A finished stereo song
The client supplied ordinary full mixes—not multitrack sessions. Vocals, guitars, bass, keys, drums, and every other element were already combined inside each WAV file.
For a developer of musician-focused apps, I designed and built a local AI-assisted system that takes a supplied full-mix WAV, isolates the drums, detects kick, snare, and closed-hi-hat events, and delivers everything needed for DAW review—without uploading the music or operating Ableton.
The input is a normal finished song containing vocals, instruments, and drums. The system separates the drum performance from that mix, converts it into an editable MIDI first pass, and keeps every delivered file aligned from the beginning of the song.
The client supplied ordinary full mixes—not multitrack sessions. Vocals, guitars, bass, keys, drums, and every other element were already combined inside each WAV file.
Each song needed a drum-only audio stem, MIDI notes for kick, snare, and closed hi-hat, the untouched original mix, and simple instructions for importing the files together at time zero.
The prototype imported a full song, separated its stems, selected the drum stem, converted that audio to MIDI, and reorganized the same project for review.
Because the full mix, generated stems, MIDI, and organizational changes lived together, the wrong audio could be selected, a stem could end early, or an old selection could carry into the next action. It also required repeated computer takeover and carried a 20–30 minute planning estimate per straightforward song.
The metered core stages averaged 2:08 per song versus the earlier Ableton-assisted planning range of 20–30 minutes. Processing ran on existing local hardware, the songs were never uploaded, and there was no per-track software charge.
A separate validator checked the actual audio and MIDI rather than trusting a success label written by the processing stage.
Technical QA could not determine whether every kick, snare, or hi-hat label was musically correct, or whether the MIDI captured the drummer’s feel well enough to save editing time.
Automate the repetitive parts, expose uncertainty, and keep the decisions that require ears in human hands.
“Another excellent project with Ambrus. He is extremely detailed-oriented and knows his stuff when it comes to all things music production. Our project was more operations-focused this time and he did a great job. Will definitely work with him again.”