Feedback That Reads Your Project, Not Just Your Audio
Drop in your Ableton .als project and TrackSensei reads what is actually inside it: how many tracks you have, how your arrangement is laid out over time, and how the structure builds. Audio-only tools never see this, because all they get is a flat rendered bounce.
Audio-only feedback stops at the bounce
When you render a track to a WAV or an MP3, you flatten everything you built into one stereo file. Every tool that only listens to that file (TrackScore.AI, Mix Check Studio, RoEx, and the rest) is working from the same flattened picture. They can hear the result, but they cannot see how you got there. They do not know you stacked eleven percussion tracks, or that your breakdown runs for thirty-two bars, or that the drop reuses the exact same elements as the verse. The bounce hides all of it. So their notes stay at the surface: this is too loud, the low end is muddy, the highs are harsh. Useful, but only half the story.
TrackSensei takes a different route. When you hand it an .als, it opens the project file itself and reads the XML inside it before it says a word about your sound. That means the feedback can talk about decisions you made in the arrangement, not just the texture of the final render.
What TrackSensei sees inside your .als
An Ableton .als file is a compressed XML document that describes your whole Live set. TrackSensei parses that XML and pulls out the structural facts a waveform can never reveal. Here is what it reads.
Track count
How many audio and MIDI tracks the set actually contains, plus return and group tracks. Eight tracks behaving like forty, or forty tracks where eight would do the job, both show up here.
Arrangement timeline
The position and length of clips along the arrangement, so it can see where your intro ends, how long the breakdown runs, and whether the drop lands where the energy needs it.
Section and structure
The shape of the track over time: intro, build, drop, break, outro. Reading clip placement across tracks lets it map your sections instead of guessing them from volume alone.
Density over time
How many elements are playing at once at each point in the track. That is how it spots a chorus that never opens up, or an intro so busy there is nowhere left to go.
None of those four facts survive a bounce to audio. They live in the project, and the project is exactly what TrackSensei reads.
Arrangement and structure feedback no waveform can give you
Once it knows your track count, your timeline and your density curve, the feedback changes shape. Instead of "the energy drops in the middle," it can say your second breakdown runs sixteen bars longer than the first while the element count stays flat, which is why the section drags. Instead of "the drop feels weak," it can point out that your drop is carried by the same four tracks as the build, so there is no lift when it hits. It can flag an arrangement that front-loads every idea in the first minute, or a track where forty layers are all fighting for the same frequency window because nothing ever drops out.
This is the kind of structural read a producer gets from a trusted second pair of ears, and it is exactly what audio-only analyzers cannot reach. It pairs naturally with the genre-aware notes on the AI track feedback page, and it sharpens fast when the structure is genre-specific, like a long hypnotic techno arrangement or a tighter, more sectioned house build.
Parsed in your browser, your project never leaves your machine
Here is the part most producers want to hear before they upload anything. The .als parsing happens entirely in your browser, on your own machine. The XML is read by client-side code, and no audio bytes and no project data are sent to a server. TrackSensei extracts the structural facts locally, then only the resulting feature summary is used to generate your feedback. Your actual project file, your samples and your ideas stay with you.
That is not a privacy promise bolted on after the fact. It is how the app is built. The heavy reading of your project runs where your project already lives, in front of you, which is also why it is fast.
100% client-side parsing
The .als is unzipped and its XML is read in the browser. No audio and no project data is uploaded. Only the extracted structural summary is used to write your feedback.
Works alongside your audio analysis
Reading the project does not replace listening to the sound, it completes it. TrackSensei still runs the full audio analysis on your render: loudness in LUFS, frequency balance, low-end behaviour, stereo image and the rest. The .als read sits on top of that. So you get the sonic picture (how it sounds) and the structural picture (how it is built) in one place, and the feedback can connect the two.
That is the combination that actually moves a track forward. When the audio analysis says the chorus feels small and the project read shows the chorus uses fewer active tracks than the verse, you stop guessing at the cause. Project plus sound, together, points straight at the fix. Upload the bounce on its own if that is all you have, or hand over the .als and get the whole story.
See what your project actually shows
Drop in your .als and read the structure audio-only tools miss. Free plan gets you 1 project and up to 2 track versions a month. No card.
Upload your .als