Topic
A new paper steers an audio diffusion model to a given melody without retraining it. The best case against getting excited is that trained, conditioned models do control better and sound better. Here is that case put fairly, and why we still think control is the feature that matters.
AI
An arXiv paper from Hsueh-Wei Fu, Yen-Tung Yeh and Yi-Hsuan Yang turns MIDI into guitar audio in two steps, first predicting editable pitch and loudness curves for each string, then rendering them to sound. The checkpoints are free to download, and listeners rated it above two comparison systems.
AIvia arXiv
A neural audio codec is a trained network that compresses audio into a short stream of discrete codes and rebuilds sound from them. EnCodec, SoundStream and DAC are the best-known examples, and their codes are the "audio tokens" many AI music generators predict.
Workflows
A diffusion model is an AI generator that starts from random noise and removes it step by step until sound (or a compressed code for sound) is left. Stable Audio Open is one documented example for music and sound.
Workflows
A new arXiv paper from Cristobal Andrade, Sebastian J. Schlecht, Ben Maman and Meinard Müller makes an existing audio diffusion model follow a given melody, using a pitch detector during sampling instead of training a new melody-conditioned model.
AIvia arXiv
One email a week: the music-tech and AI-in-music news that mattered, and what we think about it.