AI Duet

[Curtis "Fjord" Hawthorne]

Google Creative Lab just released A.I. Duet, an interactive experiment which lets you play a music duet with the computer.  This is a great way to try out Magenta without having to install anything.

To learn more, check out our blog post: https://magenta.tensorflow.org/2017/02/16/ai-duet/

Or, just head straight to the demo: https://aiexperiments.withgoogle.com/ai-duet/view/

Enjoy!

-Fjord

[Vincent M.]

So COOL.

I need to catch up on the Magenta development sphere. I've been out for awhile. This is awesome though.

[Ben Riva]

I'm so excited to show this to my class!

[Circu Virtu]

Great stuff.

Tried your modified code that should support multiple models.
I just get the same interface as with the original and nothing to select any model.
Is that not build into the interface? What is the way to change between models then?

As well on PC/Windows I can't get any external Midi keyboard recognized and play.
Is there something I have to configure to get external Midi working in the browser?

Tnx

--cv

[James Weaver]

To change models, assign the desired model to the BUNDLE_NAME variable in predict.py:

BUNDLE_NAME = 'basic_rnn'

# BUNDLE_NAME = 'attention_rnn'

# BUNDLE_NAME = 'lookback_rnn'

Regards,

James Weaver

Twitter: http://twitter/JavaFXpert

Blogs: http://JavaFXpert.com http://CulturedEar.com

[James Weaver]

Great presentation, Fjord!  I'm giving a presentation entitled Composing Music in the Cloud [1] at the DevNexus developer conference in Atlanta this week [2], and would like to highlight Project Magenta as well as demonstrate A. I. Duet.  Would you mind if I showed a few of your slides in the process from your TEDx talk?  If you're OK with that, I will take screenshots from the video and give you attribution in the presentation and on the slides themselves.

[1] http://slides.com/javafxpert/counterpoint-composer-getting-started

[2] https://devnexus.com/s/devnexus2017/presentations/17305

Please advise,

James Weaver

On Thursday, February 16, 2017 at 3:05:10 PM UTC-5, Curtis Hawthorne wrote:

[Rafael Valle]

The problem being addressed by Magenta's AI duet has been successfully addressed by researchers in the past, including François Pachet continuator, IRCAM's OMax , George Lewi's Voayger, Edmund Camtpion's Natural Selection, etc...

François Pachet Continuator, for example, solves the AI duet problem using different algorithms but producing incredibly results, considerably more mature and interesting than Google's AI Duet. As a matter of fact, Pachet's continuator is believed to have passed the musical Turin test. Other examples are IRCAM's oMax, George Lewis' Voyager, etc...

Let's please honor research being done previously and situate current research appropriately, including a background review.

Here's a youlink to Pachet's continuator.

https://www.youtube.com/watch?v=ynPWOMzossI

http://francoispachet.fr/continuator/continuator.html

--
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.

[Paul Brossier]

Hi all,

Thank you Rafael for mentioning these references.

I was also planning to mention the ­­— quite amazing — Continuator on
this thread. More examples can be found online:

https://youtu.be/hh-tpWQlPcE
https://youtu.be/RIoE4ckZYgs
https://www.youtube.com/playlist?list=PLvoqwxjRRNfmLD2kNGnp9LlSeb5CzI3Kh

Looking forward to hearing the progress on this!

Best, Paul

> Google Creative Lab <https://github.com/googlecreativelab> just
> released A.I. Duet
> <https://aiexperiments.withgoogle.com/ai-duet/view>, an

> interactive experiment which lets you play a music duet with the
> computer. This is a great way to try out Magenta without having
> to install anything.
>
> To learn more, check out our blog
> post: https://magenta.tensorflow.org/2017/02/16/ai-duet/
> <https://magenta.tensorflow.org/2017/02/16/ai-duet/>
>
> Or, just head straight to the
> demo: https://aiexperiments.withgoogle.com/ai-duet/view/
> <https://aiexperiments.withgoogle.com/ai-duet/view/>
>
> Enjoy!
>
> -Fjord
>
> --

> Magenta project: magenta.tensorflow.org <http://magenta.tensorflow.org>

> To post to this group, send email to magenta...@tensorflow.org

> <mailto:magenta...@tensorflow.org>

> To unsubscribe from this group, send email to

> magenta-discu...@tensorflow.org
> <mailto:magenta-discuss%2Bunsu...@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
> <mailto:magenta-discu...@tensorflow.org>.

>
>
> --
> 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
> <mailto:magenta-discu...@tensorflow.org>.

[Bob Sturm]

Hi James. I appreciate your enthusiasm. Since you are interested in
_composing_ music with AI, you might find our work interesting ---
which is parallel to magenta, but goes beyond with respect to actually
bringing the applied machine learning back to music practice:

- https://github.com/IraKorshunova/folk-rnn
- https://soundcloud.com/sturmen-1
- https://www.youtube.com/channel/UCL-dVzMfnJAbIwKLF2-mnHw
- https://sites.google.com/site/c4dmconcerts1617/home/fixedmedia/brief
- https://highnoongmt.wordpress.com/2015/08/15/deep-learning-for-assisting-the-process-of-music-composition-part-4/
- http://www.eecs.qmul.ac.uk/~sturm/research/RNNIrishTrad/index.html
- https://csmc2016.wordpress.com/proceedings/

We are organising a concert in May of works composed with our system.
We are also organising a workshop next month with master traditional
musicians around our first volume of tunes
(https://highnoongmt.wordpress.com/2016/09/12/folk-rnn-session-tunes-volume-1-of-10/).

We have been working hart to get magenta models to generate material
that we find compelling enough for composition. Our folk-rnn system
(which is nothing but a "vanilla" LSTM) works with transcription
tokens instead of quaver-resolution MIDI events. Surprisingly, that
seems to make a major difference.

Our research is not unique, of course. There is the fantastic work of
Pachet's group and their flow machines: http://www.flow-machines.com/
Pachet's Continuator program did what the magenta AI duet system does,
but in 2000 using completely different approaches.
And there's the fantastic work of Tom Collins:
http://www.tomcollinsresearch.net/
The work of David Cope cannot be overlooked:
http://artsites.ucsc.edu/faculty/cope/experiments.htm.

Many others have come long before magenta too, e.g., the Illiac Suite
in 1957 (https://en.wikipedia.org/wiki/Illiac_Suite), and Sister
Padberg's "Canon and Free Fugue" in 1964.

Three awesome things about magenta though: its a whole team of really
smart people working to bring these techniques mainstream; the project
is supported by Google!; they have a strong press outfit.

Cheers.
-Bob.

> --
> 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.



--
Bob L. Sturm, Lecturer in Digital Media
School of Electronic Engineering and Computer Science
Queen Mary University of London
https://highnoongmt.wordpress.com

[Lasconic]

Three awesome things about magenta though: its a whole team of really
smart people working to bring these techniques mainstream; the project
is supported by Google!; they have a strong press outfit.

Let me add a fourth important one: The code is available and under an open license

[Bob Sturm]

Good point, but that's not unique to magenta. Lots of other programs
are open and available, e.g., Nick Collins' work
(https://composerprogrammer.com/code.html). And folk-rnn has been
available since 2015. There are many other researchers and composers
who have shared their code.

[Rafael Valle]

Code and publications.

On Sun, Feb 19, 2017 at 11:57 AM, Bob Sturm <bobl...@gmail.com> wrote:

Good point, but that's not unique to magenta. Lots of other programs
are open and available, e.g., Nick Collins' work
(https://composerprogrammer.com/code.html). And folk-rnn has been
available since 2015. There are many other researchers and composers
who have shared their code.

On Sun, Feb 19, 2017 at 7:20 PM, Lasconic <lasc...@gmail.com> wrote:
>>
>> Three awesome things about magenta though: its a whole team of really
>> smart people working to bring these techniques mainstream; the project
>> is supported by Google!; they have a strong press outfit.
>
>
> Let me add a fourth important one: The code is available and under an open
> license
>
> --
> 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.

--
Bob L. Sturm, Lecturer in Digital Media
School of Electronic Engineering and Computer Science
Queen Mary University of London
https://highnoongmt.wordpress.com

--
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.

[Circu Virtu]

thanks a lot.
Unfortunately one has to restart the server to make it work.
It would be great to have just soem buttons on the client web interface to switch.
Maybe even the full interface from the NIPS demo to change other parameters as well.

--cv

[Adam Roberts]

Thanks for all of the additional citations!

We certainly do not work in a vacuum nor do we desire to. Our hope is that others will use our library to build better models that can be shared back with the community. What we have included in our repository thus far are mostly basic models meant to demonstrate a few fundamental techniques for melody generation. We want to continue to add (with your help!) additional methods inspired by external work (as we have done for DeepBach with polyphonic_rnn), while also sharing anything new and compelling that we come up with.  Finally, we hope that by building interfaces with simple APIs, you will be able to easily evaluate and "play with" your models using anything from a web browser to a professional-level production environment.

If you'd like to contribute, please have a look at the Issues tracker on our GitHub to pick something out or add your own ideas.

Thanks again for all of your feedback on our work!

-Adam

--

[Curtis "Fjord" Hawthorne]

Hey James,

Glad you liked the presentation! Feel free to screenshot some of the slides from the video. Hope your presentation goes well!

-Fjord

--

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.

[maninder singh]

Hello to all i am having some problem in training the dataset can anyone help me out

 thanks in advance

On Thu, Mar 2, 2017 at 11:40 PM, 'Curtis "Fjord" Hawthorne' via Magenta Discuss <magenta...@tensorflow.org> wrote:

Hey James,

Glad you liked the presentation! Feel free to screenshot some of the slides from the video. Hope your presentation goes well!

-Fjord

On Sun, Feb 19, 2017 at 5:49 AM James Weaver <james.l...@gmail.com> wrote:

Great presentation, Fjord!  I'm giving a presentation entitled Composing Music in the Cloud [1] at the DevNexus developer conference in Atlanta this week [2], and would like to highlight Project Magenta as well as demonstrate A. I. Duet.  Would you mind if I showed a few of your slides in the process from your TEDx talk?  If you're OK with that, I will take screenshots from the video and give you attribution in the presentation and on the slides themselves.

[1] http://slides.com/javafxpert/counterpoint-composer-getting-started

[2] https://devnexus.com/s/devnexus2017/presentations/17305

Please advise,

James Weaver

On Thursday, February 16, 2017 at 3:05:10 PM UTC-5, Curtis Hawthorne wrote:

Google Creative Lab just released A.I. Duet, an interactive experiment which lets you play a music duet with the computer.  This is a great way to try out Magenta without having to install anything.

To learn more, check out our blog post: https://magenta.tensorflow.org/2017/02/16/ai-duet/

Or, just head straight to the demo: https://aiexperiments.withgoogle.com/ai-duet/view/

Enjoy!

-Fjord

--
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.

--
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.