Automatically make a Decent Sampler Lib by a folder of samples

[Nicola Landro]

I think that Decent Sampler is very good sampler, that work also for Linux, so I made this script that can help to have a base plugin by a folder of samples. In this way it is easy to port our libraries of sound also for this sampler and it is also a good starting point for who want to create a new one. More info about the project [there](https://gitlab.com/nicolalandro/samples_to_decent_sampler_lib/-/tree/main).

requirements: python and librosa (installed by pip)

import os
import numpy as np
import librosa
import scipy
import math

SAMPLES_FOLDER = "Samples"
OUTPUT_FILE = "preset.dspreset"
BASE_SCRIPT_START = """
<?xml version="1.0" encoding="UTF-8" ?>
<DecentSampler>
    <ui bgImage="background.png" width="812" height="375">
        <tab name="main">            
            <label x="365" y="0" width="100" height="50" text="Reverb" textColor="FFFFFFFF"/>
            <control x="360" y="30" parameterName="Reverb" type="float" minValue="0" maxValue="1" textColor="FF000000" value="0.5">
                <binding type="effect" level="instrument" position="0" parameter="FX_REVERB_WET_LEVEL"/>
            </control>
        </tab>
    </ui>
    <groups>
        <group>

"""
BASE_SCRIPT_END = """
        </group>
    </groups>
    <effects>
        <effect type="reverb" wetLevel="0.5" />
    </effects>
</DecentSampler>
"""

def create_sample_line(note_number, file_path):
    return f'            <sample rootNote="{note_number}" path="{file_path}" />'

def main():
    audio_samples_names = os.listdir(SAMPLES_FOLDER)
    
    samples_lines = ""
    max_iteration = len(audio_samples_names)
    for i, sample_name in enumerate(audio_samples_names):
        print(f'{i+1}/{max_iteration}')
        path = os.path.join(SAMPLES_FOLDER, sample_name)
        
        midi_number = audio_to_note(path)
        
        sample_line = create_sample_line(midi_number, path)
        samples_lines += sample_line + 'n'
                
    lib_text = BASE_SCRIPT_START + samples_lines + BASE_SCRIPT_END
    with open(OUTPUT_FILE, 'w') as f:
        f.writelines(lib_text)
 
def audio_to_note(path):
    print('t', path)
    
    audio, sample_rate = librosa.load(path) # default semple rate 22500 that reduce original but it is more efficient for computation
    fundamental_frequencies, _, _ = librosa.pyin(audio, fmin=librosa.note_to_hz('C2'), fmax=librosa.note_to_hz('C7'))
    
    filtered_fundamental_frequencies = list(filter(lambda x: not np.isnan(x), fundamental_frequencies))
    frequency_mode = scipy.stats.mode(filtered_fundamental_frequencies).mode[0]
    frequency_median = np.median(filtered_fundamental_frequencies)
    frequency_mean = np.mean(filtered_fundamental_frequencies)
    print("ttmode:", frequency_mode, "median:", frequency_median, "mean:", frequency_mean)
    
    predicted_frequency = np.mean([frequency_mode, frequency_median])
    raw_midi_number = librosa.hz_to_midi(predicted_frequency)
    note_name = librosa.midi_to_note(raw_midi_number)
    midi_number = int(math.ceil(librosa.hz_to_midi(librosa.note_to_hz(note_name))))
    print('ttfrequency:', predicted_frequency, 
          'midi number:', f'{midi_number}({raw_midi_number})', 
          'note:', note_name)
    return midi_number

if __name__ == "__main__":
    main()

Usage:
1. create a folder
2. put background.png file (width: 812, height: 375)
3. create a folder Samples with inside all the .wav recorded
4. copy generate_decentsampler.py inside the folder
5. run python3 generate_decentsampler.py
6. it create the ‘preset.dspreset’ file that is the file to load into your decentsampler
7. you can zip the folder and share on pianobook!

That script is not perfect, but is a good starting point and maybe with the help of the community can become better and better.

[Jorge Serrano]

Thank you for sharing Nicola!

I will take a look to you script.
I think that could be useful not only for Linux users, but also for MacOS or Windows users too.

Kind regards,

Jorge

[Author]

Posts