devagarwal2/sarvam-voice-mode

Sarvam Voice Mode is a hands-free voice chat app, think of ChatGPT's live voice mode, but built specifically for Indian languages and English. You talk naturally, pause, and the AI responds out loud. No tapping required after you start. The flow is straightforward: your microphone captures speech, which gets transcribed to text, sent to an AI language model for a response, and then that text response is converted back into speech. What makes it feel live is the silence detection, when you stop talking for about 1.3 seconds, the app automatically processes your message. You can also interrupt the AI mid-sentence by tapping the on-screen orb or pressing the spacebar, just like cutting into a real conversation. The app supports over ten Indian languages plus English, auto-detecting which one you're speaking. You can pick from three preset voices, two male, one female. This would be useful for anyone building conversational AI products aimed at Indian users, especially those more comfortable speaking than typing in regional languages. A founder prototyping a voice-based customer support tool, a PM exploring voice interfaces for non-English markets, or a developer wanting a reference implementation for real-time voice-to-voice chat could all use this as a starting point. You just need Node.js installed and an API key from Sarvam AI's dashboard to run it locally. On the technical side, the project keeps things relatively simple: it's built with Next.js and React, and all three AI capabilities, speech recognition, language processing, and text-to-speech, come from Sarvam AI's own models. The API key is handled server-side only, so it's not exposed in the browser. The animated orb visualization is rendered on a canvas element using the Web Audio API to react to your voice levels in real time. It's described as a "community edition" and is not officially affiliated with Sarvam AI.