InspectNoise is a real-time sound meter with A glance into single-board computer. The main idea is to have a low-budget device to monitor and analyze acoustic noise. To do this we used a Kinobo USB microphone named Mini AKIRO.
We have also used a calibrated sound level meter and temperature, humidity, atmospheric pressure, PM10, PM2.5, PM1.0 sensors to create a dataset, and through machine learning models we have significantly improved accuracy performance (in terms of dB SPL) of the aforementioned microphone.
All tests are based on Raspberry Pi 2 model B. This projects is totally open-source (license MIT) and you can find the source code on GitHub. Following, the instruction step-by-step to use InspectNoise:
REQUIREMENTS
The file requiremets.txt contains libraries that need to be installed.
pip install -r requirements.txt
Usage and FLAGS
To know available flags use:
python3 inspect_noise --help
| Flag | Description |
|---|---|
-c/ --collect |
collect data as Min, Max and Avg |
-l/--log [file] |
log of recorded data as text file. [file] is an optional params, if not specified, program will save log on a file (name: log.log) in ~/.inspectNoise/ hidden folder |
-r/--record threshold |
record audio when dB are on average higher than the specified threshold. Timestamp used as name of audio file. |
-s/--seconds seconds |
specify recording time |
-sh/--showindex |
it is useful for know index of input audio devices available |
-se/--setindex index |
used to set index of input microphone (writing on configuration file) |
-f/--format [mp3, wav, ogg] |
define format of output record. It can be used only with –record |
-to/--thrashesoutput |
used for debug or utility, when specified permit to not show output on terminal |
-ca/--calibrate |
used to load machine learning model that try to predict db read by a calibrated phonometer. The calibration tries to predict the values read by the [UT351/352] (the calibration tries to predict the read values of the UT321 sound level meter) sound level meter |
Note
Please pay attention when use –calibrate flag, because, as reported in requirements.txt file, in our Raspberry the version of scikit-learn is 0.21.3, and the model was printed using the same version of this library.
Extension
After first use the tool creates a hidden folder with name “.inspectNoise” in ~ (user dir). In this directory will be created (after using flag –record for the first time) a new directory with name gathered_mp3. Within this folder will be saved recorded data in sub-folder with date of recoded day.
In the project directory are included 2 more utils file.
- First is plotter.py that can be ran separately to create graphs starting from a log file (created with flag –log). The output will be located in “plot_data” folder, located in “.inspect_noise” folder.
python3 plotter.py file.log my_dpi [threshold]
- Second is audio_merger.py that it can be used to merge audio in a specific folder.
python3 audio_merger.py input_dir_path export_file format
Created Dataset
Using this library for the environmental noise monitoring with a microphone and a SPL meter three different datasets were created. The following datasets are located in subdirectory named “dataset” and stored in simple csv files.
- DB_dataset_first_model.csv. This dataset contains microphone and SPL meter samples corresponding to the same second.
- DB_dataset_canarin_second_model.csv. This dataset contains microphone, SPL meter and environmental (like temperature, humidity, pression, PM ecc.) samples corresponding to the same second.
- DB_dataset_canarin_third_model.csv. This dataset contains microphone, SPL meter and environmental (like temperature, humidity, pression, PM ecc.) samples corresponding to the same minute.
In the first and second datasets are used data from a month of sampling, while in the third are used two months’ data.
Citation
Monti, L.; Vincenzi, M.; Mirri, S.; Pau, G.; Salomoni, P. RaveGuard: A Noise Monitoring Platform Using Low-End Microphones and Machine Learning. Sensors 2020, 20, 5583. DOI