Hum to Song

Problem

When you have a hummed phrase, rough vocal take, or short recording and want to turn it into editable MIDI or a fuller song, cloud-based audio pipelines can feel unnecessary. hum2song is oriented toward local workflows: it accepts WAV, MP3, M4A, FLAC, or OGG input, normalizes it to 44.1kHz mono, and extracts the melody without sending recordings to third-party services.

How It Works

• Melody extraction: Uses Spotify's Basic Pitch to convert humming or singing into MIDI, covering C1 to C8, with polyphonic note detection and roughly 10ms timing resolution.
• MIDI cleanup: Cleans and structures the extracted notes to reduce false detections and rhythmic noise.
• Song generation: Optionally uses ACE-Step for local audio generation if it has been manually installed; otherwise it can follow a MIDI + SoundFont synthesis path without an AI model.
• Output options: Supports pop, rock, jazz, classical, and electronic styles; upbeat, calm, energetic, and melancholic moods; and target durations from 30 to 300 seconds.

Boundaries

The skill requires a local Python environment. ACE-Step is optional, needs roughly 4GB RAM for the base model, and may require about 4GB of locally downloaded weights when enabled; processing a two-minute song can take 2 to 5 minutes. Output quality depends on how clearly the melody is hummed, and complex harmonies may not be fully captured.