RUMOR: leak points to custom voices and new Avocado model for Meta's Muse

[{"h":"RUMOR: Muse may get custom voices while Meta tests an Avocado model internally","p":"Nine days after officially shipping Muse Voice Transcribe, its first real-time audio perception model, Meta is already testing the next layer of its voice plan for Muse internally, according to a report from tracker TestingCatalog published on September 10. The core of the leak is on-demand custom voices: a user would ask for a specific voice style, such as a robotic voice, and Muse would generate and store that audio in its own library, accessible through a voice button inside the interface. None of this is public, and the feature is reportedly locked outside internal testing groups."},{"h":"The package does not stop at voice","p":"Alongside on-demand voices, the same report points to a Meta AI style voice mode dedicated to Muse, reasoning controls that would let the agent toggle how much it thinks before responding, and a new model update round codenamed Avocado. In parallel, the report cites internal voice models labeled MAI being tested inside Meta itself. None of these names has been confirmed by the company as a final product, and there is no official date, region or feature list for any of them."},{"h":"Why the sequence matters","p":"The more interesting read here isn't the isolated feature, it's the order of releases. Meta Superintelligence Labs shipped Muse Voice Transcribe on September 1 as the technical base, combining streaming speech to text, speaker diarization and end of speech detection in a single model. Nine days later, leaks about sound personality features started surfacing. That suggests a deliberate script: build the voice plumbing first, then hang the personalization layer on top. If that read is right, Meta would be closing the voice gap against OpenAI, Google and Microsoft in phases, not with one big single announcement."},{"h":"What backs the rumor and what's missing","p":"The source is a single tracker specialized in AI product leaks, with no independent confirmation so far and no statement from Meta on the matter. The report describes behavior observed in internal builds, not a roadmap disclosed by the company. That classifies the lead as a single sourced but traceable rumor, not a strong signal, since it lacks a second convergent source or a checkable public artifact such as a changelog entry or app store listing. The already confirmed and publicly verifiable baseline is only Muse Voice Transcribe, priced at $3 per 1,000 audio minutes with processing every 80 milliseconds."},{"h":"What changes if confirmed","p":"If Meta does ship custom voices inside Muse, the product starts competing directly with OpenAI's Advanced Voice and Google's Gemini Live in territory that today belongs to native voice assistants, not productivity agents. For anyone building on top of AI agents, the combination of customizable voice with adjustable reasoning control changes the cost per interaction calculation in customer service and voice automation products. It's still early to treat this as settled strategy: what exists today is a leaked internal test, and the distance between a feature in testing and a shipped feature remains the point to watch."},{"h":"MaxAssistant's read","p":"Worth tracking because Meta has a recent history of moving voice pieces fast, as Voice Transcribe itself showed by going straight to production with public pricing. But today's story is about what leaked, not what shipped, and the RUMOR label stays until a second source, an official changelog or an app store entry confirms the feature."},{"h":"Sources","p":"Tech Insider, Meta Muse Custom Voices Leak: 80ms Model, $3/1K Min"}]