AI4Bharat and Josh Talks AI launch Voice of India, a new evaluation platform to assess speech and language systems against India’s diverse linguistic landscape, addressing gaps in global AI benchmarks.
AI4Bharat, the IIT Madras-backed artificial intelligence company, has joined with Josh Talks AI to launch Voice of India, a new evaluation platform designed to test speech and language systems against Indian conditions rather than Western benchmarks. According to the companies, the aim is to give governments, businesses, researchers and developers an independent way to measure how well AI performs across India’s languages, accents, dialects and deployment settings.
The move comes as India pours investment into AI models, computing infrastructure and large language models, while the infrastructure for rigorous independent testing has lagged behind. The platform is meant to fill that gap by focusing on real-world speech, where linguistic variation, code-switching and noisy environments can sharply affect performance. The Voice of India site says the benchmark covers 15 languages and dialects and includes more than 36,000 speakers.
That challenge is not merely theoretical. Reporting by Business Standard has said global AI systems continue to struggle with Indian languages and dialects, with notable gaps between India-focused tools and international models. A separate description of the benchmark says it is built from unscripted telephonic conversations, with background noise and informal speech included to better reflect everyday use. That approach is intended to make the results more relevant for India’s diverse and highly localised speech patterns.
Supriya Paul, Josh Talks AI’s co-founder, said trust in AI would depend on how well systems understand the people they are intended to serve. Mitesh Khapra of AI4Bharat said evaluation was becoming as important as model-building itself, and argued that Voice of India could provide scientifically robust frameworks for India’s linguistic diversity. Josh Talks also runs a human evaluation platform for voice AI and text-to-speech systems, and has previously said such assessments can be completed in hours rather than weeks, with native speakers judging issues such as pronunciation, naturalness and intelligibility.
Disclaimer: This article is intended to inform and educate, not to recommend or endorse any financial product, investment or strategy. Please consider your own financial circumstances and seek professional advice where appropriate before making financial decisions.
