MusicVAE: Creating a palette for musical scores with machine learning.

[Adam Roberts]

We just posted on our blog about our newest work MusicVAE (https://g.co/magenta/music-vae).

Much like SketchRNN and NSynth, it learns a latent space, but in this case it is a space of musical scores (note sequences) instead of sketches or instrument timbres. This enables all sorts of creative applications which we explore in the post.

We also introduce a JavaScript library for using the model to create music in the browser along with 3 apps to demonstrate its capabilties.

We are really excited about this work and we think you will be too. Please have a look and share any music or applications you create with it!

-Adam

[Rafael Valle]

thank you for putting this together. vector arithmetic and long-term structures is something that i've been wanting to see in music for a long time.

please update the arxiv paper and related publications such that the related work section covers previous work, including tools.

for tools one can look at open music and bach for max msp. 

for publications one can look at open music composers book or papers from icmc, ismir and related conferences.

similarly to what happened with NSynth, it is our responsibility to situate MusicVAE's contribution to a practice that exist for at least two decades.

--
Magenta project: magenta.tensorflow.org
To post to this group, send email to magenta...@tensorflow.org
To unsubscribe from this group, send email to magenta-discuss+unsubscribe@tensorflow.org
---
You received this message because you are subscribed to the Google Groups "Magenta Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to magenta-discuss+unsubscribe@tensorflow.org.

[Adam Roberts]

Thanks!  We agree that there has been prior work on creative tools for composing music, and we cite many of them in our paper on the interactive tool itself: http://ceur-ws.org/Vol-2068/milc7.pdf

However, the contribution of the MusicVAE paper is the hierarchical recurrent VAE structure, so this work is relevant only to the application we demonstrate, not the core message of that particular paper.

On Thu, Mar 15, 2018 at 10:40 PM Rafael Valle <jrafae...@gmail.com> wrote:

thank you for putting this together. vector arithmetic and long-term structures is something that i've been wanting to see in music for a long time.

please update the arxiv paper and related publications such that the related work section covers previous work, including tools.

for tools one can look at open music and bach for max msp. 

for publications one can look at open music composers book or papers from icmc, ismir and related conferences.

similarly to what happened with NSynth, it is our responsibility to situate MusicVAE's contribution to a practice that exist for at least two decades.

On Thu, Mar 15, 2018 at 10:01 AM, 'Adam Roberts' via Magenta Discuss <magenta...@tensorflow.org> wrote:

We just posted on our blog about our newest work MusicVAE (https://g.co/magenta/music-vae).

Much like SketchRNN and NSynth, it learns a latent space, but in this case it is a space of musical scores (note sequences) instead of sketches or instrument timbres. This enables all sorts of creative applications which we explore in the post.

We also introduce a JavaScript library for using the model to create music in the browser along with 3 apps to demonstrate its capabilties.

We are really excited about this work and we think you will be too. Please have a look and share any music or applications you create with it!

-Adam

--
Magenta project: magenta.tensorflow.org
To post to this group, send email to magenta...@tensorflow.org

To unsubscribe from this group, send email to magenta-discu...@tensorflow.org

---
You received this message because you are subscribed to the Google Groups "Magenta Discuss" group.

To unsubscribe from this group and stop receiving emails from it, send an email to magenta-discu...@tensorflow.org.