AI
Startup Smallest.ai Raises $13M to Build Voice AI That Mimics Natural Human Conversation
A newly funded voice AI startup is rethinking how agents interact by using smaller, specialized models that listen and respond in real time rather than processing long prompts like traditional AI.
TechCrunch — AI
Most voice AI assistants today still feel robotic to users, despite advances in agent capability. Startup Smallest.ai, launched in late 2024, is pursuing a different path: rather than simply making large language models faster, the company builds lightweight, specialized models engineered specifically for natural conversation.
The core insight is straightforward—humans listen, think, and speak simultaneously, and any delay in an AI response sounds unnatural. Smallest.ai's approach mirrors that human rhythm.
How the Model Works#
The startup's small voice model processes information in real time without the latency that plague traditional LLM-based systems. Founder and CEO Sudarshan Kamath explains the distinction: whereas an LLM typically receives an entire prompt before responding, the new model generates replies incrementally as it listens, much as a human would.
When the model encounters questions outside its specialized knowledge base, it gracefully hands off to a larger foundational model—effectively placing the customer on brief hold to "research" the issue, just as human agents do.
Kamath envisions a two-tier architecture becoming standard: small voice models for real-time interaction and larger offline LLMs for complex problem-solving.
Focus on Voice-Specific Challenges#
Unlike general-purpose AI, Smallest.ai concentrates entirely on voice-specific requirements: handling diverse accents, supporting dozens of languages, and performing reliably in noisy environments.
The company's customers already include established names like RingCentral and Truecaller, with potential expansion into any customer support business.
Funding and Competition#
The $13 million Series A round, led by Seligman Ventures and including Sierra Ventures and 3one4 Capital, brings total funding to over $21 million.
Smallest.ai competes with established voice AI players like ElevenLabs and Cartesia, as well as regional specialists such as Sarvam. However, while competitors span audio dubbing and podcasting, Smallest.ai remains laser-focused on conversational agents for enterprise customers—a narrower but potentially high-value niche.
Kamath's ultimate ambition is simple: "You should speak to our model and not know it's AI or human."
