Auto Record is a Python script that records audio in WAV files. Instead of recording continuously, Auto Record creates multiple recordings, starting when noise is detected and stopping when there is a period of silence.
The WAV files can be accessed via a simple web service implemented in web_server.py
This project has the following scripts:
- auto_record.py: The noise triggered audio recorder
- web_server.py: Web server to access audio recordings
- wav2csv.py: Convert WAVE files to CSV and plot audio data
- play.py: Sample script to play audio files
- record.py: Simple example script to record WAVE files
The audio recorder requires pyaudio which requires the portaudio package. The web server requires fastapi
First, ensure you have pip and venv:
sudo apt install pip
sudo apt install python3-venv
Install the portaudio19 dev package:
sudo apt install portaudio19-dev
Create an isolated venv and install the required package
python3 -m venv venv
source ./venv/bin/activate
pip install -r requirements.txt
The recording script can be run directory from the command line:
python auto_record.py [ debug ] [ calibrate | disabled ]
- debug: print debug information
- calibrate: run an interactive audio calibration and exit
- disabled: start the recorder with recording disabled
source ./venv/bin/activate
python auto_record.py
The web server provides access to the recordings and the ability to enable and disable recording.
source ./venv/bin/activate
python web_server.py
Connect to localhost:3000 to see the list of recordings.
Calibrate the trigger threshold by running an interactive calibraiton process. The process needs 2 seconds of quiet to establish a baseline, then a 1/2 second noise at the level that should trigger a recording.
source ./venv/bin/activate
python auto_record.py calibrate
The auto_record.py script creates a data directory under the working directory. In this directory, it creates:
- .wav files - recordings
- .json files - information for each recording in json format
- record - if present, recording is enabled
- calibrated - created by the calibraiton process.
Start the web server providing access to the audio files. Note that there is no access control for this web server.
source ./venv/bin/activate
python web_server.py
In addition to being run directly from the command line, the scripts can be run with systemd for automatic start on boot up. There are issues with the default sound input device being different when run from systemd. The auto_record_log.txt file can be helpful in troubleshooting issues.
Note that on the RaspberryPi, the sound card needs to be configured in /etc/asound.conf. Run cat /proc/asound/cards to see the existing sound options.
For more info, see: [SOLVED] Help: ffmpeg and alsa - work as script not as systemd service problem
Change the User= and WorkingDirectory= lines in the auto_record.service and auto_record_server.service files. Install the files:
sudo cp auto_record.service /lib/systemd/system
sudo cp auto_record_server.service /lib/systemd/system
sudo systemctl daemon-reload
sudo systemctl enable auto_record
sudo systemctl enable auto_record_server
Start the services:
sudo systemctl start auto_record
sudo systemctl start auto_record_server
sudo systemctl status auto_record
sudo systemctl status auto_record_server
To remove the installed systemd configuration:
sudo systemctl stop auto_record
sudo systemctl stop auto_record_server
sudo systemctl disable auto_record
sudo systemctl disable auto_record_server
sudo rm /lib/systemd/system/auto_record.service
sudo rm /lib/systemd/system/auto_record_server.service
There are a few areas where the audio recorder could likey use improvement.
The auto_record.py script opens the default sound input device. When run from a desktop environment, this is usually configured correctly. However when starting from systemd for an automatic boot, the script may pick the wrong device. A few to pass in a device name may be useful.
The criteria for starting a recording is very simplistic. There may be better ways to detect when to start the recording.
WAVE files are large and require a great deal of storage. Any application that collects a significant number of hours of recording would benefit from a feature that converts WAVE files to MP3 files.
Once the auto recording script is configured to start on boot, it is inconvenient to run a calibration. You need to stop the service, run the calibraiton from the command line, and then restart the service. It would be useful to invoke calibration via the web server interface.
It also isn't clear that the environments running at boot time and from the command line are the same and would produce the same calibration values.
It is facinating to look at sound files graphically to understand the nature of the signals. The wav2csv.py module reads a wav file and writes a CSV file containing the audio samples. It has some additional routines to create simple plots of the audio data.
In order to analyze and plot sound file data, you need two additional modules not included in the requriements.txt:
pip install numpy
pip install matplotlib
To view the data as well as generate a CSV file, run the script with the -i argument to python:
python -i wav2csv.py <WAVE file>
After the CSV file is written, the sound data is in the numpy ndarray array:
python -i wav2csv.py data/2025-01-20_16:14:25.wav
Write file: data/2025-01-20_16:14:25.wav.csv
Done
>>> array
array([[ -15, -17],
[ -6, -8],
[ 18, 17],
...,
[-122, -122],
[-151, -150],
[-136, -136]], shape=(27000, 2))
Plot all audio data
plot_array(array)
Plot only the first channel of the array:
plot_array(array[:,0])
Noom in, plotting samples 9000 to 10000
plot_array(array[9000:10000,0])
Plot the volume of each block of audio:
python
>>> import wav2csv as w
>>> blocks = w.read_data_blocks("data/2025-01-20_16:14:25.wav")
>>> volume = w.get_volume_array(blocks)
>>> w.plot_stairs(volume)
