Jaipur startup Pixa AI unveils Luna, the world’s first speech-to-speech foundational model

The homegrown innovation delivers human-like, emotionally intelligent conversations without text conversion

In a breakthrough for conversational AI, Jaipur-based startup Pixa AI, founded by 25-year-old Sparsh Agrawal, has introduced Luna AI, the world’s first speech-to-speech foundational model capable of understanding and generating natural, emotionally expressive speech — including whispering, pausing, and even singing — in real time.

Unlike conventional voice systems that rely on speech-to-text and text-to-speech conversion, Luna AI directly processes and responds to audio, eliminating conversion delays and dramatically improving conversational fluidity. The model enables zero-latency, emotion-aware interactions that feel more human and contextually intelligent.

Redefining the future of voice interaction

Luna’s architecture represents a significant leap in AI design, allowing for nuanced vocal behaviors such as tone modulation, natural pauses, and emotional adaptation. Benchmark results show up to 50% lower latency and higher emotional fidelity compared to leading text-mediated systems.

In internal tests, Luna outperformed top global models, including those from OpenAI, Deepgram, and ElevenLabs. It achieved an Automatic Speech Recognition (ASR) error rate of 5.24%, a Text-to-Speech Word Error Rate (WER) of 1.3%, and a Mean Opinion Score (MOS) of 4.62, surpassing industry benchmarks in accuracy and naturalness.

“I didn’t have a research lab or a $100 million runway,” said Sparsh Agrawal, Founder of Pixa AI. “Luna proves that world-class technology can be built from India — with resourcefulness, not just resources. This is the spirit of the IndiaAI Mission: innovation that’s homegrown, open, and globally competitive.”

From India to the world

Trained on millions of hours of speech data, Luna is fine-tuned for real-time emotion recognition and contextual response, unlocking applications across entertainment, mobility, wellness, and interactive media. Pixa AI plans to license Luna to partners worldwide, supporting use cases in automotive voice systems, gaming, virtual companions, and wellness platforms.

Backed by WTFund, Kunal Shah (CRED), Shankar Narayana (ex-Carlyle Asia), and Kunal Kapoor, Pixa AI stands out as a pioneering Indian innovator challenging global incumbents. While Big Tech focuses on utility-driven voice bots, Luna aims to make conversations more expressive and human — transforming AI from a tool into a true companion.