Meta has released two new artificial intelligence models: Muse Voice Transcribe, designed for audio processing and transcription, and Muse Spark 1.3, an updated version of its large language model.
Muse Voice Transcribe is capable of performing real-time audio transcription while identifying and differentiating multiple speakers in a conversation. It can also process longer instructions and manage multiple simultaneous workflows within a single session, improving efficiency for complex tasks.
Among the model’s notable features is closer interaction with the user — the system asks clarifying questions when necessary and requests confirmation before executing irreversible actions. It also demonstrates greater resistance to prompt injection attacks and a reduced tendency to generate inaccurate information, owing to an improved ability to recognise and handle its own limitations.
Muse Spark 1.3 delivers more optimised performance, consuming approximately 25% fewer tokens and making 20% fewer tool calls compared with its predecessor. The model is available to developers via API and is expected to reach Instagram, Facebook, and the Meta AI application in the near future, according to the company.
Muse Voice Transcribe is Meta’s first real-time audio model, capable of transcribing and dictating complex or noisy audio with speed and precision. The system can natively identify and differentiate the speech of more than 20 simultaneous speakers in sessions lasting up to one hour.
The model was trained on more than 70 languages, 25 of which were validated at launch, and handles code-switching — the alternation between two or more languages within the same conversation or sentence. To balance speed and accuracy, the system adjusts response time in real time, pausing slightly longer to analyse difficult terms while confirming easily identifiable words more immediately.
The model is already integrated into the Meta AI desktop application for Mac and the Muse Code developer tool. API access is priced at three dollars per 1,000 minutes of audio for developers.
With this initiative, Meta aims to strengthen its position in the AI market and increase its competitiveness against models developed by other companies in the sector. The push includes more optimised architectures and tools capable of processing information faster and more efficiently.
Source: Diário Económico
Original article: https://www.diarioeconomico.co.mz/?p=536763
