What the business is
An audio AI startup building audio LLMs that understand and express emotion in spoken interaction.
The bet
Audio is the social-emotional layer of AGI: audio LLMs that read tone, inflection and accent and answer with real emotion can do what text-only assistants cannot.
How it started
Alexis Conneau joined OpenAI three years earlier as a Facebook researcher after emailing his AI hero; he led the research team behind ChatGPT's voice feature, which drew unflattering headlines when one voice resembled Scarlett Johansson's.
What happened
Conneau left OpenAI in September 2024, posting 'After an amazing journey at @OpenAI building #Her, I've decided to start a new company', and founded WaveForms with Coralie Lemaitre, an engineer turned business strategist.
How it ended up
WaveForms emerged from stealth on 9 December 2024 backed by $40 million in seed funding from a16z, promising audio LLMs that capture emotional subtleties, understand conversational context and respond in kind.
What has to be true
Voice emotion is orthogonal to the text-and-reasoning race the big labs are running.
The team's ChatGPT voice work is direct, hard-to-copy experience in production audio AI.
a16z's $40 million seed signals conviction that emotional audio is a category, not a feature.
Consumer familiarity with AI voice — fair or not — gives an emotional-audio product instant context.
What can be applied
The specialist lane inside a general race can be its own company: if the giants are optimising for AGI broadly, the emotional layer of voice is left unclaimed.
Aftermath
As of the 9 December 2024 launch, WaveForms was building audio LLMs for emotional understanding, with no product shipped publicly yet.
FOLLOW THE EVIDENCE