, while a hard-to-easy curriculum adds gains for spontaneous speech. These findings motivate reverse multi-stage fine-tuning (R-MFT), a training recipe that enables a parameter-efficient 244M Whisper model to match or exceed conventionally fine-tuned 769M counterparts. Representational analysis via CKA and SVD reveals effective schedules concentrate adaptation in the decoder, preserving the pre-trained encoder's acoustic geometry. We release the benchmark and models.