midiharmony

Fast, data‑driven harmony analysis for any MIDI file

Abstract

midiharmony provides fast, stand‑alone harmony analysis for MIDI files by comparing their chord and note relationships against a high‑quality database of extracted chord quads. This approach enables reliable detection of strong harmonic structure, musically coherent progressions, and potential inconsistencies, making it a practical tool for music‑AI pipelines, composition analysis, and large‑scale MIDI processing.

Install

Standard processing on CPU (default)

!pip install -U midiharmony

Accelerated processing on GPU

!pip install -U midiharmony[gpu]

Basic use examples

Using midiharmony is easy and requires just two lines of code :)

Analyze a single MIDI

import midiharmony

midi_harmony_dict = midiharmony.analyze_midi('Come To My Window.mid')

# This code will return a single dictionary with harmony stats
{'midi_path': 'Come To My Window.mid',
  'midi_name': 'Come To My Window',
  'total_chords_count': 2287,
  'bad_chords_count': 11,
  'grouped_chords_count': 1861,
  'total_quads_count': 1861,
  'unique_quads_count': 644,
  'harmonic_quads_count': 470,
  'harmony_ratio': 0.7298136645962733
}

Analyze MIDI folder(s)

import midiharmony

midi_harmony_dicts_list = midiharmony.analyze_midi_folders(['./midi_folder_1',
                                                            './midi_folder_2',
                                                            './midi_folder_3'
                                                           ])

# This code will return a list of dictionaries with harmony stats
# for all processed MIDIs from all specified folders
[
 {'midi_path': 'midi_folder_1/Bach Violin 2.mid',
  'midi_name': 'Bach Violin 2',
  'total_chords_count': 1089,
  'bad_chords_count': 71,
  'grouped_chords_count': 1045,
  'total_quads_count': 1045,
  'unique_quads_count': 974,
  'harmonic_quads_count': 867,
  'harmony_ratio': 0.8901437371663244},
 {'midi_path': 'midi_folder_2/Camping at Aylm.mid',
  'midi_name': 'Camping at Aylm',
  'total_chords_count': 417,
  'bad_chords_count': 7,
  'grouped_chords_count': 383,
  'total_quads_count': 383,
  'unique_quads_count': 369,
  'harmonic_quads_count': 326,
  'harmony_ratio': 0.8834688346883469},
 {'midi_path': 'midi_folder_3/Come To My Window.mid',
  'midi_name': 'Come To My Window',
  'total_chords_count': 2287,
  'bad_chords_count': 11,
  'grouped_chords_count': 1861,
  'total_quads_count': 1861,
  'unique_quads_count': 644,
  'harmonic_quads_count': 470,
  'harmony_ratio': 0.7298136645962733}
]

NOTES

• Most important value in each returned midi_harmony_dictionary is the "harmony_ratio"
• High harmony_ratio (>=0.75) indicates good harmony
• Exceptional harmony is indicated by harmony_ratio >= 0.9 and high unique_quads_count

Advaced use example

The code snippet below works great if you want to analyze a large MIDI dataset

# Import midiharmony, TMIDIX and tqdm
import midiharmony
from midiharmony import TMIDIX
import tqdm

# Create large MIDI dataset files list
filez = TMIDIX.create_files_list(['clean_midi'])

# Define a tine wrapper for fast MIDI multi-processing
def process(file):
    return midiharmony.process_midi(file, verbose=False)

# Use TMIDIX multi-processing wrapper (Linux)
# Or use joblib if you are on Windows
# Alternativelly, you can use a simple loop as well
output = TMIDIX.multiprocessing_wrapper(process, filez)

# Remove all empty dicts
output = [o for o in output if o]

# Analyze pre-processed MIDI dicts
all_harmony_dicts = []

for o in tqdm.tqdm(output):

    if o['quads']:
        res = midiharmony.analyze_processed_midi(o)
    
        if res:
            all_harmony_dicts.append(res)

# Final sort by harmony ratio
all_harmony_dicts.sort(key=lambda x: -x['harmony_ratio'])

# Save results to a nice json
TMIDIX.write_jsonl(all_harmony_dicts, 'all_harmony_dicts')

midiharmony API reference list

midiharmony.find_quads_fast_cupy
Count matching 4‑chord rows between two arrays using a GPU‑accelerated FNV‑1a hash.

midiharmony.find_quads_fast_numpy
Count matching 4‑chord rows between two arrays using NumPy on CPU.

midiharmony.get_trg_array
Load and cache the target harmonic‑quad array in NumPy or CuPy form.

midiharmony.process_midi
Extract chords, grouped chords, and unique 4‑chord quads from a MIDI file.

midiharmony.analyze_processed_midi
Compare extracted quads against the target database and compute harmony metrics.

midiharmony.analyze_midi
Run the full pipeline: process a MIDI file and evaluate its harmonic quality.

midiharmony.analyze_midi_folders
Batch‑analyze all MIDI files in one or more folders and return harmony reports.

midiharmony.helpers.get_package_data
Return a sorted list of packaged .npz data files with their resolved paths.

midiharmony.helpers.get_normalized_midi_md5_hash
Compute original and normalized MD5 hashes for any MIDI file.

midiharmony.helpers.normalize_midi_file
Normalize a MIDI file and write the normalized version to disk.

midiharmony.helpers.is_installed
Check whether a Debian package is installed using dpkg-query.

midiharmony.helpers._run_apt_get
Internal helper to run apt-get commands with consistent flags and timeout.

midiharmony.helpers.install_apt_package
Idempotently install an apt package with retries, optional sudo, and optional python‑apt.

Project Los Angeles

Tegridy Code 2026