← Latest news 
Speech AI breaks on Indian accents as major global systems stumble in new benchmark
Technology
Published on 24 April 2026
Top OpenAI and Microsoft models lag behind Sarvam
A new benchmark finds leading speech AI systems struggle to understand India. OpenAI and Microsoft models perform poorly on real-world audio featuring code-switching and regional variations, while Sarvam models lead in Indian language recognition. The gap threatens digital inclusion for millions who rely on voice tech, from assistants to support services, where accuracy remains uneven.
- OpenAI and Microsoft speech systems lag on Indian language recognition
- Sarvam models lead in handling Indian accents and speech patterns
- The benchmark tests real-world audio, including code-switching
- Inaccurate voice AI could limit digital inclusion for millions
Read the full story at The Economic Times
This summarization was done by Beige for a story published on 
The Economic Times