Smallest.ai, a startup founded in late 2024, has raised $13 million in a Series A round led by Seligman Ventures, with participation from Sierra Ventures and 3one4 Capital, bringing its total funding to over $21 million. The company is developing small, specialized voice models designed to make AI agents sound genuinely human in real-time conversations, a critical step for customer support applications where even a short pause feels unnatural.
“While I’m speaking to you, you’re already thinking, and you might interrupt me if I talk for too long,” said Sudarshan Kamath, founder and CEO of Smallest.ai, explaining how the model mimics human conversational patterns. Unlike large language models that process an entire prompt before responding, Smallest.ai’s model listens, thinks, and speaks simultaneously, virtually eliminating response lag.
Why small voice models are the next frontier in AI
Large foundational models like GPT-4 or Claude are powerful but inherently slow for voice interactions. Kamath argues that the future of AI agents will rely on a two-model architecture: a small, real-time voice model for natural conversation, and an offline LLM that can be called upon for complex problem-solving. This hybrid approach allows the voice model to handle routine interactions with zero latency, while handing off to a larger model when it encounters a query outside its knowledge base—briefly placing the customer on hold, just as a human would.
The startup focuses exclusively on voice-specific nuances, such as handling diverse accents, supporting dozens of languages, and operating in noisy environments—areas where general-purpose LLMs often struggle. This specialization is what allows Smallest.ai to achieve the low latency and naturalness that customers expect.
Market positioning and competition
Smallest.ai competes with voice AI leaders like ElevenLabs and Cartesia, as well as regional players such as Sarvam, which focus on local languages. However, Kamath differentiates his company by targeting real-time conversational voice agents for enterprise customers, rather than use cases like audio dubbing or podcasting. “We want our models to break the Turing test,” he said. “You should speak to our model and not know it’s AI or human. That’s the sole focus of the company.”
The startup’s existing customers include RingCentral and Truecaller, both of which rely on voice interactions. Kamath also sees potential in newer AI customer support companies like Sierra and Decagon, noting that for these startups, building a specialized voice model is a distraction from their core business.
What this means for the future of customer support
The funding comes at a time when AI agents are increasingly deployed in customer service, but user experience often suffers from robotic responses and awkward pauses. Smallest.ai’s approach could set a new standard for what “human-like” means in AI interactions, potentially making it difficult for customers to tell whether they’re speaking to a machine or a person.
For businesses, this technology promises to reduce friction in customer support, improving satisfaction while lowering operational costs. However, the ethical implications of AI that is indistinguishable from humans are significant, raising questions about transparency and disclosure. As the technology matures, regulators and companies will need to address these concerns, but for now, the race to make AI sound human is accelerating.
