ML Community Day is November 9! Join us for updates from TensorFlow, JAX, and more Learn more

Generate music with an RNN

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

This tutorial shows you how to generate musical notes using a simple RNN. You will train a model using a collection of piano MIDI files from the MAESTRO dataset. Given a sequence of notes, your model will learn to predict the next note in the sequence. You can generate a longer sequences of notes by calling the model repeatedly.

This tutorial contains complete code to parse and create MIDI files. You can learn more about how RNNs work by visiting Text generation with an RNN.

Setup

This tutorial uses the pretty_midi library to create and parse MIDI files, and pyfluidsynth for generating audio playback in Colab.

sudo apt install -y fluidsynth
The following packages were automatically installed and are no longer required:
  linux-gcp-5.4-headers-5.4.0-1040 linux-gcp-5.4-headers-5.4.0-1043
  linux-gcp-5.4-headers-5.4.0-1044
Use 'sudo apt autoremove' to remove them.
The following additional packages will be installed:
  fluid-soundfont-gm libasyncns0 libdouble-conversion1 libevdev2 libflac8
  libfluidsynth1 libgudev-1.0-0 libinput-bin libinput10 libjack-jackd2-0
  libmtdev1 libogg0 libpulse0 libqt5core5a libqt5dbus5 libqt5gui5
  libqt5network5 libqt5svg5 libqt5widgets5 libqt5x11extras5 libsamplerate0
  libsndfile1 libvorbis0a libvorbisenc2 libwacom-bin libwacom-common libwacom2
  libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-randr0
  libxcb-render-util0 libxcb-shape0 libxcb-util1 libxcb-xinerama0 libxcb-xkb1
  libxkbcommon-x11-0 qsynth qt5-gtk-platformtheme qttranslations5-l10n
Suggested packages:
  fluid-soundfont-gs timidity jackd2 pulseaudio qt5-image-formats-plugins
  qtwayland5 jackd
The following NEW packages will be installed:
  fluid-soundfont-gm fluidsynth libasyncns0 libdouble-conversion1 libevdev2
  libflac8 libfluidsynth1 libgudev-1.0-0 libinput-bin libinput10
  libjack-jackd2-0 libmtdev1 libogg0 libpulse0 libqt5core5a libqt5dbus5
  libqt5gui5 libqt5network5 libqt5svg5 libqt5widgets5 libqt5x11extras5
  libsamplerate0 libsndfile1 libvorbis0a libvorbisenc2 libwacom-bin
  libwacom-common libwacom2 libxcb-icccm4 libxcb-image0 libxcb-keysyms1
  libxcb-randr0 libxcb-render-util0 libxcb-shape0 libxcb-util1
  libxcb-xinerama0 libxcb-xkb1 libxkbcommon-x11-0 qsynth qt5-gtk-platformtheme
  qttranslations5-l10n
0 upgraded, 41 newly installed, 0 to remove and 115 not upgraded.
Need to get 132 MB of archives.
After this operation, 198 MB of additional disk space will be used.
Get:1 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libogg0 amd64 1.3.2-1 [17.2 kB]
Get:2 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libdouble-conversion1 amd64 2.0.1-4ubuntu1 [33.0 kB]
Get:3 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libqt5core5a amd64 5.9.5+dfsg-0ubuntu2.5 [2036 kB]
Get:4 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libevdev2 amd64 1.5.8+dfsg-1ubuntu0.1 [28.9 kB]
Get:5 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libmtdev1 amd64 1.1.5-1ubuntu3 [13.8 kB]
Get:6 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libgudev-1.0-0 amd64 1:232-2 [13.6 kB]
Get:7 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libwacom-common all 0.29-1 [36.9 kB]
Get:8 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libwacom2 amd64 0.29-1 [17.7 kB]
Get:9 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libinput-bin amd64 1.10.4-1ubuntu0.18.04.2 [11.2 kB]
Get:10 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libinput10 amd64 1.10.4-1ubuntu0.18.04.2 [86.2 kB]
Get:11 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libqt5dbus5 amd64 5.9.5+dfsg-0ubuntu2.5 [195 kB]
Get:12 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libqt5network5 amd64 5.9.5+dfsg-0ubuntu2.5 [635 kB]
Get:13 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libxcb-icccm4 amd64 0.4.1-1ubuntu1 [10.4 kB]
Get:14 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libxcb-util1 amd64 0.4.0-0ubuntu3 [11.2 kB]
Get:15 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libxcb-image0 amd64 0.4.0-1build1 [12.3 kB]
Get:16 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libxcb-keysyms1 amd64 0.4.0-1 [8406 B]
Get:17 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libxcb-randr0 amd64 1.13-2~ubuntu18.04 [16.4 kB]
Get:18 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libxcb-render-util0 amd64 0.3.9-1 [9638 B]
Get:19 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libxcb-shape0 amd64 1.13-2~ubuntu18.04 [5972 B]
Get:20 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libxcb-xinerama0 amd64 1.13-2~ubuntu18.04 [5264 B]
Get:21 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libxcb-xkb1 amd64 1.13-2~ubuntu18.04 [30.1 kB]
Get:22 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libxkbcommon-x11-0 amd64 0.8.2-1~ubuntu18.04.1 [13.4 kB]
Get:23 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libqt5gui5 amd64 5.9.5+dfsg-0ubuntu2.5 [2568 kB]
Get:24 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libqt5widgets5 amd64 5.9.5+dfsg-0ubuntu2.5 [2203 kB]
Get:25 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libqt5svg5 amd64 5.9.5-0ubuntu1 [128 kB]
Get:26 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 fluid-soundfont-gm all 3.1-5.1 [119 MB]
Get:27 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libsamplerate0 amd64 0.1.9-1 [938 kB]
Get:28 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libjack-jackd2-0 amd64 1.9.12~dfsg-2 [263 kB]
Get:29 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libasyncns0 amd64 0.8-6 [12.1 kB]
Get:30 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libflac8 amd64 1.3.2-1 [213 kB]
Get:31 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libvorbis0a amd64 1.3.5-4.2 [86.4 kB]
Get:32 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libvorbisenc2 amd64 1.3.5-4.2 [70.7 kB]
Get:33 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libsndfile1 amd64 1.0.28-4ubuntu0.18.04.2 [170 kB]
Get:34 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 libpulse0 amd64 1:11.1-1ubuntu7.11 [266 kB]
Get:35 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 libfluidsynth1 amd64 1.1.9-1 [137 kB]
Get:36 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 fluidsynth amd64 1.1.9-1 [20.7 kB]
Get:37 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 libqt5x11extras5 amd64 5.9.5-0ubuntu1 [8596 B]
Get:38 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libwacom-bin amd64 0.29-1 [4712 B]
Get:39 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 qsynth amd64 0.5.0-2 [191 kB]
Get:40 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic-updates/main amd64 qt5-gtk-platformtheme amd64 5.9.5+dfsg-0ubuntu2.5 [117 kB]
Get:41 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 qttranslations5-l10n all 5.9.5-0ubuntu1 [1485 kB]
Fetched 132 MB in 4s (30.1 MB/s)
Extracting templates from packages: 100%

7[0;23r8[1ASelecting previously unselected package libogg0:amd64.
(Reading database ... 246391 files and directories currently installed.)
Preparing to unpack .../00-libogg0_1.3.2-1_amd64.deb ...
7[24;0fProgress: [  0%] [..........................................................] 8Unpacking libogg0:amd64 (1.3.2-1) ...
7[24;0fProgress: [  1%] [..........................................................] 8Selecting previously unselected package libdouble-conversion1:amd64.
Preparing to unpack .../01-libdouble-conversion1_2.0.1-4ubuntu1_amd64.deb ...
Unpacking libdouble-conversion1:amd64 (2.0.1-4ubuntu1) ...
7[24;0fProgress: [  2%] [#.........................................................] 8Selecting previously unselected package libqt5core5a:amd64.
Preparing to unpack .../02-libqt5core5a_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
7[24;0fProgress: [  3%] [#.........................................................] 8Unpacking libqt5core5a:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [  4%] [##........................................................] 8Selecting previously unselected package libevdev2:amd64.
Preparing to unpack .../03-libevdev2_1.5.8+dfsg-1ubuntu0.1_amd64.deb ...
Unpacking libevdev2:amd64 (1.5.8+dfsg-1ubuntu0.1) ...
7[24;0fProgress: [  5%] [###.......................................................] 8Selecting previously unselected package libmtdev1:amd64.
Preparing to unpack .../04-libmtdev1_1.1.5-1ubuntu3_amd64.deb ...
7[24;0fProgress: [  6%] [###.......................................................] 8Unpacking libmtdev1:amd64 (1.1.5-1ubuntu3) ...
7[24;0fProgress: [  7%] [####......................................................] 8Selecting previously unselected package libgudev-1.0-0:amd64.
Preparing to unpack .../05-libgudev-1.0-0_1%3a232-2_amd64.deb ...
Unpacking libgudev-1.0-0:amd64 (1:232-2) ...
7[24;0fProgress: [  8%] [####......................................................] 8Selecting previously unselected package libwacom-common.
Preparing to unpack .../06-libwacom-common_0.29-1_all.deb ...
7[24;0fProgress: [  9%] [#####.....................................................] 8Unpacking libwacom-common (0.29-1) ...
7[24;0fProgress: [ 10%] [#####.....................................................] 8Selecting previously unselected package libwacom2:amd64.
Preparing to unpack .../07-libwacom2_0.29-1_amd64.deb ...
Unpacking libwacom2:amd64 (0.29-1) ...
7[24;0fProgress: [ 11%] [######....................................................] 8Selecting previously unselected package libinput-bin.
Preparing to unpack .../08-libinput-bin_1.10.4-1ubuntu0.18.04.2_amd64.deb ...
7[24;0fProgress: [ 12%] [#######...................................................] 8Unpacking libinput-bin (1.10.4-1ubuntu0.18.04.2) ...
7[24;0fProgress: [ 13%] [#######...................................................] 8Selecting previously unselected package libinput10:amd64.
Preparing to unpack .../09-libinput10_1.10.4-1ubuntu0.18.04.2_amd64.deb ...
Unpacking libinput10:amd64 (1.10.4-1ubuntu0.18.04.2) ...
7[24;0fProgress: [ 14%] [########..................................................] 8Selecting previously unselected package libqt5dbus5:amd64.
Preparing to unpack .../10-libqt5dbus5_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
7[24;0fProgress: [ 15%] [########..................................................] 8Unpacking libqt5dbus5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 16%] [#########.................................................] 8Selecting previously unselected package libqt5network5:amd64.
Preparing to unpack .../11-libqt5network5_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
Unpacking libqt5network5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 17%] [##########................................................] 8Selecting previously unselected package libxcb-icccm4:amd64.
Preparing to unpack .../12-libxcb-icccm4_0.4.1-1ubuntu1_amd64.deb ...
Unpacking libxcb-icccm4:amd64 (0.4.1-1ubuntu1) ...
7[24;0fProgress: [ 18%] [##########................................................] 8Selecting previously unselected package libxcb-util1:amd64.
Preparing to unpack .../13-libxcb-util1_0.4.0-0ubuntu3_amd64.deb ...
7[24;0fProgress: [ 19%] [###########...............................................] 8Unpacking libxcb-util1:amd64 (0.4.0-0ubuntu3) ...
7[24;0fProgress: [ 20%] [###########...............................................] 8Selecting previously unselected package libxcb-image0:amd64.
Preparing to unpack .../14-libxcb-image0_0.4.0-1build1_amd64.deb ...
Unpacking libxcb-image0:amd64 (0.4.0-1build1) ...
7[24;0fProgress: [ 21%] [############..............................................] 8Selecting previously unselected package libxcb-keysyms1:amd64.
Preparing to unpack .../15-libxcb-keysyms1_0.4.0-1_amd64.deb ...
7[24;0fProgress: [ 22%] [############..............................................] 8Unpacking libxcb-keysyms1:amd64 (0.4.0-1) ...
7[24;0fProgress: [ 23%] [#############.............................................] 8Selecting previously unselected package libxcb-randr0:amd64.
Preparing to unpack .../16-libxcb-randr0_1.13-2~ubuntu18.04_amd64.deb ...
Unpacking libxcb-randr0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 24%] [##############............................................] 8Selecting previously unselected package libxcb-render-util0:amd64.
Preparing to unpack .../17-libxcb-render-util0_0.3.9-1_amd64.deb ...
7[24;0fProgress: [ 25%] [##############............................................] 8Unpacking libxcb-render-util0:amd64 (0.3.9-1) ...
7[24;0fProgress: [ 26%] [###############...........................................] 8Selecting previously unselected package libxcb-shape0:amd64.
Preparing to unpack .../18-libxcb-shape0_1.13-2~ubuntu18.04_amd64.deb ...
Unpacking libxcb-shape0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 27%] [###############...........................................] 8Selecting previously unselected package libxcb-xinerama0:amd64.
Preparing to unpack .../19-libxcb-xinerama0_1.13-2~ubuntu18.04_amd64.deb ...
7[24;0fProgress: [ 28%] [################..........................................] 8Unpacking libxcb-xinerama0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 29%] [################..........................................] 8Selecting previously unselected package libxcb-xkb1:amd64.
Preparing to unpack .../20-libxcb-xkb1_1.13-2~ubuntu18.04_amd64.deb ...
Unpacking libxcb-xkb1:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 30%] [#################.........................................] 8Selecting previously unselected package libxkbcommon-x11-0:amd64.
Preparing to unpack .../21-libxkbcommon-x11-0_0.8.2-1~ubuntu18.04.1_amd64.deb ...
7[24;0fProgress: [ 31%] [##################........................................] 8Unpacking libxkbcommon-x11-0:amd64 (0.8.2-1~ubuntu18.04.1) ...
7[24;0fProgress: [ 32%] [##################........................................] 8Selecting previously unselected package libqt5gui5:amd64.
Preparing to unpack .../22-libqt5gui5_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
Unpacking libqt5gui5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 33%] [###################.......................................] 8Selecting previously unselected package libqt5widgets5:amd64.
Preparing to unpack .../23-libqt5widgets5_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
Unpacking libqt5widgets5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 34%] [###################.......................................] 8Selecting previously unselected package libqt5svg5:amd64.
Preparing to unpack .../24-libqt5svg5_5.9.5-0ubuntu1_amd64.deb ...
7[24;0fProgress: [ 35%] [####################......................................] 8Unpacking libqt5svg5:amd64 (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 36%] [#####################.....................................] 8Selecting previously unselected package fluid-soundfont-gm.
Preparing to unpack .../25-fluid-soundfont-gm_3.1-5.1_all.deb ...
Unpacking fluid-soundfont-gm (3.1-5.1) ...
7[24;0fProgress: [ 37%] [#####################.....................................] 8Selecting previously unselected package libsamplerate0:amd64.
Preparing to unpack .../26-libsamplerate0_0.1.9-1_amd64.deb ...
7[24;0fProgress: [ 38%] [######################....................................] 8Unpacking libsamplerate0:amd64 (0.1.9-1) ...
7[24;0fProgress: [ 39%] [######################....................................] 8Selecting previously unselected package libjack-jackd2-0:amd64.
Preparing to unpack .../27-libjack-jackd2-0_1.9.12~dfsg-2_amd64.deb ...
Unpacking libjack-jackd2-0:amd64 (1.9.12~dfsg-2) ...
7[24;0fProgress: [ 40%] [#######################...................................] 8Selecting previously unselected package libasyncns0:amd64.
Preparing to unpack .../28-libasyncns0_0.8-6_amd64.deb ...
7[24;0fProgress: [ 41%] [#######################...................................] 8Unpacking libasyncns0:amd64 (0.8-6) ...
7[24;0fProgress: [ 42%] [########################..................................] 8Selecting previously unselected package libflac8:amd64.
Preparing to unpack .../29-libflac8_1.3.2-1_amd64.deb ...
Unpacking libflac8:amd64 (1.3.2-1) ...
7[24;0fProgress: [ 43%] [#########################.................................] 8Selecting previously unselected package libvorbis0a:amd64.
Preparing to unpack .../30-libvorbis0a_1.3.5-4.2_amd64.deb ...
7[24;0fProgress: [ 44%] [#########################.................................] 8Unpacking libvorbis0a:amd64 (1.3.5-4.2) ...
7[24;0fProgress: [ 45%] [##########################................................] 8Selecting previously unselected package libvorbisenc2:amd64.
Preparing to unpack .../31-libvorbisenc2_1.3.5-4.2_amd64.deb ...
Unpacking libvorbisenc2:amd64 (1.3.5-4.2) ...
7[24;0fProgress: [ 46%] [##########################................................] 8Selecting previously unselected package libsndfile1:amd64.
Preparing to unpack .../32-libsndfile1_1.0.28-4ubuntu0.18.04.2_amd64.deb ...
7[24;0fProgress: [ 47%] [###########################...............................] 8Unpacking libsndfile1:amd64 (1.0.28-4ubuntu0.18.04.2) ...
7[24;0fProgress: [ 48%] [###########################...............................] 8Selecting previously unselected package libpulse0:amd64.
Preparing to unpack .../33-libpulse0_1%3a11.1-1ubuntu7.11_amd64.deb ...
Unpacking libpulse0:amd64 (1:11.1-1ubuntu7.11) ...
7[24;0fProgress: [ 49%] [############################..............................] 8Selecting previously unselected package libfluidsynth1:amd64.
Preparing to unpack .../34-libfluidsynth1_1.1.9-1_amd64.deb ...
7[24;0fProgress: [ 50%] [#############################.............................] 8Unpacking libfluidsynth1:amd64 (1.1.9-1) ...
Selecting previously unselected package fluidsynth.
Preparing to unpack .../35-fluidsynth_1.1.9-1_amd64.deb ...
7[24;0fProgress: [ 51%] [#############################.............................] 8Unpacking fluidsynth (1.1.9-1) ...
7[24;0fProgress: [ 52%] [##############################............................] 8Selecting previously unselected package libqt5x11extras5:amd64.
Preparing to unpack .../36-libqt5x11extras5_5.9.5-0ubuntu1_amd64.deb ...
Unpacking libqt5x11extras5:amd64 (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 53%] [##############################............................] 8Selecting previously unselected package libwacom-bin.
Preparing to unpack .../37-libwacom-bin_0.29-1_amd64.deb ...
7[24;0fProgress: [ 54%] [###############################...........................] 8Unpacking libwacom-bin (0.29-1) ...
7[24;0fProgress: [ 55%] [################################..........................] 8Selecting previously unselected package qsynth.
Preparing to unpack .../38-qsynth_0.5.0-2_amd64.deb ...
Unpacking qsynth (0.5.0-2) ...
7[24;0fProgress: [ 56%] [################################..........................] 8Selecting previously unselected package qt5-gtk-platformtheme:amd64.
Preparing to unpack .../39-qt5-gtk-platformtheme_5.9.5+dfsg-0ubuntu2.5_amd64.deb ...
7[24;0fProgress: [ 57%] [#################################.........................] 8Unpacking qt5-gtk-platformtheme:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 58%] [#################################.........................] 8Selecting previously unselected package qttranslations5-l10n.
Preparing to unpack .../40-qttranslations5-l10n_5.9.5-0ubuntu1_all.deb ...
Unpacking qttranslations5-l10n (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 59%] [##################################........................] 8Setting up libxcb-xinerama0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 60%] [##################################........................] 8Setting up libxcb-render-util0:amd64 (0.3.9-1) ...
7[24;0fProgress: [ 61%] [###################################.......................] 8Setting up libxcb-randr0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 62%] [####################################......................] 8Setting up libxcb-icccm4:amd64 (0.4.1-1ubuntu1) ...
7[24;0fProgress: [ 63%] [####################################......................] 8Setting up libasyncns0:amd64 (0.8-6) ...
7[24;0fProgress: [ 64%] [#####################################.....................] 8Setting up libwacom-common (0.29-1) ...
7[24;0fProgress: [ 65%] [#####################################.....................] 8Setting up libdouble-conversion1:amd64 (2.0.1-4ubuntu1) ...
7[24;0fProgress: [ 66%] [######################################....................] 8Setting up libevdev2:amd64 (1.5.8+dfsg-1ubuntu0.1) ...
7[24;0fProgress: [ 67%] [#######################################...................] 8Setting up fluid-soundfont-gm (3.1-5.1) ...
7[24;0fProgress: [ 68%] [#######################################...................] 8Setting up libxcb-util1:amd64 (0.4.0-0ubuntu3) ...
7[24;0fProgress: [ 69%] [########################################..................] 8Setting up libogg0:amd64 (1.3.2-1) ...
7[24;0fProgress: [ 70%] [########################################..................] 8Setting up qttranslations5-l10n (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 71%] [#########################################.................] 8Setting up libmtdev1:amd64 (1.1.5-1ubuntu3) ...
7[24;0fProgress: [ 72%] [#########################################.................] 8Setting up libxcb-shape0:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 73%] [##########################################................] 8Setting up libgudev-1.0-0:amd64 (1:232-2) ...
7[24;0fProgress: [ 74%] [###########################################...............] 8Setting up libxcb-keysyms1:amd64 (0.4.0-1) ...
7[24;0fProgress: [ 75%] [###########################################...............] 8Setting up libsamplerate0:amd64 (0.1.9-1) ...
7[24;0fProgress: [ 76%] [############################################..............] 8Setting up libvorbis0a:amd64 (1.3.5-4.2) ...
7[24;0fProgress: [ 77%] [############################################..............] 8Setting up libxcb-xkb1:amd64 (1.13-2~ubuntu18.04) ...
7[24;0fProgress: [ 78%] [#############################################.............] 8Setting up libqt5core5a:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 79%] [#############################################.............] 8Setting up libqt5dbus5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 80%] [##############################################............] 8Setting up libqt5network5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 81%] [###############################################...........] 8Setting up libwacom2:amd64 (0.29-1) ...
7[24;0fProgress: [ 82%] [###############################################...........] 8Setting up libxcb-image0:amd64 (0.4.0-1build1) ...
7[24;0fProgress: [ 83%] [################################################..........] 8Setting up libflac8:amd64 (1.3.2-1) ...
Setting up libinput-bin (1.10.4-1ubuntu0.18.04.2) ...
7[24;0fProgress: [ 84%] [################################################..........] 8Setting up libxkbcommon-x11-0:amd64 (0.8.2-1~ubuntu18.04.1) ...
7[24;0fProgress: [ 85%] [#################################################.........] 8Setting up libwacom-bin (0.29-1) ...
7[24;0fProgress: [ 86%] [##################################################........] 8Setting up libjack-jackd2-0:amd64 (1.9.12~dfsg-2) ...
7[24;0fProgress: [ 87%] [##################################################........] 8Setting up libvorbisenc2:amd64 (1.3.5-4.2) ...
7[24;0fProgress: [ 88%] [###################################################.......] 8Setting up libinput10:amd64 (1.10.4-1ubuntu0.18.04.2) ...
7[24;0fProgress: [ 89%] [###################################################.......] 8Setting up libsndfile1:amd64 (1.0.28-4ubuntu0.18.04.2) ...
7[24;0fProgress: [ 90%] [####################################################......] 8Setting up libqt5gui5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 91%] [####################################################......] 8Setting up qt5-gtk-platformtheme:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 92%] [#####################################################.....] 8Setting up libqt5x11extras5:amd64 (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 93%] [######################################################....] 8Setting up libqt5widgets5:amd64 (5.9.5+dfsg-0ubuntu2.5) ...
7[24;0fProgress: [ 94%] [######################################################....] 8Setting up libpulse0:amd64 (1:11.1-1ubuntu7.11) ...
7[24;0fProgress: [ 95%] [#######################################################...] 8Setting up libqt5svg5:amd64 (5.9.5-0ubuntu1) ...
7[24;0fProgress: [ 96%] [#######################################################...] 8Setting up libfluidsynth1:amd64 (1.1.9-1) ...
7[24;0fProgress: [ 97%] [########################################################..] 8Setting up fluidsynth (1.1.9-1) ...
7[24;0fProgress: [ 98%] [########################################################..] 8Setting up qsynth (0.5.0-2) ...
7[24;0fProgress: [ 99%] [#########################################################.] 8Processing triggers for hicolor-icon-theme (0.17-2) ...
Processing triggers for mime-support (3.60ubuntu1) ...
Processing triggers for libc-bin (2.27-3ubuntu1.2) ...
Processing triggers for udev (237-3ubuntu10.50) ...
Processing triggers for man-db (2.8.3-2ubuntu0.1) ...

7[0;24r8[1A[J
pip install --upgrade pyfluidsynth
pip install pretty_midi
import collections
import datetime
import fluidsynth
import glob
import numpy as np
import pathlib
import pandas as pd
import pretty_midi
import seaborn as sns
import tensorflow as tf

from IPython import display
from matplotlib import pyplot as plt
from typing import Dict, List, Optional, Sequence, Tuple
seed = 42
tf.random.set_seed(seed)
np.random.seed(seed)

# Sampling rate for audio playback
_SAMPLING_RATE = 16000

Download the Maestro dataset

data_dir = pathlib.Path('data/maestro-v2.0.0')
if not data_dir.exists():
  tf.keras.utils.get_file(
      'maestro-v2.0.0-midi.zip',
      origin='https://storage.googleapis.com/magentadata/datasets/maestro/v2.0.0/maestro-v2.0.0-midi.zip',
      extract=True,
      cache_dir='.', cache_subdir='data',
  )
Downloading data from https://storage.googleapis.com/magentadata/datasets/maestro/v2.0.0/maestro-v2.0.0-midi.zip
59244544/59243107 [==============================] - 1s 0us/step
59252736/59243107 [==============================] - 1s 0us/step

The dataset contains about 1,200 MIDI files.

filenames = glob.glob(str(data_dir/'**/*.mid*'))
print('Number of files:', len(filenames))
Number of files: 1282

Process a MIDI file

First, use pretty_midi to parse a single MIDI file and inspect the format of the notes. If you would like to download the MIDI file below to play on your computer, you can do so in colab by writing files.download(sample_file).

sample_file = filenames[1]
print(sample_file)
data/maestro-v2.0.0/2015/MIDI-Unprocessed_R1_D1-1-8_mid--AUDIO-from_mp3_02_R1_2015_wav--6.midi

Generate a PrettyMIDI object for the sample MIDI file.

pm = pretty_midi.PrettyMIDI(sample_file)

Play the sample file. The playback widget may take several seconds to load.

def display_audio(pm: pretty_midi.PrettyMIDI, seconds=30):
  waveform = pm.fluidsynth(fs=_SAMPLING_RATE)
  # Take a sample of the generated waveform to mitigate kernel resets
  waveform_short = waveform[:seconds*_SAMPLING_RATE]
  return display.Audio(waveform_short, rate=_SAMPLING_RATE)
display_audio(pm)

Do some inspection on the MIDI file. What kinds of instruments are used?

print('Number of instruments:', len(pm.instruments))
instrument = pm.instruments[0]
instrument_name = pretty_midi.program_to_instrument_name(instrument.program)
print('Instrument name:', instrument_name)
Number of instruments: 1
Instrument name: Acoustic Grand Piano

Extract notes

for i, note in enumerate(instrument.notes[:10]):
  note_name = pretty_midi.note_number_to_name(note.pitch)
  duration = note.end - note.start
  print(f'{i}: pitch={note.pitch}, note_name={note_name},'
        f' duration={duration:.4f}')
0: pitch=63, note_name=D#4, duration=0.0760
1: pitch=75, note_name=D#5, duration=0.0687
2: pitch=75, note_name=D#5, duration=0.0615
3: pitch=63, note_name=D#4, duration=0.0688
4: pitch=75, note_name=D#5, duration=0.0448
5: pitch=63, note_name=D#4, duration=0.0573
6: pitch=87, note_name=D#6, duration=0.0302
7: pitch=99, note_name=D#7, duration=0.0260
8: pitch=87, note_name=D#6, duration=0.0271
9: pitch=99, note_name=D#7, duration=0.0260

You will use three variables to represent a note when training the model: pitch, step and duration. The pitch is the perceptual quality of the sound as a MIDI note number. The step is the time elapsed from the previous note or start of the track. The duration is how long the note will be playing in seconds and is the difference between the note end and note start times.

Extract the notes from the sample MIDI file.

def midi_to_notes(midi_file: str) -> pd.DataFrame:
  pm = pretty_midi.PrettyMIDI(midi_file)
  instrument = pm.instruments[0]
  notes = collections.defaultdict(list)

  # Sort the notes by start time
  sorted_notes = sorted(instrument.notes, key=lambda note: note.start)
  prev_start = sorted_notes[0].start

  for note in sorted_notes:
    start = note.start
    end = note.end
    notes['pitch'].append(note.pitch)
    notes['start'].append(start)
    notes['end'].append(end)
    notes['step'].append(start - prev_start)
    notes['duration'].append(end - start)
    prev_start = start

  return pd.DataFrame({name: np.array(value) for name, value in notes.items()})
raw_notes = midi_to_notes(sample_file)
raw_notes.head()

It may be easier to interpret the note names rather than the pitches, so you can use the function below to convert from the numeric pitch values to note names. The note name shows the type of note, accidental and octave number (e.g. C#4).

get_note_names = np.vectorize(pretty_midi.note_number_to_name)
sample_note_names = get_note_names(raw_notes['pitch'])
sample_note_names[:10]
array(['D#4', 'D#5', 'D#4', 'D#5', 'D#4', 'D#5', 'D#6', 'D#7', 'D#6',
       'D#7'], dtype='<U3')

To visualize the musical piece, plot the note pitch, start and end across the length of the track (i.e. piano roll). Start with the first 100 notes

def plot_piano_roll(notes: pd.DataFrame, count: Optional[int] = None):
  if count:
    title = f'First {count} notes'
  else:
    title = f'Whole track'
    count = len(notes['pitch'])
  plt.figure(figsize=(20, 4))
  plot_pitch = np.stack([notes['pitch'], notes['pitch']], axis=0)
  plot_start_stop = np.stack([notes['start'], notes['end']], axis=0)
  plt.plot(
      plot_start_stop[:, :count], plot_pitch[:, :count], color="b", marker=".")
  plt.xlabel('Time [s]')
  plt.ylabel('Pitch')
  _ = plt.title(title)
plot_piano_roll(raw_notes, count=100)

png

Plot the notes for the entire track.

plot_piano_roll(raw_notes)

png

Check the distribution of each note variable.

def plot_distributions(notes: pd.DataFrame, drop_percentile=2.5):
  plt.figure(figsize=[15, 5])
  plt.subplot(1, 3, 1)
  sns.histplot(notes, x="pitch", bins=20)

  plt.subplot(1, 3, 2)
  max_step = np.percentile(notes['step'], 100 - drop_percentile)
  sns.histplot(notes, x="step", bins=np.linspace(0, max_step, 21))

  plt.subplot(1, 3, 3)
  max_duration = np.percentile(notes['duration'], 100 - drop_percentile)
  sns.histplot(notes, x="duration", bins=np.linspace(0, max_duration, 21))
plot_distributions(raw_notes)

png

Create a MIDI file

You can generate your own MIDI file from a list of notes using the function below.

def notes_to_midi(
  notes: pd.DataFrame,
  out_file: str, 
  instrument_name: str,
  velocity: int = 100,  # note loudness
) -> pretty_midi.PrettyMIDI:

  pm = pretty_midi.PrettyMIDI()
  instrument = pretty_midi.Instrument(
      program=pretty_midi.instrument_name_to_program(
          instrument_name))

  prev_start = 0
  for i, note in notes.iterrows():
    start = float(prev_start + note['step'])
    end = float(start + note['duration'])
    note = pretty_midi.Note(
        velocity=velocity,
        pitch=int(note['pitch']),
        start=start,
        end=end,
    )
    instrument.notes.append(note)
    prev_start = start

  pm.instruments.append(instrument)
  pm.write(out_file)
  return pm
example_file = 'example.midi'
example_pm = notes_to_midi(
    raw_notes, out_file=example_file, instrument_name=instrument_name)

Play the generated MIDI file and see if there is any difference.

display_audio(example_pm)

As before, you can write files.download(example_file) to download and play this file.

Create the training dataset

Create the training dataset by extracting notes from the MIDI files. You can start by using a small number of files, and experiment later with more. This may take a couple minutes.

num_files = 5
all_notes = []
for f in filenames[:num_files]:
  notes = midi_to_notes(f)
  all_notes.append(notes)

all_notes = pd.concat(all_notes)
n_notes = len(all_notes)
print('Number of notes parsed:', n_notes)
Number of notes parsed: 13606

Next, create a tf.data.Dataset from the parsed notes.

key_order = ['pitch', 'step', 'duration']
train_notes = np.stack([all_notes[key] for key in key_order], axis=1)
notes_ds = tf.data.Dataset.from_tensor_slices(train_notes)
notes_ds.element_spec
TensorSpec(shape=(3,), dtype=tf.float64, name=None)

You will train the model on batches of sequences of notes. Each example will consist of a sequence of notes as the input features, and next note as the label. In this way, the model will be trained to predict the next note in a sequence. You can find a diagram explaining this process (and more details) in Text classification with an RNN.

You can use the handy window function with size seq_length to create the features and labels in this format.

def create_sequences(
    dataset: tf.data.Dataset, 
    seq_length: int,
    vocab_size = 128,
) -> tf.data.Dataset:
  """Returns TF Dataset of sequence and label examples."""
  seq_length = seq_length+1

  # Take 1 extra for the labels
  windows = dataset.window(seq_length, shift=1, stride=1,
                              drop_remainder=True)

  # `flat_map` flattens the" dataset of datasets" into a dataset of tensors
  flatten = lambda x: x.batch(seq_length, drop_remainder=True)
  sequences = windows.flat_map(flatten)

  # Normalize note pitch
  def scale_pitch(x):
    x = x/[vocab_size,1.0,1.0]
    return x

  # Split the labels
  def split_labels(sequences):
    inputs = sequences[:-1]
    labels_dense = sequences[-1]
    labels = {key:labels_dense[i] for i,key in enumerate(key_order)}

    return scale_pitch(inputs), labels

  return sequences.map(split_labels, num_parallel_calls=tf.data.AUTOTUNE)

Set the sequence length for each example. Experiment with different lengths (e.g. 50, 100, 150) to see which one works best for the data, or use hyperparameter tuning. The size of the vocabulary (vocab_size) is set to 128 representing all the pitches supported by pretty_midi.

seq_length = 25
vocab_size = 128
seq_ds = create_sequences(notes_ds, seq_length, vocab_size)
seq_ds.element_spec
(TensorSpec(shape=(25, 3), dtype=tf.float64, name=None),
 {'pitch': TensorSpec(shape=(), dtype=tf.float64, name=None),
  'step': TensorSpec(shape=(), dtype=tf.float64, name=None),
  'duration': TensorSpec(shape=(), dtype=tf.float64, name=None)})

The shape of the dataset is (100,1), meaning that the model will take 100 notes as input, and learn to predict the following note as output.

for seq, target in seq_ds.take(1):
  print('sequence shape:', seq.shape)
  print('sequence elements (first 10):', seq[0: 10])
  print()
  print('target:', target)
sequence shape: (25, 3)
sequence elements (first 10): tf.Tensor(
[[0.5625     0.         0.034375  ]
 [0.5234375  0.52083333 0.04270833]
 [0.5        0.53854167 0.0375    ]
 [0.375      0.25       0.04166667]
 [0.28125    0.27395833 0.05520833]
 [0.5390625  0.24375    0.03958333]
 [0.484375   0.26458333 0.03333333]
 [0.375      0.24583333 0.04270833]
 [0.28125    0.26354167 0.04270833]
 [0.5234375  0.25       0.04270833]], shape=(10, 3), dtype=float64)

target: {'pitch': <tf.Tensor: shape=(), dtype=float64, numpy=36.0>, 'step': <tf.Tensor: shape=(), dtype=float64, numpy=0.25>, 'duration': <tf.Tensor: shape=(), dtype=float64, numpy=0.078125>}

Batch the examples, and configure the dataset for performance.

batch_size = 64
buffer_size = n_notes - seq_length  # the number of items in the dataset
train_ds = (seq_ds
            .shuffle(buffer_size)
            .batch(batch_size, drop_remainder=True)
            .cache()
            .prefetch(tf.data.experimental.AUTOTUNE))
train_ds.element_spec
(TensorSpec(shape=(64, 25, 3), dtype=tf.float64, name=None),
 {'pitch': TensorSpec(shape=(64,), dtype=tf.float64, name=None),
  'step': TensorSpec(shape=(64,), dtype=tf.float64, name=None),
  'duration': TensorSpec(shape=(64,), dtype=tf.float64, name=None)})

Create and train the model

The model will have three outputs, one for each note variable. For pitch and duration, you will use a custom loss function based on mean squared error that encourages the model to output non-negative values.

def mse_with_positive_pressure(y_true: tf.Tensor, y_pred: tf.Tensor):
  mse = (y_true - y_pred) ** 2
  positive_pressure = 10 * tf.maximum(-y_pred, 0.0)
  return tf.reduce_mean(mse + positive_pressure)
input_shape = (seq_length, 3)
learning_rate = 0.005

inputs = tf.keras.Input(input_shape)
x = tf.keras.layers.LSTM(128)(inputs)

outputs = {
  'pitch': tf.keras.layers.Dense(128, name='pitch')(x),
  'step': tf.keras.layers.Dense(1, name='step')(x),
  'duration': tf.keras.layers.Dense(1, name='duration')(x),
}

model = tf.keras.Model(inputs, outputs)

loss = {
      'pitch': tf.keras.losses.SparseCategoricalCrossentropy(
          from_logits=True),
      'step': mse_with_positive_pressure,
      'duration': mse_with_positive_pressure,
}

optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)

model.compile(loss=loss, optimizer=optimizer)

model.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_1 (InputLayer)            [(None, 25, 3)]      0                                            
__________________________________________________________________________________________________
lstm (LSTM)                     (None, 128)          67584       input_1[0][0]                    
__________________________________________________________________________________________________
duration (Dense)                (None, 1)            129         lstm[0][0]                       
__________________________________________________________________________________________________
pitch (Dense)                   (None, 128)          16512       lstm[0][0]                       
__________________________________________________________________________________________________
step (Dense)                    (None, 1)            129         lstm[0][0]                       
==================================================================================================
Total params: 84,354
Trainable params: 84,354
Non-trainable params: 0
__________________________________________________________________________________________________

Testing the model.evaluate function, you can see that the pitch loss is significantly greater than the step and duration losses. Note that loss is the total loss computed by summing all the other losses and is currently dominated by the pitch loss.

losses = model.evaluate(train_ds, return_dict=True)
losses
212/212 [==============================] - 6s 3ms/step - loss: 5.0686 - duration_loss: 0.1985 - pitch_loss: 4.8481 - step_loss: 0.0220
{'loss': 5.068634510040283,
 'duration_loss': 0.19852526485919952,
 'pitch_loss': 4.8480706214904785,
 'step_loss': 0.022038981318473816}

One way balance this is to use the loss_weights argument to compile:

model.compile(
    loss=loss,
    loss_weights={
        'pitch': 0.05,
        'step': 1.0,
        'duration':1.0,
    },
    optimizer=optimizer,
)

The loss then becomes the weighted sum of the individual losses.

model.evaluate(train_ds, return_dict=True)
212/212 [==============================] - 1s 3ms/step - loss: 0.4630 - duration_loss: 0.1985 - pitch_loss: 4.8481 - step_loss: 0.0220
{'loss': 0.46296781301498413,
 'duration_loss': 0.19852526485919952,
 'pitch_loss': 4.8480706214904785,
 'step_loss': 0.022038981318473816}

Train the model.

callbacks = [
    tf.keras.callbacks.ModelCheckpoint(
        filepath='./training_checkpoints/ckpt_{epoch}',
        save_weights_only=True),
    tf.keras.callbacks.EarlyStopping(
        monitor='loss',
        patience=5,
        verbose=1,
        restore_best_weights=True),
]
%%time
epochs = 50

history = model.fit(
    train_ds,
    epochs=epochs,
    callbacks=callbacks,
)
Epoch 1/50
212/212 [==============================] - 2s 4ms/step - loss: 0.3816 - duration_loss: 0.1560 - pitch_loss: 4.1258 - step_loss: 0.0193
Epoch 2/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3563 - duration_loss: 0.1439 - pitch_loss: 3.8772 - step_loss: 0.0185
Epoch 3/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3529 - duration_loss: 0.1421 - pitch_loss: 3.8537 - step_loss: 0.0181
Epoch 4/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3478 - duration_loss: 0.1398 - pitch_loss: 3.7948 - step_loss: 0.0183
Epoch 5/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3480 - duration_loss: 0.1426 - pitch_loss: 3.7489 - step_loss: 0.0180
Epoch 6/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3459 - duration_loss: 0.1414 - pitch_loss: 3.7280 - step_loss: 0.0180
Epoch 7/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3389 - duration_loss: 0.1379 - pitch_loss: 3.6642 - step_loss: 0.0178
Epoch 8/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3352 - duration_loss: 0.1352 - pitch_loss: 3.6510 - step_loss: 0.0175
Epoch 9/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3337 - duration_loss: 0.1347 - pitch_loss: 3.6349 - step_loss: 0.0172
Epoch 10/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3319 - duration_loss: 0.1333 - pitch_loss: 3.6339 - step_loss: 0.0169
Epoch 11/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3280 - duration_loss: 0.1300 - pitch_loss: 3.6183 - step_loss: 0.0171
Epoch 12/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3239 - duration_loss: 0.1267 - pitch_loss: 3.6040 - step_loss: 0.0170
Epoch 13/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3217 - duration_loss: 0.1250 - pitch_loss: 3.5992 - step_loss: 0.0168
Epoch 14/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3161 - duration_loss: 0.1200 - pitch_loss: 3.5855 - step_loss: 0.0168
Epoch 15/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3143 - duration_loss: 0.1192 - pitch_loss: 3.5692 - step_loss: 0.0166
Epoch 16/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3117 - duration_loss: 0.1164 - pitch_loss: 3.5670 - step_loss: 0.0169
Epoch 17/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3080 - duration_loss: 0.1134 - pitch_loss: 3.5653 - step_loss: 0.0164
Epoch 18/50
212/212 [==============================] - 1s 4ms/step - loss: 0.3084 - duration_loss: 0.1123 - pitch_loss: 3.5965 - step_loss: 0.0163
Epoch 19/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2988 - duration_loss: 0.1055 - pitch_loss: 3.5410 - step_loss: 0.0162
Epoch 20/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2980 - duration_loss: 0.1028 - pitch_loss: 3.5758 - step_loss: 0.0164
Epoch 21/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2915 - duration_loss: 0.0993 - pitch_loss: 3.5235 - step_loss: 0.0160
Epoch 22/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2975 - duration_loss: 0.1024 - pitch_loss: 3.5846 - step_loss: 0.0159
Epoch 23/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2831 - duration_loss: 0.0911 - pitch_loss: 3.5186 - step_loss: 0.0161
Epoch 24/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2787 - duration_loss: 0.0883 - pitch_loss: 3.5001 - step_loss: 0.0154
Epoch 25/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2792 - duration_loss: 0.0889 - pitch_loss: 3.4912 - step_loss: 0.0158
Epoch 26/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2795 - duration_loss: 0.0897 - pitch_loss: 3.4919 - step_loss: 0.0153
Epoch 27/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2705 - duration_loss: 0.0810 - pitch_loss: 3.4829 - step_loss: 0.0153
Epoch 28/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2707 - duration_loss: 0.0819 - pitch_loss: 3.4695 - step_loss: 0.0154
Epoch 29/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2696 - duration_loss: 0.0804 - pitch_loss: 3.4733 - step_loss: 0.0156
Epoch 30/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2681 - duration_loss: 0.0804 - pitch_loss: 3.4568 - step_loss: 0.0149
Epoch 31/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2569 - duration_loss: 0.0700 - pitch_loss: 3.4456 - step_loss: 0.0146
Epoch 32/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2516 - duration_loss: 0.0647 - pitch_loss: 3.4387 - step_loss: 0.0149
Epoch 33/50
212/212 [==============================] - 1s 5ms/step - loss: 0.2556 - duration_loss: 0.0674 - pitch_loss: 3.4626 - step_loss: 0.0150
Epoch 34/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2418 - duration_loss: 0.0559 - pitch_loss: 3.4249 - step_loss: 0.0146
Epoch 35/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2352 - duration_loss: 0.0506 - pitch_loss: 3.4185 - step_loss: 0.0136
Epoch 36/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2559 - duration_loss: 0.0633 - pitch_loss: 3.5603 - step_loss: 0.0145
Epoch 37/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2411 - duration_loss: 0.0557 - pitch_loss: 3.4405 - step_loss: 0.0134
Epoch 38/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2358 - duration_loss: 0.0521 - pitch_loss: 3.4078 - step_loss: 0.0132
Epoch 39/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2312 - duration_loss: 0.0477 - pitch_loss: 3.3963 - step_loss: 0.0137
Epoch 40/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2319 - duration_loss: 0.0486 - pitch_loss: 3.3890 - step_loss: 0.0139
Epoch 41/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2338 - duration_loss: 0.0512 - pitch_loss: 3.3824 - step_loss: 0.0135
Epoch 42/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2383 - duration_loss: 0.0549 - pitch_loss: 3.4013 - step_loss: 0.0133
Epoch 43/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2454 - duration_loss: 0.0615 - pitch_loss: 3.3930 - step_loss: 0.0142
Epoch 44/50
212/212 [==============================] - 1s 4ms/step - loss: 0.2313 - duration_loss: 0.0498 - pitch_loss: 3.3607 - step_loss: 0.0134
Restoring model weights from the end of the best epoch.
Epoch 00044: early stopping
CPU times: user 56.9 s, sys: 12.1 s, total: 1min 8s
Wall time: 42.2 s
plt.plot(history.epoch, history.history['loss'], label='total loss')
plt.show()

png

Generate notes

To use the model to generate notes, you will first need to provide a starting sequence of notes. The function below generates one note from a sequence of notes.

For note pitch, it draws a sample from softmax distribution of notes produced by the model, and does not simply pick the note with the highest probability. Always picking the note with the highest probability would lead to repetitive sequences of notes being generated.

The temperature parameter can be used to control the randomness of notes generated. You can find more details on temperature in Text generation with an RNN.

def predict_next_note(
    notes: np.ndarray, 
    keras_model: tf.keras.Model, 
    temperature: float = 1.0) -> int:
  """Generates a note IDs using a trained sequence model."""

  assert temperature > 0

  # Add batch dimension
  inputs = tf.expand_dims(notes, 0)

  predictions = model.predict(inputs)
  pitch_logits = predictions['pitch']
  step = predictions['step']
  duration = predictions['duration']

  pitch_logits /= temperature
  pitch = tf.random.categorical(pitch_logits, num_samples=1)
  pitch = tf.squeeze(pitch, axis=-1)
  duration = tf.squeeze(duration, axis=-1)
  step = tf.squeeze(step, axis=-1)

  # `step` and `duration` values should be non-negative
  step = tf.maximum(0, step)
  duration = tf.maximum(0, duration)

  return int(pitch), float(step), float(duration)

Now generate some notes. You can play around with temperature and the starting sequence in next_notes and see what happens.

temperature = 2.0
num_predictions = 120

sample_notes = np.stack([raw_notes[key] for key in key_order], axis=1)

# The initial sequence of notes; pitch is normalized similar to training
# sequences
input_notes = (
    sample_notes[:seq_length] / np.array([vocab_size, 1, 1]))

generated_notes = []
prev_start = 0
for _ in range(num_predictions):
  pitch, step, duration = predict_next_note(input_notes, model, temperature)
  start = prev_start + step
  end = start + duration
  input_note = (pitch, step, duration)
  generated_notes.append((*input_note, start, end))
  input_notes = np.delete(input_notes, 0, axis=0)
  input_notes = np.append(input_notes, np.expand_dims(input_note, 0), axis=0)
  prev_start = start

generated_notes = pd.DataFrame(
    generated_notes, columns=(*key_order, 'start', 'end'))
generated_notes.head(10)
out_file = 'output.mid'
out_pm = notes_to_midi(
    generated_notes, out_file=out_file, instrument_name=instrument_name)
display_audio(out_pm)

You can also download the audio file by adding the two lines below:

from google.colab import files
files.download(out_file)

Visualize the generated notes.

plot_piano_roll(generated_notes)

png

Check the distributions of pitch, step and duration.

plot_distributions(generated_notes)

png

In the above plots, you will notice the change in distribution of the note variables. Since there is a feedback loop between the model's outputs and inputs, the model tends to generate similar sequences of outputs to reduce the loss. This is particularly relevant for step and duration, which has uses MSE loss. For pitch, you can increase the randomness by increasing the temperature in predict_next_note.

Next steps

This tutorial demonstrated the mechanics of using an RNN to generate sequences of notes from a dataset of MIDI files. To learn more, you can visit the closely related Text generation with an RNN tutorial, which contains additional diagrams and explanations.

An alternative to using RNNs for music generation is using GANs. Rather than generating audio, a GAN-based approach can generate a entire sequence in parallel. The Magenta team has done impressive work on this approach with GANSynth. You can also find many wonderful music and art projects and open-source code on Magenta project website.