Audio-to-MIDI & Advanced Renderer

Upload a Audio for transcription-then-rendering, or a MIDI for rendering-only.

This application combines piano audio transcription with a powerful MIDI transformation and rendering toolkit. Based on the work of asigalov61.

1. Upload File

Input Audio or MIDI File

2. Results

MIDI Title

MIDI Description

Rendered Audio Output

MIDI Score Plot

Download Processed MIDI File

Output MIDI MD5 Hash

MIDI metadata summary

Transcription Settings

Note: This entire section is for audio-to-MIDI conversion. All settings here are ignored if a MIDI file is uploaded.

Audio Transcription Method

Choose 'General Purpose' for most music (vocals, etc.). Choose 'Piano-Specific' only for solo piano recordings.

General Purpose Piano-Specific

For stereo audio files only. When enabled, transcribes left and right channels independently, then merges them. Note: This will double the transcription time.

Enable Stereo Transcription

If checked, separates the audio into its component stems (vocals, drums, etc.) before processing.

Enable Source Separation

Transcription Profile Preset

Select a profile to auto-fill settings for different instrument types.For reference only; it is recommended to test and adjust for optimal results.

On-set Threshold

Sensitivity for detecting the start of a new note. Lower values will detect more notes (even faint ones), but may create false positives. Higher values are stricter and cleaner, but might miss subtle notes.

0 1

Frame Threshold

Sensitivity for determining if a note is 'on' or 'off'. Lower values will sustain notes longer, but can merge distinct notes. Higher values create shorter, more separated notes, but might cut off tails.

0 1

Minimum Note Length (ms)

Filters out notes shorter than this duration. Increase this to remove fast, noisy artifacts or clicks. Decrease it if the transcription is missing very short, staccato notes.

10 500

Minimum Frequency (Hz)

Ignores any detected pitches below this frequency. Increase this to filter out low-frequency noise like rumble or hum. Set it just below your target instrument's lowest note (e.g., ~80Hz for guitar).

0 500

Maximum Frequency (Hz)

Ignores any detected pitches above this frequency. Decrease this to filter out high-frequency noise like hiss or cymbals. Set it just above your target instrument's highest note (e.g., ~1200Hz for vocals).

501 10000

When enabled, the model actively looks for and emphasizes the start of each note (the 'attack'). Recommended for percussive or clear, rhythmic music. Disable for very smooth, legato music like vocal pads.

Infer Onsets (Boost Onsets)

When enabled, uses a secondary melody-detection algorithm to refine the main pitch contour. Highly recommended for most melodic content. Disable if you are transcribing non-melodic noise or complex polyphony.

Melodia Trick (Contour Optimization)

When enabled, allows a single note to have multiple, continuous pitch bends within it. Essential for transcribing vocals, slides, or vibrato-heavy instruments. Disable for clean, discrete notes like a standard piano.

Allow Multiple Pitch Bends

Audio-to-MIDI & Advanced Renderer

1. Upload File

2. Results

Transcription Settings

MIDI Transformation & Rendering Settings

8-bit Synthesizer Settings