songhash 0.1.1

Deterministic musical hash — turn a file's SHA-256 into ≤30 s of pleasant, music-theory-correct audio.

mhash

A deterministic musical fingerprint for any file. Pipe in some bytes, get back ≤30 seconds of pleasant, music-theory-correct audio. Same hash → identical track.

Like an identicon — but for your ears.

pip install songhash
mhash some/file        # plays a unique 30 s song for that file

╭─[ mhash ]──────────────────────────────────────────────────────────────────────╮
│  hash    10d59bff…89f1    source LICENSE    mime text/plain                        │
│  mood    M3 sub-flavor 1    tempo 106.6 BPM    key G ionian    meter 4/4           │
│  form    theme_var    groove straight_4_4    voicing close_triad    curve terraces │
│  matrix  4floor_house    humanize machine (jitter ±0)    mix balanced              │
╰────────────────────────────────────────────────────────────────────────────────────╯
  bars    A · A · A · A · Avar · Avar · Avar · Avar
  energy  ▃▃▄▄▅▆▆▇
  chords  Gmaj | Dmaj | Emin | Cmaj | Gmaj | Dmaj | Emin | Cmaj
  roman   I | V | vi | IV | I | V | vi | IV

  arrangement (rows × sections):
            A     B  fill
   drums       ◆     ◆     █
   bass        ◆     ◆     ▪
   comp        ▣     ▣     ·
   lead        ◆     ◆     ·
   riser       ▪     ▪     ·

  L1 drums   ch10 kit=latin-conga  low=lc-d2-tumbao / high=lc-d4-peak-mambo
          ▶ perc_1     X··X ··X· ··X· X···
          ▶ tom_high   ···· ··XX ···· ··XX
          ▶ tom_mid    X··· X··· ···· X···
  L2 bass    ch1  prog 032 Acoustic Bass     pat=bossa_bass
  L3 comp    ch2  prog 004 Electric Piano 1  pat=dotted_8th_pad   role=piano_stab
  L4 lead    ch3  prog 011 Vibraphone        motif=m44_fnk_01
  L5 pad     ch4  prog 089 Warm Pad
  L6 counter ch5  prog 073 Flute             mode=parallel_6th

  fx        Chorus(0.7Hz) → Delay(375ms) → Reverb(0.55)    audio 44100/16-bit @ -16 LUFS

  rendered  notes 359    peak-poly 12    midi 3.1 KB    wav 4.1 MB    dur 18.0s    took 0.82s

The dashboard shows everything the renderer is doing: the I-V-vi-IV roman numerals next to chord names, an arrangement matrix for which layers play in which form sections (· silent, ▪ sparse, ▣ normal, █ dense, ◆ energy-scaled), a one-bar drum-pattern grid for L1, and a render-stats footer with note count, peak polyphony, and timings.

Why

You can show someone an identicon and they'll spot a wrong digit at a glance. mhash does the same with sound: a deterministic, musical, memorable signature of a file's contents. Different files → different songs. Same file → exact same song, every time, on every machine.

Install

python3 -m venv .venv && source .venv/bin/activate
pip install songhash
brew install fluid-synth      # mac;  apt install fluidsynth  on debian/ubuntu
mhash some/file               # auto-downloads a SoundFont (~50 MB) on first run

What you can do

mhash path/to/file                  # hash + render + play (default)
mhash -o path/to/file               # write path/to/file.mhash.wav, no playback
mhash --out my.wav path/to/file     # write to a specific path
mhash --mood M14 path/to/file       # force a mood (M0..M14)
mhash -                             # read stdin
cat foo.bin | mhash                 # ditto, auto-detected
mhash -c 4 some.iso                 # chunk mode: split into 4-MB pieces,
                                    #   play one song per chunk in order
mhash --sf fluidr3 path/to/file     # try the FluidR3 SoundFont (auto-downloads ~141 MB)
mhash --sf ~/my.sf2 path/to/file    # use any .sf2/.sf3 you already have
mhash -q path/to/file               # silence dashboard + progress (just play)

# Power CLI (full flag surface — emit midi/mp3/flac, scores, etc.):
mhash --audio --mp3 --score path/to/file

# Per-mood demo: one sample per M0..M14, stitched into showcase/showcase.wav:
python3 -m mhash.showcase --score

A taste of the spectrum

The same file rendered through three very different moods (forced via --mood).

--mood M14 — chiptune (LSDj-flavored)

╭─[ mhash ]───────────────────────────────────────────────────────────────────╮
│  hash    7c496489…65e5    source README.md    mime text/plain                   │
│  mood    M14 sub-flavor 1    tempo 153.7 BPM    key C mixolydian    meter 4/4   │
│  form    pyramid    groove straight_4_4    voicing sus_open    curve arc        │
│  matrix  breakdown_pad    humanize groove-loose (jitter ±9)    mix lead_forward │
╰─────────────────────────────────────────────────────────────────────────────────╯
  bars    i · A · A · B · B · B · B · A · A · o
  energy  ▃▄▅▆▆▆▆▅▄▃
  chords  Amin | Fmaj | Cmaj | Gmaj | Amin | Fmaj | Cmaj | Gmaj | Amin | Fmaj
  roman   vi | IV | I | V | vi | IV | I | V | vi | IV

  L1 drums   kit=trap-808   esc=linear_add
          ▶ hat_closed X·X· X·X· X·X· X·X·
          ▶ kick       X··· ···· ··X· ····
          ▶ snare      ···· ···· X··· ····
  L2 bass    prog 039 Synth Bass 2   pat=montuno_bass
  L3 comp    prog 080 Square Lead    pat=jazz_freddie    role=synth_arp
  L4 lead    prog 084 Charang Lead   motif=m44_pop_08
  L5 pad     prog 088 New-Age Pad
  fx         HiShelf(-4dB) → Drive(2dB) → Reverb(0.10)

  rendered   notes 325    peak-poly 10    midi 3.1 KB    wav 4.3 MB    dur 15.6s

--mood M0 — ambient

╭─[ mhash ]─────────────────────────────────────────────────────────────────╮
│  hash    10d59bff…89f1    source LICENSE    mime text/plain                   │
│  mood    M0 sub-flavor 1    tempo 65.0 BPM    key G ionian    meter 4/4       │
│  form    pulse_only    groove straight_4_4    voicing power    curve flat_mid │
│  matrix  bass_lead_duo    humanize machine (jitter ±0)    mix balanced        │
╰───────────────────────────────────────────────────────────────────────────────╯
  bars    A · A · A · A · A · A · A
  energy  ▄▄▄▄▄▄▄
  chords  Gmaj | Dmaj | Gmaj | Dmaj | Gmaj | Dmaj | Gmaj
  roman   I | V | I | V | I | V | I

  L1 drums   kit=acoustic-studio   esc=reverse_cymbal_sweep
          ▶ hat_closed XXXX XXXX XXXX XXXX
          ▶ kick       X··· ·X·· ··X· ····
          ▶ snare      ···· X··· ···· X···
  L2 bass    prog 038 Synth Bass 1   pat=halfnote_fifth_walk
  L3 comp    prog 091 Choir Pad      pat=sustain_whole    role=harp_gliss
  L5 pad     prog 088 New-Age Pad
  L7 drone   prog 089 Warm Pad   tonic+5th pedal
  fx         Reverb(0.85) → Chorus(0.6Hz) → LoShelf(-1dB)

  rendered   notes 64    peak-poly 8    midi 0.9 KB    wav 5.3 MB    dur 25.9s

--mood M11 — lofi (jazzy)

╭─[ mhash ]────────────────────────────────────────────────────────────────╮
│  hash    c32a49aa…2e01    source BYTES.md    mime text/plain                 │
│  mood    M11 sub-flavor 0    tempo 70.1 BPM    key A# dorian    meter 4/4    │
│  form    AABA    groove dilla_feel    voicing open_triad    curve arc        │
│  matrix  band_basic    humanize groove-loose (jitter ±9)    mix lead_forward │
╰──────────────────────────────────────────────────────────────────────────────╯
  bars    A · A · A · A · B · B · A · A
  energy  ▃▄▅▆▆▅▄▃
  chords  A#maj | Gmin | Cmin | Fmaj | A#maj | Gmin | Cmin | Fmaj
  roman   I | vi | ii | V | I | vi | ii | V

  L1 drums   kit=jazz-kit   esc=reverse_cymbal_sweep
          ▶ ride       X··X X··· X··X X···
          ▶ kick       X··· ···· X··· ····
          ▶ snare      ···· X··· ···· X···
          ▶ hat_closed ···· X··· ···· X···
  L2 bass    prog 038 Synth Bass 1   pat=root_fifth_alt
  L3 comp    prog 024 Nylon Guitar   pat=montuno_2_3    role=piano_arp
  L4 lead    prog 071 Clarinet       motif=m44_pop_04
  L6 counter prog 005 Electric Piano 2   mode=parallel_3rd
  fx         Drive(6dB) → HiShelf(-3dB) → Chorus(0.4Hz) → Reverb(0.40)

  rendered   notes 193    peak-poly 9    midi 1.9 KB    wav 5.3 MB    dur 27.4s

How it works

1. Hash. SHA-256 of the input bytes.
2. Decode. Walk a deterministic HKDF-SHA256 byte stream over 32 macro decisions: mood, tempo, key, mode, time-signature, form, energy curve, chord progression, voicing, drum kit / pattern, bass pattern, comp role, melody motif, contour, FX, mix… A lookup table is pre-filtered by all prior decisions, so every combination produces music-theory-correct output.
3. MIDI. Render 9 simultaneous layers — drums (ch10), bass, comp, lead, pad, counter-melody, drone, riser, ear-candy stabs.
4. Audio. fluidsynth synthesizes the MIDI through a General-MIDI SoundFont; pedalboard applies a per-mood FX chain (reverb / delay / chorus / saturation / EQ); pyloudnorm normalises to -16 LUFS.
5. Play. macOS afplay, Linux paplay/aplay, Windows PowerShell SoundPlayer.

15 moods (M0 ambient → M14 gameboy/chiptune), 16 forms, 16 energy curves, 16 grooves, 12 drum kits, 39 chord progressions, 10 voicing styles, 8 escalation algorithms, ~2200 GM samples — all selected deterministically from the hash.

The MIME type of the input file biases the mood selection, so .json and .wav and .mkv tend to sound recognizably different even before content kicks in. (Override with --mime off for pure content-driven output.)

Setup notes

• pip install songhash pulls every Python dep (mido, numpy, python-magic, pyloudnorm, pedalboard, soundfile).
• Optional extras: pip install songhash[quality] for mosqito psychoacoustic scoring; pip install songhash[dev] for the test suite + build/twine.
• The fluidsynth system binary is required for audio render. macOS: brew install fluid-synth. Debian/Ubuntu: apt install fluidsynth. Windows: scoop install fluidsynth.
• The default GM SoundFont (MS-Basic.sf3, ~50 MB) is auto-downloaded on first mhash run into assets/v1/sf2/. To skip the per-render OGG-decode tax, the installer also produces an uncompressed MS-Basic.sf2 (~440 MB additional disk, ~12× faster renders). Override via SOUNDHASH_SOUNDFONT=/path/to/your.sf2.

Dev

git clone https://github.com/tabledevil/mhash
cd mhash
pip install -e ".[dev]"
pytest -q                      # 23 fast tests
mhash --self-test          # baseline MIDI SHA check
python -m build                # builds dist/*.whl + .tar.gz

See DESIGN.md for the full spec, BYTES.md for the byte-by-byte map of how each SHA-256 byte drives a musical decision, AESTHETICS.md for per-mood design intent, and CHANGELOG.md for the journey.

License

MIT. © 2026 tabledevil.