Soket AI Labs Unveils DHRITH, India’s First Emotion-Aware ASR Built for Multilingual, Code-Mixed Speech

Soket AI Labs has launched DHRITH, an India-built automatic speech recognition (ASR) system designed to understand not just words, but the emotion, tone and multilingual complexity of how Indians actually speak.

The company says the system fills a long-standing gap in India’s AI ecosystem, where global ASR tools often fail to capture cultural and linguistic nuance.

“In India, most data doesn’t live in text – it lives in voice,” the company wrote in a LinkedIn post, adding that existing ASR technologies struggled to “capture tone, emotion, code switching and speaker context with precision.”

DHRITH was developed under Project EKΛ, an open-source initiative supported by IndiaAI and the Ministry of Electronics and IT.

The project aims to build foundational AI capabilities rooted in India's diverse audio landscape—where millions speak fluidly across languages, dialects, and emotional contexts.

According to Soket AI Labs, DHRITH includes emotion tagging, code-mixed fluency, speaker diarisation, and context-aware transcription, enabling it to analyse speech more naturally and intelligently.

The system interprets how Indians shift between languages, adjust tone and express sentiment—capabilities critical for sectors like customer service, governance, healthcare and education.

DHRITH is now the firm’s in-house ASR engine, “designed to listen the way India speaks.”

The startup says it will soon support researchers and enterprises, with plans to offer developers an API and a detailed technical blog.

Soket AI Labs is a Bengaluru-based startup founded in 2019 by Abhishek Upperwal. It is one of the select ventures under India’s IndiaAI Mission, tasked with developing a 120-billion-parameter open-source Indic large-language model.