pypam
is a python package to analyze underwater sound.
It is made to make easier the processing of underwater data stored in audio files in chunks.
The main classes are AcousticFile, AcousticSurvey and DataSet. The first one is a representation of an audio file
together with all the metadata needed to process the data (such as hydrophone used).
The second one is the representation of a folder where all the files are stored for one deployment.
Here we consider a deployment as a measurement interval corresponding to the time when a hydrophone was in the water,
without changing any recording parameters.
The Dataset is a combination of different AcousticSurveys in one dataset. This is to be used if the user has made
several deployments and wants to process them with the same parameters.
Then pypam
allows to go through all the audio files from the deployments only with one line of code and store
the output in netCDF files, including metadata. The package can be used to analyze a single file,
a folder with files or a group of different deployments.
pypam
deals with the calibration directly, so the output obtained is already in uPa or dB!
All the documentation can be found on readthedocs
Warning This package is under active development, use with caution.
pip install lifewatch-pypam
- Clone the package
git clone https://github.com/lifewatch/pypam.git
- Use poetry to install the project dependencies
poetry install
- Build the project
poetry build
In version 0.2.0 we removed the detectors, because there are better maintained packages for these purposes. In version 0.2.1 we added the frequency-dependent calibration! In version 0.3.0 we changed quite some plot function to make them more flexible with other data products. We also changed the mean functions to median, so the output in db is the same than the output in uPa
The package is imported as pypam
. The audio file names must comply with the needs of
pyhydrophone to be able to read the datetime information.
The user can choose a window chunk size (parameter binsize, in seconds), so all the features / methods are applied to that window. If set to None, the operations are performed along an entire file.
The available methods and features are:
- Acoustic Indices:
- ACI
- BI
- SH
- TH
- NDSI
- AEI
- ADI
- Zero crossing (average)
- BN peaks
- time-domain features:
- rms
- dynamic_range
- sel
- peak
- rms_envelope
- spectrum_slope
- correlation coefficient
- frequency-domain
- spectrogram (also octave bands spectrogram)
- spectrum (density or power)
- 1/n-octave bands
- hybrid millidecade bands
- long-term spectrogram
- time and frequency
- SPD
Futhermore, there are several plotting functions
- SPD
- spectrum with standard deviation
- boxplots of time series aggregated data
- daily patterns
- LTSA
and some signal-based operations:
- Signal operations
- Noise reduction
- Downsample
- Band filter
- Envelope
- DC noise removal
PBP (PyPAM Based Processing) is a wrapper for PyPAM to make its use to compute hybrid millidecade bands more effective. PBP allows for local or cloud computing. For more information, you can find it in this repository.
See the documentation in readthedocs for a complete reference manual and example gallery.
In this collection of notebooks, some extra examples can be found of how to use pypam (under development).
Planned:
- Add function to generate files per included folder (too big deployments)
- Add options for the user to choose what to do when the blocksize is not multiple of the frames, and to deal with time keeping
- Add a logger that logs the code that was run and the warnings together with the output
- Add deep learning feature extraction (vggish and compatibility with koogu and AVES)
- Add parallel processing options
- Add support for frequency calibration
- Support for reading detections
Parcerisas, C. (2023). PyPAM: a package to process long-term underwater acoustics data in chunks (0.3.0). Zenodo. https://doi.org/10.5281/zenodo.10037826
We want to thank Stan Panier for the beautiful logo. This project was possible due to the funding on LifeWatch Belgium