Overview
WaveForms AI developed an audio model intended to make spoken conversations with computers more natural. Founded by Alexis Conneau and Coralie Lemaitre, the company raised $40 million in seed funding led by Andreessen Horowitz. Meta acquired WaveForms in 2025; the price was not disclosed.[1][2]
Founding Story
Conneau had worked on language and speech research at Meta and OpenAI. Lemaitre brought product and business strategy experience from Google. In a May 2025 interview, Conneau said the company launched the previous September. Its premise was that voice carries information about tone and emotion that can be lost when a system converts speech to text before generating a reply.[3][4]
Timeline
- September 2024: Conneau said WaveForms launched.[3]
- December 2024: The company announced its $40 million seed round. It had five employees and was still developing its models.[1]
- May 2025: Conneau described plans for a consumer voice interface and a business API.[3]
- August 2025: The Information reported Meta had acquired WaveForms. Meta's later proxy identifies WaveForms among its 2025 acquisitions.[2][5]
What They Built
Traditional voice assistants can chain speech recognition, a text model, and speech synthesis. WaveForms pursued an end-to-end audio language model that would process and generate speech directly. Its investor described the goal as a conversation that felt indistinguishable from speaking with a person. That was a goal, not a demonstrated performance result.[4]
Conneau described early tests in which a model produced extended improvised audio and unexpected vocal mannerisms. He also described a consumer interface and an API for other organizations. Those accounts explain the intended experience and distribution; they do not establish a public launch or customer adoption.[3]
Market Position
Target Customers
The proposed consumer interface targeted people speaking directly with an assistant. The proposed API targeted businesses integrating voice into their own services. These routes have different needs: consumer use depends on the quality of the conversation, while an API also needs reliability and predictable integration.
Market Size
WaveForms entered a market where major AI companies were developing voice interfaces. Its available record does not establish revenue or paying customers.
Competition
OpenAI's Realtime API later offered direct audio processing and generation for voice agents. Amazon's Nova 2 Sonic also offered speech-to-speech interaction. These are active alternatives to building a new foundation model from scratch.[6][7]
Business Model
Conneau described both a consumer product and a business API, but no public pricing or revenue is established. Training a model and supporting a voice service would also require ongoing computing and engineering resources. The size of those costs at WaveForms is unknown.[3]
Outcome and evolution
Meta acquired the company less than a year after its reported launch. The acquisition brought specialized audio research into a much larger organization. The available sources do not establish which WaveForms models shipped independently or how its technology was integrated into Meta products.[2][5]
Key Lessons
- Voice quality is a product choice. Conneau targeted tone, timing, and interaction, not only accurate transcription.
- Model research and distribution are separate challenges. A consumer interface and an API each require a route to users.
- A research acquisition does not reveal adoption. Funding and a sale do not establish how widely a standalone product was used.