Play a song to a trained musician and, without thinking hard about it, they can tell you a lot: roughly what key it's in, how fast it's going, where the chorus lands, what the bass is doing under the vocal. That fluency is what "understanding a song" means, and it's exactly the kind of understanding we set out to give a machine.
Here's what an AI ear can hear from a single track.
The musical layer
- Key and chords. Detect the harmonic content: what key the piece is in and how the chords move through it.
- Tempo and beat. Get the global tempo, the beat times, and how steady (or not) the tempo is across the track.
- Structure. Map the song's sections: intro, verse, chorus and so on, along with beats, downbeats, and meter.
- Notes. Turn a pitched performance into discrete musical notes you can export as MIDI.
The sources underneath
A finished mix is many things layered into one waveform. Understanding it often means pulling those layers apart.
- Stems. Separate a song into its sources: vocals, bass, drums, and more, as individual audio files.
- Isolate or remove an element. Get "just the vocals," or "the song minus the vocals," as a processed file.
The words and the sounds
- Speech. Transcribe with word-level timestamps, generate captions, tell speakers apart, and translate to English.
- Sound events. Detect and tag what sounds are present across any kind of audio, not just music.
See it, don't just read it
Numbers are hard to feel. Analysis results can be rendered as interactive charts and tables: a chord timeline, a tempo curve, a feature table, so you can actually see the shape of what the ear found. And when you want to take the work with you, export is real files and data: MIDI, stems, processed renders, JSON, and CSV.
One pass, many answers
The point isn't any single one of these capabilities. It's that they compose. A song can come in and go out understood along several dimensions at once: its harmony, its rhythm, its structure, its sources, its words. That's what it means to give a machine an ear: not one clever trick, but the fluency to hear a track the way an attentive listener does, and hand you back something you can read, see, and build on.
Every capability named here: key/chords, tempo/beat, structure, note transcription, stem separation, element isolation, speech transcription, sound tagging, interactive charts, and file/data export: all live today. Export is files and data only (MIDI, stems, renders, JSON, CSV), not formatted reports. See Features.