Meta has acquired WaveForms AI, an audio AI startup focused on systems that can recognize and mimic emotions in speech, according to a report from The Information. The deal is part of a broader AI overhaul at Meta following recent setbacks, with the company reorganizing staff, technology, and investments.
Back in April, Meta CEO Mark Zuckerberg said that everyone will eventually have their own AI they talk to every day - a vision the WaveForms acquisition is meant to support.
Audio as a strategic focus
WaveForms AI first came onto the scene in December after raising $40 million in a round led by Andreessen Horowitz. The company was co-founded by Alexis Conneau, who spent nearly eight years on audio research at Meta before leading audio projects for OpenAI's GPT-4o, and Coralie Lemaitre, previously involved in business strategy for Google's advertising division.
Both Conneau and Lemaitre are joining the new Meta Superintelligence Labs, led by Chief AI Officer Alexandr Wang and former GitHub CEO Nat Friedman.
The WaveForms deal fits into Meta's recent audio push. Other recent hires include Johan Schalkwyk, who led the machine learning team at voice AI startup Sesame. On top of that, Meta acquired the voice AI startup PlayAI last month.
MSL and TBD Lab: Llama 4.x, reasoning, agents
Meta is now consolidating its AI work under the new Meta Superintelligence Labs (MSL) umbrella. Within MSL is a special unit called "TBD Lab" ("to be determined"), which, according to The Wall Street Journal, is leading development of the next Llama model generation and has recruited top researchers from rival companies. The project is headed by Jack Rae, a recent hire from Alphabet's Google. Internally, names like "Llama 4.5" and "Llama 4.X" are already circulating, the WSJ reports.
In a memo to staff, Wang wrote that TBD Lab is working with other AI teams on upcoming model releases, advancing reasoning capabilities, and building AI agents. The new structure is designed to enable more ambitious technical goals, allow for parallel development, and accelerate high-performance results, according to the WSJ.