Overview
- Launched on October 30, Luna generates voice directly from audio input without converting to text, enabling features such as singing, whispering, pausing, and emotion-aware responses.
- Company tests cite sub-600 millisecond end-to-end latency, low error rates, and higher mean opinion scores than OpenAI’s GPT-4 TTS and ElevenLabs, though these results are internal claims.
- Proof-of-concept trials include an Indian automaker for in-car entertainment and a US wellness startup for interactive storytelling.
- The model is available in English at launch, with the company targeting multilingual capabilities, including Indian languages, within two to three months.
- Founder Sparsh Agrawal, 25, has seed backing from investors such as Kunal Shah, Shankar Narayana, Kunal Kapoor, and WTFund, and the company plans an API-first rollout focused on entertainment, wellness, and automotive use cases.