How to get started

From song file to scored take

Pitcheon takes a song you already have, lifts the singer's voice out of it, reads the melody off that voice, and then draws your own voice over the melody while you sing along. Nothing you do in it leaves your Mac. This page walks through the whole thing once, slowly. By the end you will have prepared a song, listened to its parts, sung it, and read your own take back.

Before you begin

You need four things:

• A Mac. Apple silicon (macOS 14 or later) is the quick one; an Intel Mac (macOS 10.15 or later) works too and takes its time preparing songs.
• A song as a file you own: mp3, m4a, wav or flac. Pitcheon cannot fetch music for you, from anywhere.
• Headphones. Without them the microphone hears the song as well as you, and the scoring becomes meaningless.
• An internet connection, once. The first song you prepare downloads the listening models, about 500 MB. After that everything runs offline.

There is no account and nothing to sign up for.

81Install it

Download Pitcheon, open the file, and drag the bird into your Applications folder. The first time you open it, macOS asks whether the app may use the microphone. Say yes: that is how it hears you sing. It records nothing until you sing on a song, and what it records stays on your disk.

82Add your first song

Drop your song file anywhere on the machine, or press the two keys at the bottom left: songs opens your library, and the + key beside it asks for a file.

The library and add keys, at the bottom left of the machine

83Let it work

Preparing a song takes a minute or two on Apple silicon, and a good deal longer on an Intel Mac. The display tells you which of the four stages it is on. The machine is working, not stuck. In order, it separates the singer from the band, reads the melody off the singer, names the song's key, and writes down the words it hears. The first song also downloads the models, so it is the slowest one you will ever prepare.

The display while a song prepares

84Know what you are looking at

When the song is ready, the machine shows it in three places:

• The big glass is the roll. The grey blocks are the notes the singer sings, scrolling towards a fixed line. When you sing, your pitch is drawn over them as a line.
• The display strip below it shows the whole song as a waveform between two clocks. Click anywhere in it to jump there.
• The play key and the volume knob do exactly what they say.

85Choose what you hear

Preparing split the song into parts, called stems, and three keys at the bottom left choose which one plays. One key is down at a time.

The stem keys. A lit lens means that part is playing

Start with the singer alone and just listen: this is the melody you are about to learn. Switch to the record to hear the song as it was. When you are ready to sing, put the band down, so the instruments back you and the original singer stays out of your way.

86Sing

Put your headphones on, press play, and sing. Your voice appears on the roll as a line drawn over the grey note blocks: on the block is on pitch, above it is sharp, below it is flat. The words appear under the roll as they come. Do not chase perfection on the first pass. The two numbers on the ledge keep the score.

How much of your singing was on pitch, and your longest run without falling off

87Fix the words if they are wrong

The words were written down by Whisper, a model trained on speech, and singing is harder to hear than speech: vowels stretch, consonants disappear, the band plays over everything. Apple silicon listens with the medium model; an Intel Mac starts with the smaller, quicker one and hears a little less. Either way, expect a few wrong words, and expect fixing them to be easy. Press edit lyrics and the editor opens beside the roll.

Editing is just typing

The whole lyric is one piece of text. Retype a word, break a line with Enter, join two lines into one, reflow a verse. While you type, every word keeps its place in the song: a corrected word inherits the timing of the word it replaced, so changing "bridge door" to "fridge door" fixes the word and moves nothing. Only brand new words need a place, and Pitcheon spreads them into the room between their neighbours when you save.

Every line knows when it starts

Left of the text sits a gutter of times, one per line. Click a time and the song plays from just before it, so you can check a line against the singing. If a line starts in the wrong place, play to the moment it should start and press the mark beside its time: the line is stamped to the playhead. A stamped line counts as placed by you, and nothing the machine does later will move it.

Words are boxes under the roll

While the song plays, each word is drawn as a box under the roll. Drag a box to move the word, drag its edge to stretch or shorten it, exactly like arranging clips. Where somebody is singing and no words are on it, the roll shows the empty stretch: click it to start a new line right there.

When the timing is a guess, retime

A line you typed into a gap arrives with no timing of its own, so it goes in at a guess. The retime key asks the machine to listen again for exactly those lines and settle them into the singing. It changes no words, and it leaves every line you placed by hand alone.

When the words are hopeless, rescan

The rescan key starts over, and asks on the glass first:

The rescan ask. A bigger model hears more singing and takes longer

Everything on screen is replaced by what the new model hears, so rescan before you start correcting, not after.

Save keeps it

The lamp beside the save key lights while there are unsaved changes and goes dark when the words are kept. The lyrics section has the editor itself to try, wired the way the app wires it.

88Read your take back

Every run with the microphone on is kept as a take. Press takes, open one, and it plays back with your pitch drawn as you sang it. The list below the roll names the exact moments that went off, worst first. Click one and the machine jumps there, so you can drill the two seconds that need it instead of repeating the whole song.

The take inspector's trouble list. Cents are hundredths of a semitone

89Balance what you hear

Under the take inspector sit three knobs. The singer sets how loud the original vocal is, my voice sets how loud you are in your own ears, and voice sync nudges your recorded voice earlier or later if it plays back out of step. If a take says it was recorded late, one key, sync with mic, measures that lag and corrects for it from then on.

If something seems wrong

• No sound. Check the volume knob, and check which stem key is down. The singer alone can be very quiet in parts where nobody sings.
• It cannot hear you. Open System Settings, then Privacy & Security, then Microphone, and allow Pitcheon.
• Preparing takes forever. The first song carries the model download. After that, a couple of minutes per song on Apple silicon is normal, and twenty on an Intel Mac is too.
• Your voice sounds late when a take plays back. Open the take and press sync with mic.

That is the whole machine. Everything it makes, the stems, the words, the takes and the scores, stays in a folder on your Mac, and nothing leaves it. When you are ready: download Pitcheon and bring a song.