Key takeaways
- Smallest.ai launches Hydra, the world's first asynchronous speech-to-speech AI model.
- Hydra enables real-time conversations by listening and responding simultaneously, unlike traditional models.
- The model preserves emotional nuances and urgency by operating entirely in the audio domain.
- Sudarshan Kamath emphasizes Hydra's ability to handle complex, emotional conversations seamlessly.
- The company raised $8 million in seed funding to develop Hydra and expand its market presence.
Smallest.ai has beta-launched a speech-to-speech AI model called Hydra. Founder and CEO Sudarshan Kamath claimed this is the world’s first asynchronous model.
“OpenAI's speech-to-speech preview is impressive. But it still has the same flaw every cascaded system has had for a decade. It waits for you to finish your sentence. It waits for silence. It waits for the text layer to decode context. And in that wait, calls drop, cues get missed, and the conversation stops feeling like one. Humans don't act like that. Humans are listening, reasoning, predicting, and responding all at once. That's the difference between what was previewed this week and what we've built with Hydra,” he said in a LinkedIn post.
Hydra preserves tone, hesitation, and urgency because it never leaves the audio domain, the founder noted. It starts forming a response mid-conversation. It listens and speaks at the same time, like a human trying to understand you, the company said. The structured tool calling from live speech enables it to take action mid-utterance.
While present models operate synchronously, Hydra works asynchronously, Kamath said.
In October last year, the firm raised $8 million in seed funding to build Hydra and expand its presence across India and the United States. The round was led by Sierra Ventures, with participation from 3one4 Capital and Better Capital.
Kamath also posted a video showcasing the demo of the product, where the model is seen empathising with him when he said that his friend passed away.
“It could have complex conversations. It could laugh. It could cry. It could exhibit emotions while it was chatting with me. It would also do tool calling extremely accurately and handle complex instructions very, very quickly. And it could be interrupted very seamlessly. So we believe Hydra is the next generation of speech-to-speech model,” Kamal said.
