Tools for quick classification of acoustic recordings
Analysing and classifying underwater acoustic signals can be a challenging task. To simplify the process, we have developed standardised procedures and dedicated software. These innovative tools help researchers and operators create detailed spectrograms and visual analyses, enabling them to extract critical information quickly and efficiently.
Most of the following procedures are for MS Windows machines, but we are working on extending the packages to OSX as well!
What our processes offer:
Impeccable organisation: forget about messy audio files! Our procedures standardise file naming, ensuring a clear and easily interpretable timeline.
Guaranteed reliability: an automatic audio file integrity check (WAV) ensures that data is error-free and ready for analysis.
Customised spectrograms: generate spectrograms with a standard 15-minute time window optimised for your research needs. Adjust the basic parameters for optimal signal visualisation.
Overview: create extended timelines (as an option) to view longer periods of time in a single image, making it easier to identify long-term patterns.
Intuitive classification: visually classify the content of spectrograms to identify classes of signals of interest. Generate data tables ready for graphical representation and statistical analysis.
MOST of the procedures and software described in this page were drafted and coded within two research projects (TursioNet and CLAPS) within PELAGOS initiative projects funded by the Prince Albert II of Monaco Foundation.
File naming conventions: towards a shared standard
In the field of underwater acoustics, several file naming conventions have been proposed, each with its own rationale. After careful consideration, we have chosen to adopt a widely recognised format, in line with that used by devices such as SoundTrap:
- File name structure: the file name consists of two segments separated by a full stop (.). The first segment includes identifying information such as the recorder ID, distribution campaign or other relevant metadata. The second segment shows the date (YYMMDD) and time (HHmms) of the start of the recording.
- Format and extension: All files are in uncompressed WAV format with the extension .WAV.
Examples of compliant file names:
blackboard-44D45.241211090655.wav
test_33D566.250214153411.wav
You can easily find your preferred file renaming utility if you need to rename an entire batch of files and/or if your recordings do not yet comply with this standard. We recommend BulkRenameUtility, which has been tested, is free for personal use, and is easy to use and highly configurable.
Always keep in mind that changing file names is a critical operation. If you make mistakes, you could lose crucial information.
This standard (acb_acb_.YYMMDDhhmmss.wav) is mandatory.
When you have a list of files with SOUND SIGNALS
When you have a list of audio files, it’s time to download and install a set of tools that will be necessary to manage the files and extract the information you need. All links (and projects) are external, unless otherwise specified, so we have nothing to do with their fantastic developers.
You only need a few tools:
- SoX, the Swiss Army knife of audio processing. This is a set of tools for sound processing. You can find it here.
Once you have installed SoX, remember to add its path to your system variables. (If you have no idea what we are talking about, ask a friend for help). - The ffmpeg library to process part of the spectrograms. The Windows libraries for ffmpeg are here. Download the necessary package (three executable files), unzip it and put all the files in the same folder as SoX
(SoX should already be in your system’s search path, so your computer will also find the ffmpeg files when needed). - Download our Spectrogram Generator interface from HERE (go to the latest version!).
This is a simplified interface that helps you work with SoX. A Windows executable will be downloaded that installs the necessary libraries. If you get any security warnings, just authorise the file to run without worrying.
Keep in mind that the generator will only work if the Sox and ffmpeg libraries are already installed on your system.
Spectrogram Generator
SELECT FOLDER:
Only select folder name, without opening it.
Once you have installed the Spectrogram Generator, run it and you should see something like this.
This software is written in Python.
You will need to select the folder you want to operate on. Press FOLDER–> SELECT FOLDER (and include all subfolders if you want to).
Conder that you need to select a folder WITHOUT OPENING IT, so do NOT double-click on its name. Click just once.
VERIFY AUDIO FILES
Corrupted files will be renamed in their extension (from .wav to .wavcorr).
Press VERIFY AUDIO FILE to check if all audio files are consistent. This operation is needed in order to rename corrupted files (mainly due to battery problems) excluding them from the further processing.
Sometimes audio files have a glitch at their beginning. This might be related to errors in writing the first bytes of each file.
The FILE TRIM function can be used to cut the first milliseconds of each file, correcting the problem when performing statistical analysis of the files (like checking max levels).
It easier if files have all the same duration
Considering that you will generate a spectrogram that is always the same size in pixels, it easier if we keep the horizontal axis always at the same duration. We like 15′ (900 seconds), shown on a 4096 pixel axis over a 1024 y axis.
If you have files with longer duration (for example 1 hour each) you can get 4 files of 15′ each automatically using the CHUNK SIZE (minutes) function. When you ENABLE this function all processed files will be chunked in smaller parts of 15′ each (if you set the minutes to 1… you will get 1′ chunks. 60 chunks per each hour). All chunks will be saved separately in a dedicated folder, and to their name will be added a trailing CNK_.
And now?
It is time to GENERATE SPECTROGRAMS
SO ZUMMARAIZING:
- 4096 time resolution x
- 1024 frequency resolution y
- Select SPECTROGRAM SCALE LINEAR
- CHUNK SIZE at 15′ if your files are longer than that
- SPECTROGRAM PALETTE set at RAINBOW (we like it)
- press GENERATE SPECTROGRAM and get a coffe
Spectrogram generation can be a rather long process
There will be a long long delay from your click to the SUCCESSFULL message.
You should see a progress bar, slowly increasing depending on the total number of files to be processed.
From a list of Spectrograms to a table
The spectrograms will be created in a new folder named SPECTRA. The folder SPECTRA will be created under the same folder containing the audio files. If INCLUDE SUBFOLDERS is selected in the folder selection tab, then each sub-folder will have its new SPECTRA folder, one per folder. In the TursioNet repository, all folders already contain their SPECTRA folder with spectrograms already generated. Now it is time to download one SPECTRA folder, with spectrograms in it, and check their contents.
Oue Helper for Human-based classification
Download and install our helper. It will make things much easier. It is called SPECTROGRAM VIEWER
DOWNLOAD FROM HERE (go to the latest version!)
Once installed our software works this way:
- Tell the system who you are, so each line of the resulting tables will retain the identity of who carried out that task;
- Select a folder containing spectrograms. SELECT THE FOLDER AND NOT THE FILES!! Filenames must be consistent with the rules of the above Spectrogram Generator.
- Define and Use you specific list of tags while browsing your spectrograms and when searching for signals. Your list of tags is a sort of configuration that can be saved and shared.
- Go, check the spectrograms, quickly look at them and do the classification, get tables with the results.
Write or confirm your name.
select THE FOLDER that contains spectra with consisting filenaming
Add the tags that you will be using, or load the ones of your workgroup
Tags will be standardized in your workgroup (?), you can add the quick key (one letter or number) that you can use on your keyboard. A button can also be used if you prefer a mouse-click of a touchscreen.
It’s time to browse the spectrograms!
Use the right/left arrows to go to the next one, click on the keys if you see signals that match your tags and interests, two clicks=2, three clicks=3 (this is a score you can give this particular spectrograms you are watching), SHIFT+CLICK=2 again. SHIFT+KEY rolls back your scoring.
When you have finished with one image just go to the next. You can ADD A NOTE, if you need to.
EXPORT PREVIEW takes you to the table you populating. so that you can check what’s going on.
You can PAUSE the classification using the EXPORT SESSION (and later reimport using IMPORT SESSION), so your progress can be divided into different sessions…
EXPORT FILE allows you to save the table, so that you will be able to do some statistics over it.
