Automatic Music Transcription

Automatic music transcription turns audio into structured musical descriptions such as note pitches and timings, instrument parts, MIDI, sheet music, or guitar tablature. Systems use spectrogram-based neural networks, sequence models, and audio-to-score pipelines, often strengthened by synthetic training data, source separation, or musical constraints such as pitch range and playability. Piano transcription is well served by established benchmarks, but dense polyphony, changing timbres, diverse genres, and assigning notes to the correct instrument or vocal part remain difficult. Progress also depends on better datasets and evaluation: scores can differ in notation accuracy and playback similarity, and the metric chosen can change which system appears strongest.