Magenta RT offers flexible multimodal input, allowing users to either enter text cues (e.g. "light jazz") or upload audio clips as the basis for generation. Even more powerful is the ability to mix text and audio cues by adjusting the weighting, e.g. "50% jazz + 50% input audio". This hybrid input allows creators to precisely control the style and tempo of the music, providing a higher degree of control and creativity in music creation.
The result of the model generation is 48kHz stereo audio that can be used as an inspiration clip or to smoothly articulate the first 10 seconds of audio context (using cross-fade technology to reduce border distortion). The output supports real-time playback or saving as a WAV file.
This answer comes from the articleMagenta RealTime: an open source model for generating music in real timeThe
English 
简体中文 
日本語 
Deutsch 
Português do Brasil