Sample packs: zoned sampling, CC0 recordings downloaded on demand

.github/workflows/ci.yml

@@ -482,6 +482,14 @@ jobs:
         run: |
           jq -n --arg relay "$JAM_RELAY" '{ relay: $relay }' > build-web/dist/config.json
 
+      # The sample packs, from the release scripts/packs.py names (made by
+      # the `sample packs' workflow, packs.yml), into the site's packs/,
+      # which the page downloads from when asked. Before that release is
+      # published the site keeps the empty index the build wrote.
+      - name: The sample packs
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/master'
+        run: python3 scripts/packs.py fetch build-web/dist/packs --missing-ok
+
       - name: The packaged site, for GitHub Pages
         if: github.event_name != 'pull_request' && github.ref == 'refs/heads/master'
         uses: actions/upload-pages-artifact@v3
@@ -530,6 +538,7 @@ jobs:
             build-web/gen
             build-web/patches
             build-web/icons
+            build-web/packs
             build-web/cairo2d/cairo2dtest.js
             build-web/cairo2d/cairo2dtest.wasm
 

.github/workflows/packs.yml

@@ -0,0 +1,77 @@
+# The sample packs, as a release of their own.
+#
+# scripts/packs.py builds them from their CC0 sources -- a couple of hundred
+# megabytes of downloads and an ffmpeg pass -- and this publishes the result
+# as the release scripts/packs.py names in RELEASE (packs-1, packs-2, ...):
+# a tar per pack and the index, about 17 MB. The site's deploy and
+# `packs.py install' download that and never touch the sources.
+#
+# By hand, and apart from a thinksynth release: the packs change when a pack
+# is added or rebuilt, not when the program does. To publish, bump RELEASE,
+# merge, and run this on master. It refuses a tag that already exists -- a
+# release the deploy points at is not replaced under it -- marks its
+# release as not the latest, so the program's own releases stay what
+# /releases/latest is, and then runs the ci workflow on master, whose
+# deploy is what puts the release on the site: the deploy that came with
+# the merge ran before the release existed.
+
+name: sample packs
+
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: write
+  actions: write
+
+jobs:
+  publish:
+    name: build and publish the sample packs
+    if: github.ref == 'refs/heads/master'
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: ffmpeg
+        run: |
+          sudo apt-get update -q
+          sudo apt-get install -y --no-install-recommends ffmpeg
+
+      # The tag, not only the release: a bare tag left behind would pass a
+      # release check and the release would be made on that old commit.
+      - name: The release this would make, not made already
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          tag=$(python3 scripts/packs.py tag)
+          # Free only on a 404: any other failure is not an answer.
+          if out=$(gh api "repos/$GITHUB_REPOSITORY/git/ref/tags/$tag" 2>&1); then
+            echo "::error::the tag $tag exists already; bump RELEASE in scripts/packs.py"
+            exit 1
+          elif ! grep -q "HTTP 404" <<< "$out"; then
+            echo "::error::could not look up the tag $tag: $out"
+            exit 1
+          fi
+          echo "TAG=$tag" >> "$GITHUB_ENV"
+
+      - name: The packs, from their sources
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+        run: |
+          python3 scripts/packs.py build out
+          python3 scripts/packs.py archive out
+
+      - name: The release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          gh release create "$TAG" out/*.tar out/index.json \
+            --target "$GITHUB_SHA" --latest=false \
+            --title "Sample packs ($TAG)" \
+            --notes "Recorded instruments for the sampled_*.dsp graphs, built by scripts/packs.py at $GITHUB_SHA. Public domain (CC0): VSCO 2 Community Edition (Versilian Studios / Sam Gossner) and Big Rusty Drums (Karoryfer Samples)."
+
+      - name: The site, deployed with it
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: gh workflow run ci.yml --ref master

README.md

@@ -182,6 +182,17 @@ browser, the page's MIDI out button does the same through Web MIDI
 (Chromium and Firefox). Either can also send MIDI clock, so a drum machine or
 a DAW follows the piece's tempo and transport.
 
+The sampled instruments (`sampled_*.dsp`: strings, horn, trumpet, flute,
+upright piano, a drum kit) play public-domain recordings that are not in this
+repository: they are published as a release of their own, built by the
+`sample packs` workflow. `scripts/packs.py install` (it needs ffmpeg)
+downloads it into this user's data directory, where thinksynth looks —
+`~/.local/share/thinksynth` on Linux, `~/Library/Application Support/thinksynth`
+on macOS, `%LOCALAPPDATA%\thinksynth` on Windows, and the first of those
+for the Flatpak too. A running thinksynth finds a newly installed pack the
+next time it loads an instrument; the page offers each pack for download
+when something in play needs it.
+
 If you want JACK on Linux, start `jackd` before thinksynth — RtAudio will use
 the running server.
 

docs/NODES.md

@@ -1064,7 +1064,7 @@ Sample Player (a wav at a voice's pitch)
 
 | Arg | Dir | Description | Default | Range | Units | Values |
 |---|---|---|---|---|---|---|
-| `file` | in | The wav to play, found under samples/ on THINK_DSP_PATH |  |  |  |  |
+| `file` | in | The wav to play, found under samples/ on THINK_DSP_PATH; or zones, `a.wav@48 b.wav@55', the one nearest the note played from its own |  |  |  |  |
 | `file2` | in | Middle layer wav; an empty slot uses file |  |  |  |  |
 | `file3` | in | Upper layer wav; an empty slot uses file2 or file |  |  |  |  |
 | `freq` | in | The note to play it at |  |  | Hz |  |

dsp/sampled_cellos.dsp

@@ -0,0 +1,70 @@
+# Cello section (sampled) -- recorded, from the `cellos' sample pack.
+#
+# VSCO 2 Community Edition, Versilian Studios / Sam Gossner; CC0. The
+# recordings are not in this repository: scripts/packs.py builds them
+# from their source into a release of their own, and the page offers the
+# pack for download when something in play needs it. Without it, this
+# plays silence.
+#
+# One osc::sample, its `file' a set of zones -- each recording with the
+# note it was made at -- so every note is played from the nearest
+# recording. 2 velocity layers, chosen by how hard the note is played.
+#
+# The recordings are trimmed to a few seconds, and the last second
+# loops with a 300 ms crossfade, so a held note holds.
+
+name "Cello section (sampled)";
+author "Misha Nasledov";
+description "Cello section, recorded (VSCO 2 Community Edition, Versilian Studios / Sam Gossner).";
+category "Strings and pads";
+
+    @a = 120 ms;
+    @a.widget = 1;
+    @a.min = 1 ms;
+    @a.max = 2000 ms;
+    @a.label = "Attack";
+
+    @r = 450 ms;
+    @r.widget = 1;
+    @r.min = 70 ms;
+    @r.max = 4000 ms;
+    @r.label = "Release";
+
+node ionode {
+    channels = 2;
+    out0 = vca->out;
+    out1 = vca->out;
+    play = env->play;
+};
+
+node freq misc::midi2freq {
+    note = ionode->note;
+};
+
+node smp osc::sample {
+    file = "cellos/cellos_36_1.wav@36 cellos/cellos_40_1.wav@40 cellos/cellos_43_1.wav@43 cellos/cellos_47_1.wav@47 cellos/cellos_50_1.wav@50 cellos/cellos_53_1.wav@53 cellos/cellos_57_1.wav@57 cellos/cellos_60_1.wav@60 cellos/cellos_64_1.wav@64 cellos/cellos_67_1.wav@67 cellos/cellos_71_1.wav@71 cellos/cellos_74_1.wav@74 cellos/cellos_77_1.wav@77";
+    file2 = "cellos/cellos_36_2.wav@36 cellos/cellos_40_2.wav@40 cellos/cellos_43_2.wav@43 cellos/cellos_47_2.wav@47 cellos/cellos_50_2.wav@50 cellos/cellos_53_2.wav@53 cellos/cellos_57_2.wav@57 cellos/cellos_60_2.wav@60 cellos/cellos_64_2.wav@64 cellos/cellos_67_2.wav@67 cellos/cellos_71_2.wav@71 cellos/cellos_74_2.wav@74 cellos/cellos_77_2.wav@77";
+    freq = freq->out;
+    trigger = ionode->trigger;
+    select = ionode->velocity;
+    split1 = 0.6;
+    split2 = 1;
+    loop = 1000 ms;
+    xfade = 300 ms;
+};
+
+node env env::adsr {
+    a = @a;
+    d = 0;
+    s = 0.5 + 0.5 * ionode->velocity;
+    r = @r;
+    p = 0.5 + 0.5 * ionode->velocity;
+    trigger = ionode->trigger;
+};
+
+node vca mixer::mul {
+    in0 = smp->out;
+    in1 = env->out * 0.75;
+};
+
+io ionode;

dsp/sampled_drums.dsp

@@ -0,0 +1,56 @@
+# Drum kit (sampled) -- recorded, from the `drums' sample pack.
+#
+# Big Rusty Drums, Karoryfer Samples; CC0. The recordings are not in
+# this repository: scripts/packs.py builds them from their source into a
+# release of their own, and the page offers the pack for download when
+# something in play needs it. Without it, this plays silence.
+#
+# One osc::sample, its `file' a set of zones -- each recording with the
+# note it was made at -- so every note is played from the nearest
+# recording. 3 velocity layers, chosen by how hard the note is played.
+#
+# General MIDI drum notes: 36 kick, 38 snare, 41 43 45 48 toms,
+# 42 closed and 46 open hi-hat, 49 crash, 51 ride. Every other
+# note plays its nearest drum, retuned.
+
+name "Drum kit (sampled)";
+author "Misha Nasledov";
+description "Drum kit, recorded (Big Rusty Drums, Karoryfer Samples).";
+category "Drums";
+
+node ionode {
+    channels = 2;
+    poly = 16;
+    out0 = vca->out;
+    out1 = vca->out;
+    play = max(smp->play, least->play);
+};
+
+# At least a tenth of a second, so a voice whose file is not
+# there yet still lives long enough to be one.
+node least env::ad {
+    a = 0;
+    d = 100 ms;
+};
+
+node freq misc::midi2freq {
+    note = ionode->note;
+};
+
+node smp osc::sample {
+    file = "drums/k_36_1.wav@36 drums/sn_center_38_1.wav@38 drums/t22_41_1.wav@41 drums/ht_cl_42_1.wav@42 drums/t18_43_1.wav@43 drums/t15_45_1.wav@45 drums/ht_open_46_1.wav@46 drums/t14_48_1.wav@48 drums/cr_49_1.wav@49 drums/rd_51_1.wav@51";
+    file2 = "drums/k_36_2.wav@36 drums/sn_center_38_2.wav@38 drums/t22_41_2.wav@41 drums/ht_cl_42_2.wav@42 drums/t18_43_2.wav@43 drums/t15_45_2.wav@45 drums/ht_open_46_2.wav@46 drums/t14_48_2.wav@48 drums/cr_49_2.wav@49 drums/rd_51_2.wav@51";
+    file3 = "drums/k_36_3.wav@36 drums/sn_center_38_3.wav@38 drums/t22_41_3.wav@41 drums/ht_cl_42_3.wav@42 drums/t18_43_3.wav@43 drums/t15_45_3.wav@45 drums/ht_open_46_3.wav@46 drums/t14_48_3.wav@48 drums/cr_49_3.wav@49 drums/rd_51_3.wav@51";
+    freq = freq->out;
+    trigger = 1;
+    select = ionode->velocity;
+    split1 = 0.4;
+    split2 = 0.75;
+};
+
+node vca mixer::mul {
+    in0 = smp->out;
+    in1 = (0.4 + 0.6 * ionode->velocity) * 0.85;
+};
+
+io ionode;

dsp/sampled_flute.dsp

@@ -0,0 +1,70 @@
+# Flute (sampled) -- recorded, from the `flute' sample pack.
+#
+# VSCO 2 Community Edition, Versilian Studios / Sam Gossner; CC0. The
+# recordings are not in this repository: scripts/packs.py builds them
+# from their source into a release of their own, and the page offers the
+# pack for download when something in play needs it. Without it, this
+# plays silence.
+#
+# One osc::sample, its `file' a set of zones -- each recording with the
+# note it was made at -- so every note is played from the nearest
+# recording. 2 velocity layers, chosen by how hard the note is played.
+#
+# The recordings are trimmed to a few seconds, and the last second
+# loops with a 300 ms crossfade, so a held note holds.
+
+name "Flute (sampled)";
+author "Misha Nasledov";
+description "Flute, recorded (VSCO 2 Community Edition, Versilian Studios / Sam Gossner).";
+category "Leads and stabs";
+
+    @a = 30 ms;
+    @a.widget = 1;
+    @a.min = 1 ms;
+    @a.max = 2000 ms;
+    @a.label = "Attack";
+
+    @r = 200 ms;
+    @r.widget = 1;
+    @r.min = 70 ms;
+    @r.max = 4000 ms;
+    @r.label = "Release";
+
+node ionode {
+    channels = 2;
+    out0 = vca->out;
+    out1 = vca->out;
+    play = env->play;
+};
+
+node freq misc::midi2freq {
+    note = ionode->note;
+};
+
+node smp osc::sample {
+    file = "flute/flute_60_1.wav@60 flute/flute_64_1.wav@64 flute/flute_69_1.wav@69 flute/flute_72_1.wav@72 flute/flute_76_1.wav@76 flute/flute_81_1.wav@81 flute/flute_84_1.wav@84 flute/flute_88_1.wav@88 flute/flute_93_1.wav@93 flute/flute_96_1.wav@96";
+    file2 = "flute/flute_60_2.wav@60 flute/flute_64_2.wav@64 flute/flute_69_2.wav@69 flute/flute_72_2.wav@72 flute/flute_76_2.wav@76 flute/flute_81_2.wav@81 flute/flute_84_1.wav@84 flute/flute_88_1.wav@88 flute/flute_93_2.wav@93 flute/flute_96_1.wav@96";
+    freq = freq->out;
+    trigger = ionode->trigger;
+    select = ionode->velocity;
+    split1 = 0.6;
+    split2 = 1;
+    loop = 1000 ms;
+    xfade = 300 ms;
+};
+
+node env env::adsr {
+    a = @a;
+    d = 0;
+    s = 0.5 + 0.5 * ionode->velocity;
+    r = @r;
+    p = 0.5 + 0.5 * ionode->velocity;
+    trigger = ionode->trigger;
+};
+
+node vca mixer::mul {
+    in0 = smp->out;
+    in1 = env->out * 3.2;
+};
+
+io ionode;

dsp/sampled_horn.dsp

@@ -0,0 +1,71 @@
+# French horn (sampled) -- recorded, from the `horn' sample pack.
+#
+# VSCO 2 Community Edition, Versilian Studios / Sam Gossner; CC0. The
+# recordings are not in this repository: scripts/packs.py builds them
+# from their source into a release of their own, and the page offers the
+# pack for download when something in play needs it. Without it, this
+# plays silence.
+#
+# One osc::sample, its `file' a set of zones -- each recording with the
+# note it was made at -- so every note is played from the nearest
+# recording. 3 velocity layers, chosen by how hard the note is played.
+#
+# The recordings are trimmed to a few seconds, and the last second
+# loops with a 300 ms crossfade, so a held note holds.
+
+name "French horn (sampled)";
+author "Misha Nasledov";
+description "French horn, recorded (VSCO 2 Community Edition, Versilian Studios / Sam Gossner).";
+category "Leads and stabs";
+
+    @a = 40 ms;
+    @a.widget = 1;
+    @a.min = 1 ms;
+    @a.max = 2000 ms;
+    @a.label = "Attack";
+
+    @r = 250 ms;
+    @r.widget = 1;
+    @r.min = 70 ms;
+    @r.max = 4000 ms;
+    @r.label = "Release";
+
+node ionode {
+    channels = 2;
+    out0 = vca->out;
+    out1 = vca->out;
+    play = env->play;
+};
+
+node freq misc::midi2freq {
+    note = ionode->note;
+};
+
+node smp osc::sample {
+    file = "horn/horn_36_1.wav@36 horn/horn_39_1.wav@39 horn/horn_43_1.wav@43 horn/horn_46_1.wav@46 horn/horn_50_1.wav@50 horn/horn_53_1.wav@53 horn/horn_57_1.wav@57 horn/horn_60_1.wav@60";
+    file2 = "horn/horn_36_2.wav@36 horn/horn_39_2.wav@39 horn/horn_43_2.wav@43 horn/horn_46_2.wav@46 horn/horn_50_2.wav@50 horn/horn_53_2.wav@53 horn/horn_57_2.wav@57 horn/horn_60_2.wav@60";
+    file3 = "horn/horn_36_2.wav@36 horn/horn_39_2.wav@39 horn/horn_43_2.wav@43 horn/horn_46_2.wav@46 horn/horn_50_3.wav@50 horn/horn_53_2.wav@53 horn/horn_57_2.wav@57 horn/horn_60_3.wav@60";
+    freq = freq->out;
+    trigger = ionode->trigger;
+    select = ionode->velocity;
+    split1 = 0.4;
+    split2 = 0.75;
+    loop = 1000 ms;
+    xfade = 300 ms;
+};
+
+node env env::adsr {
+    a = @a;
+    d = 0;
+    s = 0.5 + 0.5 * ionode->velocity;
+    r = @r;
+    p = 0.5 + 0.5 * ionode->velocity;
+    trigger = ionode->trigger;
+};
+
+node vca mixer::mul {
+    in0 = smp->out;
+    in1 = env->out * 0.5;
+};
+
+io ionode;