PyPAM

pypam is a python package to analyze underwater sound. It is made to make easier the processing of underwater data stored in *.wav files. The main classes are AcousticFile, AcousticSurvey and DataSet. The first one is a representation of a wav 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 wav 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.

Installation

Using pip distribution

pip install lifewatch-pypam

Using git clone

1. Clone the package

   git clone https://github.com/lifewatch/pypam.git

2. Use poetry to install the project dependencies

   poetry install

3. Build the project

   poetry build

News from version 0.2.1

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!

Quickstart

The package is imported as pypam. The wav files 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

Examples and documentation

See the documentation in readthedocs for a complete reference manual and example gallery.

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 features (vggish and compatibility with koogu and AVES)
• Add parallel processing options
• Add support for frequency calibration
• Support for reading detections

Cite