docs/core concepts/Listening

listening:
play it, don't describe it.

Everywhere else in these docs, tonee works from the structure of your set. Listening is the exception: it asks you to play something, captures what reaches your speakers, and answers from the audio itself. Nothing in your project changes.

01 /How a listen works

tonee can't start listening on its own and it can't listen quietly: arming the capture takes over the message box, so there is never a moment where it's recording and you don't know.

tonee asks

It tells you what it wants to hear and why ("play me the chorus", "let me hear the low end"), then arms the capture.

The composer turns red

The message box is replaced by a recording bar: the reason, a level meter, an elapsed timer, and Stop.

You play

Fire the clip, run the section, play the part in by hand. tonee is capturing whatever your speakers are getting.

You press Stop

Capture ends, the audio is analysed, and the result goes back to tonee, which picks the conversation up from what it heard.

Stop is yours

The timer counts up, not down. tonee waits as long as you take, up to a sixty-second safety cap, so a forgotten capture ends by itself.

Cancel means no

The ✕ discards the capture, and tonee treats it as a decision rather than a glitch. It won't re-arm and try again; it asks what you'd rather do.

Nothing is recorded into your project: no temporary track, no resampling input, no armed record button. tonee taps your computer's audio output, well outside Ableton. Your set is exactly as you left it.

02 /What it can hear

Because it captures your system output rather than an Ableton track, the source doesn't have to be Ableton at all. If it comes out of your speakers, it can be listened to.

A MIDI clip

Fire it and tonee hears the instrument playing it, the one thing reading the notes could never tell it.

The whole mix

Run the section. Balance, dynamics and width of everything together, not track by track.

Hardware

A synth or drum machine coming back through your interface counts as audio out. So does a guitar.

A reference

Play the track you're chasing and ask how yours compares.

You

Hum the line you want. It's audio like anything else.

Solo anything

Solo a track before you hit play and the capture is that track alone.

It hears everything, not just Ableton: a notification chime, a browser tab, a video playing behind your session all land in the same capture. Mute what you don't want analysed before you press play.

03 /What comes back

tonee doesn't get a recording to play back. It gets a measurement of one, so its answer is anchored to numbers rather than to a vibe.

Frequency balance

Level in six bands (sub bass, bass, low mid, mid, upper mid, air) in dBFS. This is what "it sounds muddy" gets checked against.

Dynamics

Peak, RMS and crest factor. How loud, how squashed, how much headroom is left.

Stereo image

Width, plus mid and side levels, when the capture is stereo.

Tone

Spectral centroid and dominant frequency: where the weight of the sound actually sits.

Anomalies

Flagged problems with a band and a severity: a resonant build-up, a hollow band, clipping.

A description

A short read of the audio in the language a mixing engineer would use, alongside the numbers.

It can't transcribe a melody or tell you the chords. What it's good at is the question you'd otherwise answer by squinting at a spectrum analyser: where the energy is, what's fighting what, and whether the thing you just changed did what you hoped.

04 /Analysing a clip instead

An audio clip already has a file behind it, so it doesn't need playing. tonee reads the source directly and returns the same analysis, with no capture step and no waiting for you.

Session clips

Any audio clip in a slot. You can ask for a window inside it rather than the whole thing.

Arrangement clips

Audio clips on the timeline work the same way.

The Analyze button

The Live session panel puts this on each audio track. One click sends the analysis request for you.

MIDI clips

There is no file to read. Fire the clip and let tonee listen instead. That's what listening is for.

05 /Permissions & privacy

On macOS, the first listen asks for Screen Recording permission. That's the system permission that covers capturing audio output, and macOS has no narrower one. On Windows it uses the built-in loopback device and there's nothing to grant or install. Capture runs at 48 kHz, 16-bit stereo.

Only while you see the bar

Capture starts when the recording bar appears and ends when it disappears. There is no background listening.

Measured on your machine

The bands, dynamics and anomalies are all computed locally by the app.

What leaves for the description

To describe the audio in words, up to the first 30 seconds is downmixed to mono, resampled to 16 kHz and sent to the AI backend with the numbers.

Not kept

The capture goes to a temporary file that is used for the analysis and then discarded. It is never added to your project.

This is the one time audio leaves your machine, and only when you press Stop on a capture you started. It's a different boundary from Project Sense, which never sends audio at all, and from voice input, which is transcribed on-device. Your project file and your sample library are not touched by any of it.

← Safe deletes Next · Plans & checklists →

listening:play it, don't describe it.