Moises

Moises uses deep learning models to split any song into up to 5 individual stems: vocals, drums, bass, guitar, and other instruments. The AI models are continuously retrained on diverse music datasets to improve accuracy across genres ranging from classical to electronic. Processing happens in the cloud, meaning users don't need powerful hardware — any smartphone or browser can submit tracks and receive separated stems within minutes. The quality of separation is strongest on professionally mixed and mastered tracks, with vocals and drums typically achieving the cleanest isolation.

The intelligent metronome automatically analyzes any uploaded track to detect its tempo (BPM) and beat positions, then generates a click track that locks precisely to the song's rhythm. This feature handles tempo fluctuations common in live recordings and older tracks recorded without a click, making it far more useful than a standard fixed-tempo metronome. Musicians can use it to practice along with songs at their original tempo or adjust the speed while keeping the click aligned, which is invaluable for learning complex rhythmic passages at a comfortable pace before building up to full speed.

Moises analyzes the harmonic content of uploaded tracks and displays chord symbols in real time as the song plays. The detection engine identifies major, minor, seventh, and extended chords, presenting them in a scrolling timeline synchronized with playback. This allows musicians to learn chord progressions without pausing to figure out each change by ear, dramatically speeding up the process of learning new songs. The feature works best on songs with clear harmonic structures and standard tuning, and provides a reliable starting point even for jazz and gospel music with more complex harmony.

Musicians can transpose any song up or down in semitone increments without introducing chipmunk effects or other pitch artifacts, which is critical for vocalists adapting songs to their range. The independent speed control allows tracks to be slowed down for learning difficult passages or sped up for advanced practice, all while maintaining natural tone quality. These two features work together so a musician can, for example, slow a guitar solo to half speed and shift it down two semitones to match their instrument's tuning — all in real time without re-processing the audio.

The lyrics transcription feature uses speech recognition AI to automatically generate text from vocal tracks, synced to the audio timeline. This is particularly useful for learning songs in unfamiliar languages, deciphering unclear vocal passages, or quickly creating lyric sheets for cover performances. The transcription appears alongside the waveform and chord display, creating a unified view of melody, harmony, and lyrics. Accuracy is highest for clearly enunciated vocals in English, with reasonable results for other major languages, though heavily stylized or screamed vocals may require manual correction.