VCs Are Keeping an Ear Out for Voice AI Startups

Each new month of the AI boom has brought along with it a new “AI startup flavor of the month” for venture capitalists to drool over. The latest one, VCs tell me, are startups building AI voice agents for various verticals, from hotels to hospitals.

Up until recently, AI software that can understand and respond with human speech has largely been too slow or error-prone to use in any real way beyond simple phone trees. In some cases, startups have found workarounds by taking the spoken request of a customer, transcribing that to text to feed to a large language model, and then taking the LLM’s response and converting it back into audio. That process, however, can be very slow.

In the last year or two, though, AI model developers including OpenAI, Deepgram and most recently Google have released new AI that can understand speech and respond with spoken comments rather than text. These advances are creating new opportunities for startups building real-time voice agents to communicate with customers. While these models tend to be more expensive than text-only LLMs, prices have dropped significantly over the last year: Google’s recently-released Live API is less than one-ninth the price of OpenAI’s equivalent model.