Vividh-ASR: Diagnosing and Fixing Studio-Bias in Whisper for Indic Languages
• 11
AI & ML interests
Speech and NLP systems for Indian legal infrastructure
Recent Activity
View all activity
Papers
Vividh-ASR: A Complexity-Tiered Benchmark and Optimization Dynamics for Robust Indic Speech Recognition
CourtNav: Voice-Guided, Anchor-Accurate Navigation of Long Legal Documents in Courtrooms
Articles
kavyamanohar
authored a
paper 3 days ago
Kush0610
authored a
paper 3 days ago
orgh0
authored a
paper 3 days ago
kavyamanohar
authored 5
papers 3 days ago
Romanized to Native Malayalam Script Transliteration Using an Encoder-Decoder Framework
Paper • 2412.09957 • Published
Vividh-ASR: A Complexity-Tiered Benchmark and Optimization Dynamics for Robust Indic Speech Recognition
Paper • 2605.13087 • Published • 3
Multistage Fine-tuning Strategies for Automatic Speech Recognition in Low-resource Languages
Paper • 2411.04573 • Published
What is lost in Normalization? Exploring Pitfalls in Multilingual ASR Model Evaluations
Paper • 2409.02449 • Published
Syllable Subword Tokens for Open Vocabulary Speech Recognition in Malayalam
Paper • 2301.06736 • Published
janaab
authored a
paper 3 days ago
Request for Datasheet
1
#1 opened 8 days ago
by 
janaab
Request for Datasheet
1
#2 opened 8 days ago
by janaab
kavyamanohar
updated 2
datasets 5 days ago
kavyamanohar
posted an update 5 days ago
Post
4541
Releasing Vividh-ASR — an open benchmark and models for Hindi and Malayalam ASR.
Vividh-ASR is built from public data, stratified by complexity:
→ Clean recordings
→ Noisy and accented speech
→ Spontaneous, conversational audio
Alongside the benchmark, we release:
→ Open models for Hindi and Malayalam
→ A training recipe with two counterintuitive choices that moved the needle
→ What failed, not just what worked
The stratified evaluation methodology transfers directly to any low-resource language setup — beyond Hindi and Malayalam.
Built at @adalatai , where we build speech tech for Indian courts. This is our first open contribution back to the community. @janaab @Kush0610 @orgh0
Link: https://huggingface.co/blog/adalat-ai/vividh-benchmark
Vividh-ASR is built from public data, stratified by complexity:
→ Clean recordings
→ Noisy and accented speech
→ Spontaneous, conversational audio
Alongside the benchmark, we release:
→ Open models for Hindi and Malayalam
→ A training recipe with two counterintuitive choices that moved the needle
→ What failed, not just what worked
The stratified evaluation methodology transfers directly to any low-resource language setup — beyond Hindi and Malayalam.
Built at @adalatai , where we build speech tech for Indian courts. This is our first open contribution back to the community. @janaab @Kush0610 @orgh0
Link: https://huggingface.co/blog/adalat-ai/vividh-benchmark
Request for Datasheet
1
#1 opened 8 days ago
by janaab
Request for Datasheet
1
#2 opened 8 days ago
by janaab
kavyamanohar
updated 4
models 10 days ago
adalat-ai/whisper-small-ml-rmft
Automatic Speech Recognition • 0.2B • Updated • 60
adalat-ai/whisper-small-ml-high-lr
Automatic Speech Recognition • 0.2B • Updated • 54
adalat-ai/whisper-small-hi-rmft
Automatic Speech Recognition • 0.2B • Updated • 75
adalat-ai/whisper-small-hi-high-lr
Automatic Speech Recognition • 0.2B • Updated • 79