SophieXin9636 / AIoT-Automatic-Sound-Recognition

Clip audio algorithm, and sound recognition implemented by combination of STFT and CNN training model.

Sound Analysis

By Sophie Shin

Introduction Analysis Step

Install the dependencies

pip3 install -r requirements.txt
sudo apt-get install ffmpeg

or

sudo bash env.sh

Usage

Step0. Convert .amr(or others) file into .wav

Method 1 Only Convert file extension

ffmpeg -i sound3.amr -ar 22050 sound3.wav

Method 2 Convert file extension and edit particular seconds of Audio

python3 editAudio.py -r <beginSec>:<endSec> -i <inputAudioFile> -o <outputAudioFile>

Take this command as an example,

python3 editAudio.py -r 5:20 -i soundfile/0601/30s/sound1.wav -o s.wav

Step1. Split Frame from a wave file

It will split frames automatically when it detects the sound.
Detection standard: dBFS > -15 (max dBFS is 0), and trim sound interval = [sec-50, sec+50]

python3 editAudio.py -a -i <inputAudioFile> -p <outputAudioPath>

Take this command as an example,

python3 editAudio.py -a -i soundfile/0715_pat_correct.wav -p real_training_data/correct

Step2. Create Spectrogram by particular Fourier Transform Method

• Create Spectrogram

python3 spectrogram.py -T <Transform> -i <inputAudioPath> -o <saveImagePath>

Take this command as an example,

python3 spectrogram.py -T stft -i 0727_data/testing/ -o 0727_data/testing/
python3 spectrogram.py -T stft -i real_training_data/incorrect/ -o real_training_data/incorrect/img

Output

Input File ：  sound/
Output Path：  img/stft/

STFT Spectrogram Has Created!

Step3. CNN Training

python3 cnn.py -i <CorrectImagePath> <InCorrectImagePath>

Take this command as an example,

python3 cnn.py -i ./real_training_data/

or Binary Classification

usage: python3 binary_cnn.py --correct <CorrectImagePath> --incorrect <CorrectImagePath>

Example

python3 binary_cnn.py --correct 0727_data/training/img/stft/ --incorrect 0727_data/incorrect/img/stft/

Step4. Validation

Validate a Audio file

usage: python3 validation.py -i <AudioFile>

Take this command as an example,

python3 validation.py -i ./0727_data/testing/frame_0.wav

or Validate numerous Audio files

usage: python3 test.py --path0 <IncorrectAudioPath> --path1 <CorrectAudioPath>

Take this command as an example,

python3 validation.py --path0 ./0727_data/incorrect/ --path1 ./0727_data/validation/

Others

Step0 to Step4

$ bash run.sh

Wave plot

usage: python3 wave_plot.py -r <begin>:<end> -i <inputAudioFile> -o <outputAudioFile>

Take this command as an example,

python3 wave_plot.py -r 1:5 -i sound_data/0727_record/1.wav -o example.png