mlUtilities

A suite of common utilities to support machine learning data collation, inference and visualisation

Background

To validate datasets, prepare new datasets and generate fake data, this collection of utilities assists Audio analysis. Examples to demonstrate use of functions listed below include:

Examples (Assumed use of Google Colab)

Example of MNIST generation from ESC50

%cd /content
import os
if not os.path.exists('ml_utilities'):
    !git clone --quiet https://github.com/victorromeo/mlUtilities.git ml_utilities
if not os.path.exists('ml_utilities/sets/ESC50'):
    !git clone --quiet https://github.com/karolpiczak/ESC-50.git ml_utilities/sets/ESC50
%cd /content/ml_utilities
!pip install -q -r requirements.txt

from ml_utilities.esc50_utils import ESC50

esc50 = ESC50('/content/ml_utilities/sets/ESC50')
x_train, y_train, s_train, x_test, y_test, s_test = esc50.to_mnist(train_folds= [1,2,3,4], test_folds=[5], n_fft=1024, hop_length=512, cache_path='/content/ml_utilities/sets' )

Example of conversion from ESC50 to MNIST using Jonnor presets

Uses preset of 22050 Hz sample rate, 60 melfilter banks, 1024 fft window length, 512 fft hop, 12 data augmentations including time_stretching and pitch shifting

%cd /content

import os
import numpy as np

# Download and install prerequisites
if not os.path.exists('ml_utilities'):
    !git clone --quiet https://github.com/victorromeo/mlUtilities.git ml_utilities
if not os.path.exists('ml_utilities/sets/ESC50'):
    !git clone --quiet https://github.com/karolpiczak/ESC-50.git ml_utilities/sets/ESC50
%cd /content/ml_utilities
!pip install -q -r requirements.txt

from ml_utilities.esc50_utils import ESC50

# Load ESC dataset
esc50 = ESC50('/content/ml_utilities/sets/ESC50')

# Preprocess audio into spectrogram numpy arrays using mel scale
preprocessed = esc50.generate_jonnor_mel_spectrograms(cache_path='/content/ml_utilties/sets')

# Generate MNIST output
x_train, y_train, s_train, x_test, y_test, s_test = esc50.generate_jonnor_mnist(preprocessed, train_folds= [1,2,3,4], test_folds=[5])

# Save outputs
with open('/content/ml_utilties/sets/jonnor_esc50_1024_512_mel.npy', 'wb') as f:
    np.save(f, np.array(preprocessed))

with open('/content/ml_utilties/sets/jonnor_mnist_esc50_1024_512_mel.npy', 'wb') as f:
    np.save(f, np.array((x_train,y_train,s_train,x_test,y_test,s_test)))

Example of Spectrogram generation into ESC50/Audio/spectrograms folder

%cd /content

import os
import numpy as np

# Download and install prerequisites
if not os.path.exists('ml_utilities'):
    !git clone --quiet https://github.com/victorromeo/mlUtilities.git ml_utilities
if not os.path.exists('ml_utilities/sets/ESC50'):
    !git clone --quiet https://github.com/karolpiczak/ESC-50.git ml_utilities/sets/ESC50
%cd /content/ml_utilities
!pip install -q -r requirements.txt

from ml_utilities.esc50_utils import ESC50

# Load ESC dataset
esc50 = ESC50('/content/ml_utilities/sets/ESC50')

# Generate spectrogram images in ESC50/Audio/spectrograms folder
# destination can be specified using argument dest_path='/<fullpath>'
esc50.generate_spectrograms(n_fft=1024,hop_length=512,cmap='gray_r',raw=True,y_axis='linear',dest_exists_ok=True, image_exists_mode='skip')

Notebooks for worked examples

Module	Object	Operation	Description
ml_utilities.activation_utils	-	see file	Various ML activation functions, for reference and experimentation
ml_utilities.audio_utils	Audio	load_audio	Load an audio file
ml_utilities.audio_utils	Audio	load_noise	Generate audio noise (white, pink, blue, brown, violet)
ml_utilities.audio_utils	Audio	load_tone	Generate audio tones (sine, square, sawtooth)
ml_utilities.audio_utils	Audio	dump	Writes audio file in WAV format to disk
ml_utilities.audio_utils	Audio	trim	Truncates the audio signal
ml_utilities.audio_utils	Audio	gain	Modify signal amplitude (multiply, additive, subtractive, ...)
ml_utilities.audio_utils	Audio	resample	Resample the audio signal to a new sample rate
ml_utilities.audio_utils	Audio	time_stretch	Modify speed of audio, but not frequency
ml_utilities.audio_utils	Audio	pitch_shift	Modify frequency of audio, but not speed
ml_utilities.audio_utils	Audio	mix	Combine two equal length signals
ml_utilities.audio_utils	Audio	insert	Add a signal into another signal, matching target amplitude
ml_utilities.audio_utils	Audio	get_sample_box	Fetch a sample of audio in a time,frequency domain bounding box
ml_utilities.audio_utils	Audio	plot_waveform	Generate waveform, to display or disk
ml_utilities.audio_utils	Audio	plot_spectrogram	Generate spectrogram, to display or disk
ml_utilities.audio_utils	Audio	plot_linear_spectrogram	Generate linear y-axis spectrogram, to display or disk
ml_utilities.audio_utils	Audio	plot_mel_spectrogram	Generate mel scale y-axis spectrogram, to display or disk
ml_utilities.audio_utils	Audio	plot_mel_dB_spectrogram	Generate mel Db scale y-axis spectrogram, to display or disk
ml_utilities.audio_utils	Audio	get_spectrogram_array	Generate numpy spectrogram array in linear, or mel format
ml_utilities.audio_utils	Audio	get_spectrogram_array_uint8	Generate numpy spectrogram array with amplitude 0-255 in linear, or mel format
ml_utilities.audio_utils	Audio	get_statistics	Gets the max,min,mean,samples, sample_rate,duration, mean_norm and RMS amplitude of signal
ml_utilities.deepsqueak_utils	-	read_calls	Internal operation to parse a DeepSqueak Matlab mat detection file using h5py
ml_utilities.deepsqueak_utils	-	read_matlab	Internal operation to read a Matlab mat file using h5py
ml_utilities.deepsqueak_utils	DeepSqueak	get_audio	Fetches an Audio object for a specified DeepSqueak call
ml_utilities.deepsqueak_utils	DeepSqueak	get_call	Fetches a single call record from a DeepSqueak detection mat file
ml_utilities.deepsqueak_utils	DeepSqueak	get_calls	Fetches a count of all calls from a DeepSqueak detection mat file
ml_utilities.moth_utils	-	datetime_to_datetime_string	Creates a datetime string which conforms to AudioMoth standards
ml_utilities.moth_utils	-	parse_filename_to_datetime_local	Creates an tz aware datetime from an AudioMoth filename (local TZ)
ml_utilities.moth_utils	-	parse_filename_to_datetime_utc	Creates an tz aware datetime from an AudioMoth filename (utc TZ)
ml_utilities.moth_utils	-	parse_filename_to_timestamp	Creates a timestamp from an AudioMoth filename
ml_utilities.moth_utils	-	rename_files_local_to_utc	Modifies the filename to change timezone of audio from local to utc
ml_utilities.moth_utils	-	rename_files_utc_to_local	Modifies the filename to change timezone of audio from utc to local
ml_utilities.esc50_utils	ESC50	generate_jonnor_mel_spectrograms	Using a preset, creates a preprocessed data set containing mel spectrograms as numpy arrays from the ESC50 repository
ml_utilities.esc50_utils	ESC50	generate_jonnor_mnist	Converts the output of the generate_jonnor_mel_spectrograms preprocessing step, to generate an MNIST dataset from ESC50
ml_utilities.esc50_utils	ESC50	generate_spectrograms	Generates linear or mel spectrograms for the ESC50 repository
ml_utilities.esc50_utils	ESC50	get_audio	Fetches the Audio object for an ESC50 audio entry
ml_utilities.esc50_utils	ESC50	get_audio_files	Retrieves a sorted list of DirEntry objects to audio files, for the ESC50 repository
ml_utilities.esc50_utils	ESC50	get_categories	Retrieves all the categories from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	get_esc10s	Retrieves all the esc10 reference from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	get_filenames	Retrieves all the filename from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	get_folds	Retrieves all the fold values from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	get_meta_data	Internal operation to read the metadata from the ESC50 csv file
ml_utilities.esc50_utils	ESC50	get_meta_file	Attempts to detect the meta CSV file for the ESC50 repository
ml_utilities.esc50_utils	ESC50	get_takes	Retrieves all the take values from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	get_targets	Retrieves all the target values from the meta CSV file, optional unique flag
ml_utilities.esc50_utils	ESC50	to_mnist	Simplified one operation tranformation into MNIST format
ml_utilities.esc50_utils	ESC50	validate_meta	Iteratively checks the metadata to ensure that audio and metadata match
ml_utilities.noise_utils	-	blue	Generates n samples of blue noise (+6dB power/octave : +3dB Power density/octave)
ml_utilities.noise_utils	-	brown	Generates n samples of brown noise (-3dB power/octave : -6dB Power density/octave)
ml_utilities.noise_utils	-	pink	Generates n samples of pink noise (+0dB power/octave : -3dB Power density/octave)
ml_utilities.noise_utils	-	violet	Generates n samples of violet noise (+9dB power/octave : +6dB Power density/octave)
ml_utilities.noise_utils	-	white	Generates n samples of white noise (+3dB power/octave : +0dB Power density/octave)
ml_utilities.noise_utils	-	mean_square	Calculates the mean square of a signal (mean of square of values)
ml_utilities.noise_utils	-	normalise	Normalised a signal to ensure all values are within a predefined range

Dependencies include

h5py==2.10.0 numpy==1.18.3 librosa==0.7.2 mat73==0.40 parse==1.15.0 pytest==5.4.2 scikit-learn==0.23.0 sox==1.3.7 tqdm==4.46.0

victorromeo / mlutilities Goto Github PK

mlutilities's Introduction

mlUtilities

Background

Examples (Assumed use of Google Colab)

Example of MNIST generation from ESC50

Example of conversion from ESC50 to MNIST using Jonnor presets

Example of Spectrogram generation into ESC50/Audio/spectrograms folder

Notebooks for worked examples

Contents

Dependencies include

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent