basic-pitch

Install the current release of basic-pitch using pip. To update to the latest version, use the --upgrade flag.

Compatible Environments:

• OS: macOS, Windows, and Ubuntu
• Python: 3.7, 3.8, 3.9, 3.10, 3.11
• Note for Mac M1: Only Python 3.10 is currently supported; otherwise, use a virtual machine.

pip install basic-pitch

By default, basic-pitch does not install TensorFlow to save time and space. It installs platform-specific runtimes (CoreML on macOS, TensorFlowLite on Linux, ONNX on Windows). To explicitly install TensorFlow along with the default model inference runtime, use the [tf] extra.

pip install basic-pitch[tf]

Use the scripts located in the datasets folder to download and process datasets for training the Basic Pitch model. These scripts utilize Apache Beam pipelines to handle data processing.

When running these scripts, you can specify several keyword arguments to control data storage, processing methods, and partitioning.

Depending on which model format you intend to use, you may need to install additional dependencies. If you encounter warnings about missing runtimes, use the following commands:

• CoreML: pip install 'basic-pitch[coreml]'
• TensorFlow / TFLite: pip install 'basic-pitch[tf]' or pip install 'basic-pitch tflite-runtime'
• ONNX: pip install 'basic-pitch[onnx]'

To avoid the overhead of reloading the model for every file, load the Model object once and pass it to the predict() function inside your loop.

import tensorflow as tf

from basic_pitch.inference import predict, Model
from basic_pitch import ICASSP_2022_MODEL_PATH

basic_pitch_model = Model(ICASSP_2022_MODEL_PATH)

for x in range():
    model_output, midi_data, note_events = predict(
        <loop-x-input-audio-path>,
        basic_pitch_model,
    )

Import predict from basic_pitch.inference to run transcription directly in Python.

Signature: model_output, midi_data, note_events = predict(<input-audio-path>, [basic_pitch_model])

Returns:

• model_output: The raw model inference output.
• midi_data: The transcribed MIDI data derived from the output.
• note_events: A list of note events derived from the output.

Note: You can pass an optional basic_pitch_model object to avoid reloading the model on every call, which is recommended when running in a loop.

from basic_pitch.inference import predict
from basic_pitch import ICASSP_2022_MODEL_PATH

model_output, midi_data, note_events = predict(<input-audio-path>)

Use predict_and_save to automatically handle the generation and saving of various output file types (MIDI, WAV, NPZ, CSV).

from basic_pitch.inference import predict_and_save

predict_and_save(
    <input-audio-path-list>,
    <output-directory>,
    <save-midi>,
    <sonify-midi>,
    <save-model-outputs>,
    <save-notes>,
    <model-path>
)

Basic Pitch accepts audio files compatible with librosa.

Supported Codecs:

• .mp3
• .ogg
• .wav
• .flac
• .m4a

Processing Details:

• Channels: While stereo is accepted, input is down-mixed to mono for analysis.
• Sample Rate: All audio is resampled to 22050 Hz.
• Length: Can process any length, but very large files may be limited by disk space; streaming in windows is recommended for long files.

The predict_and_save function accepts the following arguments:

• <input-audio-path-list>: Directory paths for basic-pitch to read from.
• <output-directory>: Directory path to write results to.
• <save-midi> (bool): Control generating and saving a MIDI file.
• <sonify-midi> (bool): Control saving a WAV audio rendering of the MIDI file.
• <save-model-outputs> (bool): Control saving the raw model output as an NPZ file.
• <save-notes> (bool): Control saving predicted note events as a CSV file.
• <model-path> (str or pathlib.Path): Local path to load the model from (e.g., use ICASSP_2022_MODEL_PATH).

The command line tool generates a MIDI file transcription of audio files.

Basic Usage:

basic-pitch <output-directory> <input-audio-path>

To process multiple files at once:

basic-pitch <output-directory> <input-audio-path-1> <input-audio-path-2> <input-audio-path-3>

CLI Flags:

• --sonify-midi: Save a .wav audio rendering of the MIDI file.
• --save-model-outputs: Save raw model outputs as an .npz file.
• --save-note-events: Save predicted note events as a .csv file.
• --model-serialization: Specify a non-default model type (e.g., use CoreML instead of TF).
• --help: Discover all available parameter controls.