data package

This package provides data and annotations with the corresponding api for 1d detection algorithm, which in my case is searching for wave signals of different tones.

Instructions:

  • All the data is reformed from some instrument records that are denoised and normalized to the same frequency.

  • First of all run the script data_preparation.sh to generate data and annotations. The script should be executed only once. It will not take effect if the data is already generated.

Example

sudo chmod +x data_preparation.sh
./data_preparation.sh

Module Name

Module function

wav_reconstruct

reconstruct wav files

data_generate

generate data and annotations

data_utils

api for data utilization
_images/paragraph.png

Submodules

data.data_generate module

Generate dataset and annotations with the reconstructed wav files. Some random noise is added to each data sample.

Caution:

  • This module should only be called by the script data_preparation.sh

Example

python3 wav_reconstruct.py

See also

Module data.wav_reconstruct

Warning

do not call this module manually!

data.data_generate.create_annotation(annotation_id)

Create annotations for the dataset using the COCO annotation structure.

data.data_generate.gen_signal(file_no)

Generate a new wave with random noise and random source wave

:param int file_no : the generated file name is wave_{file_no} :return:

data.data_generate.load_signal(name)

Load signal from a .wav file

Parameters

name (str) – the name of the wav file (without extension)

Returns

the signal data wrapped in numpy ndarray

Return type

numpy.ndarray

data.data_utils module

class data.data_utils.DataUtils(annotation_file=None)

Bases: object

Custom dataset helper class for reading or writing annotations

property dataset
display(text)

A test function that print text on screen

Parameters

text (str) – input text

Returns

return the input text

Return type

str

data.wav_reconstruct module

Reconstruct the wave to eliminate noise and normalize base frequency. This module will load name.wav files in ./wav folder and generate corresponding name_rc.wav files This is a longer explanation, which may include math with latex syntax \(\alpha\).

Caution:

  • This module should only be called by the script data_preparation.sh

Example

python3 wav_reconstruct.py

Note

an important source of 1d signal

See also

Module data.signal_generation

Warning

do not call this module manually!

data.wav_reconstruct.combine_overtones(frequencies, amplitudes, duration=2)

Generate a wave with particular overtones (particular amplitudes on particular frequencies)

Parameters

  • frequencies (array) – the frequencies of overtones

  • amplitudes (array) – the amplitudes of overtones

Returns

generated wave

Return type

array

data.wav_reconstruct.find_peaks(frequencies, amplitudes, width, lookaround)

Find the frequencies that are local extrema in amplitude

Parameters

  • frequencies (array) – frequncies of a signal

  • amplitudes (array) – amplitudes of a signal

  • width (float64) – range of searching the local extremum

  • lookaround (float64) – range of refining the local extremum

Returns

frequencies and amplitudes of the local extrema

Return type

tuple

data.wav_reconstruct.load_signal_wav(name)

Load wave as an python array from a wav file

Parameters

name (str) – the name of the wav file without extension

Returns

wave array

Return type

array

data.wav_reconstruct.reconstruct_wave(file)

Reconstruct the wave to eliminate noise and normalize base frequency

Parameters

file (str) – name of the wav file without extension

Returns

reconstructed wave signal

Return type

array

data.wav_reconstruct.sine_wave(frequency, duration)

Generate a unit sine wave with the given frequency and duration

Parameters

  • frequency (float64) – frequency of the sine wave

  • duration (float64) – duration time of the sine wave