Customer Support & Dictation Loom Large As Voice Startup Leaders Prepare for a World of Talking Machines
Voice models are a few steps behind text models but the industry is preparing for takeoff. Our takeaways from the Cerebral Valley Voice Summit.
Voice computing true believers gathered at our Cerebral Valley Voice Summit on Wednesday in San Francisco. We gave them hats that said, “VOICEPILLED.”
On stage, Sierra CEO Bret Taylor explained how his customer support-oriented company is taking Silicon Valley by storm raising $950 million this week at a $15.8 billion valuation led by Tiger Global and GV. He shared that Sierra’s agents are widespread enough that they’ve already talked to each other on the phone — “they weren’t conspiring, as far as I know,” he joked.
Wispr Flow, the rapidly growing voice dictation software, captured the audience’s attention. Tanay Kothari told the crowd how his company takes great care in turning your stream of consciousness into correctly punctuated prose. “What Wispr does as you’re using it the first five, six times, is it learns your comma patterns. That really small detail is what adds to the delight.” (Ramp recently found that Wispr Flow is the third-fastest growing software vendor.)
Meanwhile, OpenAI’s head of realtime AI Justin Uberti charted the history of voice technology without trying to give away the open secret that on Thursday OpenAI would reveal a suite of new voice models, giving them the power to reason and properly follow the context of a conversation even when interrupted.
A few takeaways:
Cascade systems where companies cobble together a chain of text to speech and speech to text models are dominant right now, but everyone is eyeing improvements coming from voice-to-voice models.
Granola and Wispr Flow are probably the two buzziest user-facing applications. Several speakers brought up the potential for better voice therapists or companions. Brandon Yang at Cartesia shouted out Tolan, an AI alien “best friend,” as an under-the-radar startup to watch.
Even true believers are divided between people who see voice as a return to a more natural form of human communication and others who are optimistic but see it as something to be paired with graphical interfaces or reserved for moments like when people are driving.
Olivia Moore, the summit’s resident consumer whisperer, said she had turned her focus to business applications. “On the consumer side, I’m still very optimistic, I think consumer just takes longer to mature,” she said. “It cannot be, in my opinion, predicted — only observed.”
Several of the speakers Wednesday envisioned a clear division between AI agents in the workplace, which will mostly be aimed at efficiency, and personal companion agents that might be assistants, therapists, and friends all at once.
“By 2030, we will have two general purpose AIs,” said Eugenia Kuyda, founder of Wabi and previously the founder of AI companion-maker Replika. One will be for knowledge work, she said, and the other will be a personal agent that will be getting to know you and “helping you flourish.”
The Cerebral Valley Voice Summit is co-hosted by Eric Newcomer, Max Child, and James Wilsterman.
Special thanks to our sponsors Baseten, Felicis, Nebius, AssemblyAI, and Weekend for making it possible.
Newcomer subscribers can scroll down to get the full breakdown of our onstage content from yesterday.
