Aug. 29, Thursday.MicrosoftThe Artificial Intelligence division has officially launched its first two self-developed AI Models --MAI-Voice-1 Voice ModelThe new MAI-Voice-1 speech model requires only a single GPU to generate 1-minute-long audio in less than 1 second, while the MAI-1-preview model "lets the user advance to the next level," according to Microsoft. According to Microsoft, the new MAI-Voice-1 voice model generates one-minute audio in less than a second on a single GPU, while the MAI-1-preview model "gives users a sneak peek at what Copilot's future capabilities will look like".
Currently, Microsoft has applied MAI-Voice-1 to a number of functions: for example, in the "Copilot Daily" function, the AI presenter will use the model to broadcast the day's top news; at the same time, it can also generate podcast-style conversations to help users understand various topics.
Ordinary users can experience MAI-Voice-1 for themselves on the Copilot Labs platform: not only can they input what they want the AI to say, but they can also customize their voice tone and speaking style. In addition to this voice model, Microsoft is also launching the MAI-1-preview model, which utilizes approximately 15,000 NVIDIA H100 GPUs in its training process and is designed for users with specific needs, with the ability to follow commands and "provide practical responses to everyday inquiries".
Speaking on an episode of the Decoder podcast last year, Microsoft AI lead Mustafa Suleiman said that the company's internally developed AI models do not have enterprise application scenarios as a core direction, explaining, "My thinking is that we have to build a product that's a great experience for the consumer and is deeply optimized for its own application scenarios. In areas such as advertising business and consumer behavior data, we have a huge amount of highly predictive and useful data, so my focus is on building models that truly fit the 'consumer partner' positioning."
It is reported that Microsoft AI plans to apply the MAI-1-preview model to specific text usage scenarios for Copilot assistant (Copilot currently relies on OpenAI's large-scale language model). In the meantime, the model has begun public testing on the AI benchmarking platform LMArena.
1AI noted that the Microsoft AI team wrote in a blog post, "We have ambitious plans for future development. Next, we will not only continue to promote technological breakthroughs, but also believe that by integrating a series of specialized models for different user needs and application scenarios, we will unleash great value."