Start with one sound

Tap Start Listening, hum or tap, then watch Neural Bloom light up.

Short sounds spark quick signals. Longer sounds grow steady electric paths.

0 sounds

Use the tools below when you are ready to record, tap, compose, and send feedback.

Record a hum

Tap Record, make a short sound, then play it back. The digest preview is just a receipt for what the model would read; you do not need to use it.

Babbled Notes Neural Bloom is listening Your sound will become a music-code seed when you record.

Optional voice control: tap Voice start once, then say "start recording" or "stop recording."

Recording quality Waiting for sound
Music features No digest yet
# Your starter .lilt code will appear after recording.

Music as access

Some people have songs inside but no easy way to get them out. Babbled Notes is for autistic users, non-speaking users, people with speech or motor differences, and anyone who needs a gentler path into composing.

Voice

Hum one note or a short phrase.

Touch

Tap notes with large buttons.

Switch

Build music from one intentional choice.

Partner

Share readable text for another person to edit.

Tap a melody

No voice required. Use touch, keyboard focus, or a switch control to place notes one at a time, then hear the phrase.

tempo 80
feel straight
key C major

mood gentle

voice voice:
  rest 4

Try a demo

 

Make your music win

Write one tiny idea, choose a classical style, then let it replay while you change notes. The goal is to hear your idea become worth keeping.

Classical style
Sound space
86tempo
3voices
11tokens
4srough length
Warm depthdepth
Roomspace
Soft rubatomotion

Local AI voice render is available from the CLI for server-side use: python -m lilt.cli voice examples\three_note_hum.json --style chopin

Why it is different

Babbled Notes is not a chatbot that talks about music. It turns a sound, a tap pattern, or a simple gesture into a small music program that can be edited, replayed, exported, and shared.

The goal is musical agency for people who may hum, tap, use a switch, use assisted speech, or communicate without speech at all. The first playback should feel like proof: your sound can become music.

Build a song

A full Babbled Notes build starts with one small musical thought, then grows through plain text. This is the submission workflow.

  1. Capture. Hum the hook, tap notes, or make a rhythm.
  2. Translate. Babbled Notes turns the gesture into readable code.
  3. Arrange. Pick instruments and add voices for melody, bass, or drums.
  4. Share. Send the `.lilt` text or MIDI to another musician.
# Brooke's first idea
tempo 86
feel straight
key C major

mood gentle, warm

voice melody:
  C4 ! soft E4 ! mf G4 hold ! mf

voice pulse:
  x . . . x . . .

voice bass:
  C2 hold ! soft

Work with others

Babbled Notes is collaboration-friendly because the song is text. A partner can review the idea, change one line, and send it back without opening a DAW.

Vocal idea

Brooke records or taps the hook.

Readable file

Babbled Notes creates `.lilt` text and MIDI.

Partner edit

A collaborator changes the bass, drums, key, or instrument.

Shared build

GitHub shows exactly what changed, line by line.

- voice bass:
-   C2 hold ! soft
+ voice bass:
+   C2 ! soft G1 ! soft C2 hold ! mf

Sound alphabet

Babbled Notes treats small sounds and gestures as musical letters. Speech is not required; a hum, breath, tap, click, or switch press can be enough.

HumC4 ! mf
Long humG4 hold
Pauserest 1
Tap or clickx
Soft breatho
Rising soundC4 ~ E4
Loud sound! loud
Short sound/ staccato

Audio digest

This is the builder view: Babbled Notes turns a short WAV into timing and pitch facts, then Gemma uses those facts to write the music program.

python -m lilt.cli digest clip.wav
python -m lilt.cli audio clip.wav --digest clip.digest.json --backend gemini

Tell Brooke how it sounded

How did it sound?
Technical preview of what gets shared

      

What is this?

Hum, beatbox, whistle, or play a phrase. Babbled Notes turns it into a short, human-readable program. Edit the program. Hear the result. Share the text.

Built for the Gemma 4 challenge by Brooke Chauntel. The model and the audio pipeline live offline; this page is the in-browser preview.

See the code on GitHub »