MixAssist

This dataset contains the complete audio recordings for the MixAssist project, as detailed in the paper "MixAssist: An Audio-Language Dataset for Co-Creative AI Assistance in Music Mixing."

This repository includes two main types of audio files:

• Raw Session Recordings: The complete, unprocessed audio recordings from the 7 co-creative mixing sessions between expert and amateur music producers. These recordings contain the full dialogue and simultaneous playback from the Digital Audio Workstation (DAW). They are provided to support research in areas like end-to-end conversational speech recognition or fine-grained interaction analysis.

• Processed Audio Segments: These are the music-only audio segments that have been extracted and temporally aligned with the conversational turns in the main MixAssist dataset. These segments represent the specific audio the participants were discussing at each point in the conversation.  You'll need to download these if you wish to train or fine-tune an audio language model on the MixAssist dataset, as the audio paths provided within the dataset refer back to these audio segments.

The full conversational MixAssist dataset is available on Hugging Face.

Please cite our paper if you use this dataset in your research.