Python

Overview

Use the Python bindings for development, validation, descriptor extraction, and research workflows.

Installation

pip install OpenScofo

Minimal Example

from OpenScofo import OpenScofo

scofo = OpenScofo(48000, 4096, 1024)
scofo.load_score("score.scofo")

Reference Table

Constructor

Argument	Meaning
`sr`	sample rate
`fft_size`	FFT/window size
`hop`	hop size

Core methods

Method	Purpose
`load_score(path)`	load a score file
`process_block(audio)`	process one NumPy audio block
`activate_all_descriptors()`	enable every descriptor; ONNX inference requires a loaded model
`set_db_threshold(value)`	set silence threshold
`set_tuning(value)`	set A4 tuning
`set_current_event(event)`	force score position
`set_current_section(section)`	reset to the first event of a named section
`set_harmonics(value)`	set pitch-template harmonics
`set_pitch_template_sigma(value)`	set pitch tolerance
`get_live_bpm()`	return estimated BPM
`get_event_index()`	return current event index
`get_states()`	return score states
`get_pitch_template(freq)`	return pitch template
`get_block_duration()`	return block duration in seconds
`get_audio_description(audio)`	return descriptors for one block

Score Actions

Use the binding's score state/action APIs when you need host-side action handling. See the C++ integration for the underlying action structure.

Descriptors

Call activate_all_descriptors() before processing audio to enable YIN, MFCC/log-mel, chroma, ZCR, onset, extended technique, and ONNX descriptors. This initializes the required buffers and increases processing cost. ONNX inference runs only after a model is loaded.

scofo.activate_all_descriptors()
scofo.process_block(segment)
desc = scofo.get_description()

To select a subset later, use scofo.set_requested_descriptors([...]) with values from OpenScofo.Descriptors.

Common Description attributes include mfcc, chroma, onset, silence_prob, loudness, spectral_flux, spectral_flatness, harmonicity, db, rms, and power.

Complete Example

import librosa
from OpenScofo import OpenScofo

scofo = OpenScofo(48000, 4096, 1024)
scofo.load_score("score.scofo")

y, _ = librosa.load("audio.wav", sr=48000)
fftsize = 4096
hopsize = 1024

for pos in range(0, len(y) - fftsize, hopsize):
    block = y[pos:pos + fftsize]
    scofo.process_block(block)

Remarks

Use Python for offline validation and training workflows. For real-time performance, prefer Pd, Max, Csound, SuperCollider, or an embedded C++ host.