Step StarAnnouncing the launch of Step-1o's 100 billion parameter end-to-end in publicVoice big modelThe "allegedly"The first end-to-end speech macromodel with 100 billion parameters in China".
According to Step Star, the traditional voice model uses a cascade program, the user input voice information needs to be converted into text, and then converted into voice output, this process will not only reduce the transmission efficiency, but also in the process of loss of information, including emotions, resulting in the voice model to extend the slow response, the quality of the answer and the level of intelligence is insufficient, the expression of emotion is empty stereotypes of the shortcomings. However, an end-to-end voice solution thatIntegration of speech understanding and generation is possible, raising the upper limits of the model's IQ and EQ.
1AI learned from the official presentation thatStep-1o supports mixed forms of input and output such as voice and text.It can respond quickly and interrupt at any time, and it also understands and imitates in depth vocal characteristics such as timbre, rhythm, dialect, and individualized habits of spoken expression;
Step-1o is able to continuously improve the quality of its responses through self-learning and imitation, both by providing professional advice on problem solving and as a companion providing high emotional value. In addition, Step-1o has inherited the ability to create a large model of the Step-Star language.
Step-1o will be connected to the Leapfrog App terminal in the near future, Step-Star revealed.Provide real-time voice call service for users.