Index

Are you a composer, musician, sound designer, or artist who works primarily with audio samples? Do you like experimenting with new tools for sound design?

As part of my PhD at the Centre for Digital Music, QMUL, I’ve been exploring ways of using AI systems creatively in sample-based composition. I’ve made a few tools for my own composition practice (documentation soon come), but now I’m trying to see what that’s like for other composers.

I invite you to participate in our research study exploring how these tools can be used for creative manipulation of the timbre of audio samples.

We are offering compensation equivalent to £100¹ for participation.

Context + Aims

When audio is uploaded into an AI model, it is initially represented by an intermediate representation (sometimes called a ‘latent representation‘). Often, these representations are designed for automate-able process like text->music generation, or genre classification, which is a bit naff for actual composers and creatives (more on that in this post: https://www.noelhirst.net/research/Is%20depth%20real/index/)

We want to see if these representations can be exposed for manipulation to users in a way that leads to increased creative control for composition. We have designed a special sample-based tool to facilitate this exploration. Now we are looking at how people use and break these tools in their own practice, or what they would want from other tools.

I have a year left in the PhD, and I want to use all of it to make stuff for and with the sample-based community. This is a first step on that path.

Commitments

Participation requires that you:

1. Complete a pre-study questionnaire (25-mins)
2. Collect 5-20 of your own samples (either recorded by you or used by you), each under 40 seconds long.
3. Attend a single in-person session in our studios around 4hrs long (including several breaks), using our study tool on a machine that we have already set up for you. The session runs as follows:
   1. A short introduction, consent, and getting your samples onto the machine (15 mins)
   2. A practice go with the tool (20-25 mins)
   3. Three music-making tasks, (20-25 mins each), making 30-60 seconds of music in response to a prompt. You will use a different version of the tool each time. As you work, you jot down the time and a few words whenever something stands out. The tool also logs some basic interaction data, such as clicks, time in use, etc.
   4. A short questionnaire after each task (10 mins each)
   5. A closing questionnaire (15 mins)
   6. An interview (45 mins) in which we talk through the experience, particularly referencing of interest to you

Contact

If you’re interested, please write to the PhD conducting this research: Ashley Noel-Hirst.

This work is conducted at the Centre for Digital Music as part of the Communication Acoustics Lab