Where we started, where we got to

The idea

Your phone is always in your pocket, and it has a microphone. The idea was to use it as a musical notebook: hum a bass line or beatbox a groove, and the app works out rhythm and notes, has a real instrument play them, and lets you put the pieces together into a song. Then you export everything, track by track, and finish it on the computer, in Reaper, singing over it with a proper microphone.

Record your voice → the app transcribes → pick the instrument → arrange the parts → listen to the mix → export → finish in Reaper.

The phone was never meant to replace the studio. It was meant to do the first half of the job well: catch the idea and give it a shape that sounds.

The choices made at the start

Before the first line was written there was a document of decisions not to be reopened:

• Fully offline, no exceptions. The app does not even ask for permission to go online. Recordings of someone singing are private: the simplest way to protect them is to be unable to send them anywhere.
• Android only, in Italian and English. One phone, two languages, done well.
• The first milestone was small on purpose: a step sequencer that plays, mixes and exports. Everything else would be added on top of a base already running on the phone.
• Precision first. A hit the app makes up is noticed and deleted; a missed one is not. When the app is not sure, it stays quiet instead of guessing.

The milestones

1. Before the first version

   The foundations

   One after another, the parts everything else stands on: the grid that plays and exports to WAV; the mixer; recording with a count-in, a click that turns into vibration on the speaker (otherwise it would end up in the microphone) and a measurement of the phone's latency; the transcription of hits and of notes; the instruments; a human touch that moves every hit by a hair, because perfect drums sound fake.

2. 7 September 2026 · version 1.0

   The first complete build

   From voice to pattern, grid, mixer with effects, chords, arrangement, albums, export of stems, MIDI, MP3, guide vocal and click. The same day the app learns to open on the library instead of inside a song: opening inside a track, nobody suspected there was a level above.

3. 8 September

   The busiest day

   Albums with covers, sound packs to import, EQ and compressor per track, odd metres, one MIDI file per track, sidechain, four reverb spaces, volume curves. And above all, beatbox that stops landing entirely on the snare: tested on twenty seconds of real beatbox, hit by hit.

4. 9–11 September

   Words, and real sounds

   Lyrics are written on the metre of the melody: the app tells you how many syllables fit and where the stress falls. You can ask an AI for the lyrics, and what comes back is checked line by line. Karaoke arrives. Latency is measured once for each pair of headphones.

   And the synthesised placeholder sounds make way for recorded instruments: Lars Muldjord's MuldjordKit drums and the Salamander piano, a Yamaha C5 sampled note by note.

5. 14–17 September

   The app learns your mouth

   To tell whether a hit was a kick, the app compared it with a real kick drum, and a “boom” breathed into a phone sounds nothing like one. Hence the idea of teaching it: make your sound five or six times, and from then on it recognises your hits. Tested on recordings of 28 people: hits recognised correctly go from 26.6% to 68.1%.

   The same week: a twelve-piece orchestra, MIDI parts and tabs brought back in, audio tracks recorded elsewhere, the sections view and the song map, triplets, breaths recognised and kept out of the pattern, song versions, the tuner, quick idea capture, and recording over the song instead of over the click alone.

6. 24–29 September

   Free instruments, and testing on the phone

   Electric bass, guitars and a Hammond organ join the built-in pack, all recorded and free (FreePats, CC0). SoundFonts and SFZ libraries can be imported straight from the phone. Guitar and bass scores come out at the right pitch.

   It is also the time of systematic testing on a real phone, screen by screen: whatever does not behave as expected goes back, gets fixed, and is tried again.

7. 2–9 October · version 1.5.99

   Ready for someone else

   Undo and redo for each mixer track; several takes picked and chained into one transcription; comparing two versions of a song one difference at a time, listening before deciding. And finally the guide inside the app, with a tutorial for every feature: the step needed to hand it to someone who did not watch it being made.

8. 9 October · versions 1.5.100 and 1.6.0

   What other people ask for

   Before handing it out, a look at what people write about similar apps: thousands of reviews and discussions, to find what they were missing and this app was missing too. The strongest request was automatic accompaniment, the way an app now withdrawn used to do it: now, under a hummed melody, the app suggests the chords and puts bass and drums underneath in the style you pick.

   Then the others: seeing the note you are singing while you sing it; takes up to a minute long; several versions of a lyric line, to choose by singing them; the rhyme scheme with words that rhyme, from two dictionaries built into the app, one Italian and one English; and the library sorted by name, with labels.

9. 9 October · version 1.6.2

   A song to listen to right away

   Until now, opening the app for the first time gave you eight bars of drums and bass: a test bench, not a song. Now there is “Oltre l'ombra”, symphonic power metal in D minor: strings and violin, distorted guitar, double kick in the chorus, lyrics with their rhymes. The mix is measured to stay below zero even without the limiter: a phone speaker crackles long before a studio does.

   In the same round, three fixes: the lyrics have their own language, Italian or English, even when the app is set to the other one; a bass imported from a MIDI file plays at its own pitch instead of two octaves down; and exported tracks start exactly together with the click.

10. 11 October · version 1.6.3

    Slower, and recordings change key too

    To learn a part you listen to the song at three quarters or half speed, at the same pitch. Recordings too: they are stretched without dropping in pitch. And when the song changes key, audio tracks now follow instead of staying where they were.

    The export measures the level between one sample and the next as well, where a phone or an MP3 can distort a file that looked fine. Stretching and measuring come from SoundCraft, an open-source project by ArtCraft, rewritten for the app.

11. 11 October · version 1.6.4

    The neck that plays

    From a video of a real guitar next to a drawn one: while the song plays, on the app's neck every note lights up its fret and the stretch of string that actually vibrates, from the fret to the bridge. Bends push the string towards the next one. And the string can be bent with your finger: press a fret, drag across, and the note rises as you write it.

    In the same round recordings follow tempo changes, notes played together all have a chord name, and the export makes a vertical video for social media, with the lyrics lighting up as they are sung.

12. 11 October · version 1.6.5

    A real keyboard

    A MIDI keyboard plugged into the phone writes into the grid like the on-screen one: every key puts down its note, with the strength you press it with, and several keys together a chord. Playing over the song live will come after a test on the phone, because there every millisecond of delay is heard.

    The video for social media also shows the guitar neck with its notes lighting up, and the app opens lighter: screens used rarely load only when they are needed.

13. 11 October · version 1.6.6

    Studio sounds, for free

    Two electronic drum kits arrive, the 808 and the 909, rebuilt like the circuits of the original machines: the app computes them as it plays, so they weigh nothing. You pick them in the mixer, and picked on one row the whole kit follows. With them come two new synth sounds, the trance Supersaw and a soft pad that sits under the voice.

    The guitar now strums: the strings of a chord start one after the other, down on the beat and up on the offbeat, like a real hand. And when a pack has recorded the same note more than once, the app alternates them, so a repeated chord no longer sounds like a machine gun. It is the first step towards free but professional sounds: a drum kit recorded with real microphones, a guitar recorded direct, a bass and a string section, all free of rights, are already being tried out.

14. 12 October · version 1.6.7

    Sections, like in a studio

    Looking at how AI music services let you work on a song one section at a time, the app learnt to do the same without a network. Any section can be moved by a semitone: the last chorus a step up, melody and chords included.

    A section's drums and bass can be redone fuller, lighter, at half time or with a fill at the end, and every version is heard before you keep it. At section changes one tap adds a riser that peaks exactly on the first beat, at any tempo, or a drop for when the energy falls.

15. 12 October · version 1.6.8

    Lyrics in their own language

    The language of the lyrics is not the app's: a song in English can be written with the app in Italian. Now you also choose it at the moment you ask an AI for the lyrics, and rhymes and syllables follow it.

16. 12 October · version 1.6.9

    A section's chords, and transitions in one tap

    A section's chords get their versions too: long, strummed or arpeggiated just there, or each replaced by its relative, where the melody fits. And one tap adds transitions across the whole song: a riser before the choruses, a reverse cymbal before the bridge, a drop after.

    From here the app also says it plainly: what you make is yours. Songs can be published and sold, and of the built-in sounds only the drums and the piano ask you to credit their author.

17. 12 October · version 1.6.10

    A second voice, and a song from a melody

    A section's melody can be doubled: a second voice a third above, in the song's scale, the way a backing singer would sing it, or the same melody an octave up for the last chorus. Like the other versions, you hear it before you keep it.

    From a melody you have just hummed, «Make it a song» prepares a draft with chords, a band in the chosen style and sections, on a copy: the original stays as it was. And every export carries the credits of the sounds the song actually uses, with the line to paste already written.

18. 12 October · version 1.6.11

    Music that keeps going, and a voice that swells

    The loop now keeps playing with the screen off or while you write somewhere else, with a notification to pause or stop it. And an idea can be caught from the phone's quick settings too, with the «An idea» tile: one tap, and you are at the microphone.

    When you sing a note that swells or fades, the instrument swells and fades with you. As soon as a melody is transcribed, the app offers by itself to make it a song. A section's versions have several takes to hear one after another, and the second voice can sit below the melody or on the notes of the chord.

19. 12 October · version 1.6.12

    The whole library in one file, and a draft that sounds like a song

    One tap puts every song, recording and album into a single file, which on a new phone comes back adding what is missing and leaving what is there untouched. While you write the lyrics, a section can loop on its own, and you try the line by singing it.

    «Make it a song» now builds a whole piece: an intro before the melody comes in, an ending, the last chorus fuller and a step up. And a section's chords have several takes too: the acoustic guitar strum, the reggae offbeat, the arpeggio from the top.

20. 13 October · version 1.6.13

    A song with a bridge, and lyrics that follow the music

    The draft now comes short, classic or with a bridge: a bridge with changed chords and the drums at half time, and after it the last chorus, higher. While a section loops under the lyrics, the note and syllable playing light up on the line.

    The karaoke slows down and loops the section for practice; a single section exports as a WAV or a video to share; and the library copy is written one piece at a time, with a reminder when it is time to make one.

21. 13 October · version 1.6.14

    A sound engineer and a guitarist look at the app

    To improve the sound and the guitar parts the app now has two expert eyes: one measures how a song sounds from a phone speaker, the other puts every guitar part on the neck and says what a real hand plays and what it doesn't. The first things they found are already fixed.

    The tab takes the tuning the part asks for — a guitar in Drop D no longer loses its low Ds — and writes each chord in a single shape that suits the hand. There is palm mute, the muted metal chug, in the grid and in the tab. And «Oltre l'ombra» sounds cleaner from a phone: the bass makes room for the kick and the snare cracks.

22. 13 October · version 1.6.15

    Metal has its subgenres, and the bass comes through a phone

    The band under the melody no longer just says «rock» or «metal»: you pick the family first and then the subgenre, twenty-five in all. Metal is heavy, thrash, power, doom or metalcore; punk is punk rock, hardcore, pop punk or ska punk; and the same goes for pop, rock, funk, Latin, electronic and blues. Where the style wants one, a guitar comes in too, with its muted power chords or its open chords, and every style starts with levels the sound engineer measured one by one.

    A phone plays nothing below 200 Hz: from there, bass and kick all but vanished. Now a new effect, harmonics, adds above what the speaker can play, and the note comes through. And the mixer has «Listen like a phone»: on headphones, you hear the song the way someone opening it on a phone will.

23. 15 October · version 1.6.16

    Hear a style before you pick it, and a guitar that plays like a guitarist

    Twenty-five styles can't be chosen by name: with “Listen before choosing” each style plays four bars of your song, which stays as it is, and you take the right one with “Use”. The band's guitar stops hitting the whole chord every time: in pop it strums up on the high strings, the ballad is picked, in bossa nova the thumb plays the bass and the fingers the high strings, funk fills the gaps with muted strings.

24. 15 October · version 1.6.17

    A bass player, a drummer, an arranger and a songwriter

    Four more expert eyes look at what the app writes on its own, and the first things they found are fixed. The draft's drums are no longer one bar repeated: sections end with a fill, the chorus is fuller than the verse, the crash opens instead of hitting every bar. The bass goes down to the low strings, an octave under the guitar, and leads into the chord changes.

    Chords and strings no longer cover the melody: the voicing now looks at it too. And under every line of lyrics, typed by hand or not, the app says where a word's stress falls wrong and where a closed vowel sits on the highest note.

25. 15 October · version 1.6.26

    The parts reach the players right

    Six new trades look at what leaves the app and reaches a real musician. The ukulele part no longer loses notes: every chord becomes the shape every ukulele player knows, C 0003, A minor 2000, F 2010. Strings and piano come out as a keyboard part, not as guitar tab, and the transition effects no longer end up on the drummer's stand.

    The MIDI opens in a DAW with every track on its own instrument and a name you can read, and the stems have clean names. Electronic music sounds electronic: a synth bass in house, the 808 and hi-hat rolls in trap, the pad pumping with the kick.

What it does not do, and why

Some things were left out on purpose: no cloud, no online collaboration, no splitting a finished track into stems, no plugins. Each would have needed the network, or pulled the app away from what it has to do well: catch an idea and make it sound.

Where we are

Today Motivetto records, transcribes, arranges, mixes and exports a whole song, and explains itself. The next step is the most important one: putting it in the hands of people who did not make it, listening to where they stumble, and from there getting ready to publish.

Read the manual