Processing Options

The processing options available include:

  • Normalization: is applied automatically to the data after loading data, and additionally is re-applied after any of the processing steps.
  • Cropping. to remove portions of the waveform at the start and end of the recording that are not relevant.
  • Muting. to supress the recording along a selected time interval
  • Filtering: to remove frequencies that are outside of the frequency range of the bird song.
  • Concatenate WAV files. To concatenate files, then the files to be concatenated should all be in the same Species folder,  This task can also be used for appending descriptive voice recording at the front of a bird song recording.

Normalization is applied automatically.

Cropping, Muting and Filtering  processing can be found by right-clicking directly on the WAV file in the directory tree. Concatenation processing can be found by right-clicking on the Species folder in the directory tree (one level above the WAV files).

The Crop and Mute processes can also be started by drawing a time window with the mouse on the spectrogram, then clicking either the "C" or "M" keys on the keyboard.

Normalization

Normalization will multiply the entire waveform by a constant factor, so that the maximumamplitude falls within the range of +/- 0.7071. This also corresponds to -3 dB. The default normalization is -3dB, with an option to select -10dB.

The data is automatically normalized upon load. It is re-normalized again after any of the processing steps, since processing will likely have removed noise spikes or noisy frequencies that are outside the bird song frequencies and influenced the initial normalization during loading.

An example showing the difference between -3dB and -10dB normalization. Find the section of the recording that has the maximum amplitude, and you will note that the amplitude scale varies between +/- 0.7071 for -3dB and between +/- 0.3162 for -10 dB.

 

Cropping

Cropping is used to remove unnecessary portions of the recording that either precede or follow on after the bird song recording. If there are undesirable clicks or cultural noise in the recordings, then cropping can be used to remove them also. The preceding or trailing noise sections, will also affect the normalization if they are loud, so removing them will allow the normalization to be more effective.

Hint: if there is some undesireable noise at the beginning or end of the recording, then a quick way to remove this is to draw a time pick window with the mouse cursor that either starts or stops less than 1 second from the begining or end of the data, then click 'M' (for Mute). If a Mute is applied less than 1 sec from the data extremities, then the Mute window is automatically extended to the end of the data, and the data is removed. This is effectively doing a "Crop".

Muting

Muting is used to suppress internal portions of the recording that are undesirable. There are three options for applying muting.

  1. Select a time window in the recording, then the amplitude of the signal in this window is reduced to zero. There is no change in the length of recording.
  2. Force the muted Section to be 1 second long. This is similar to the first method, but if the section to mute is very long, then this option allows to reduce the muted section to be only 1 second long. The length of the recording will be changed.
  3. Remove the muted section. This will collapse the latter section onto the section before the mute.. The length of the recording will be reduced. This option is not recommended, since it wont be obvious that it has been applied, and timings of the bird song may be mis-leading.
Filtering

Filtering can be a destructive process if the filter range overlaps that of the desired recorded signal, so it needs to be applied with caution. Filter options include Low Cut, High Cut and Bandpass FIR filters, with selectable length. Generally, it is recommended to limit filtering to a Low Cut filter with a cut value of less than 250 Hz. If a steeper cutoff is required use a longer filter length.

Filtering can only be applied once to a file, so if a different filter range is desired, then apply it to the original dataset, and not one of the previously filtered datasets. Cropping can be re-applied to an earlier cropped file for a tighter crop range if desired.

Although low-cut (high pass) will have made it "nicer" to listen too, it may not make any difference as far "Bird Identification Software" is concerned. I have observed the Bird Identification sometimes works better on filtered waveforms, and sometimes it is worse ... which implies that it is best to upload the un-filtered (or minimally filtered) waveforms to internet archives, since once applied, filtering cannot be un-done.

If filtering is applied, be sure to keep the original earlier version of the WAV file.

It is only possible to save filtered waveforms back to a disk file if the applied low-cut frequency is less then 250 Hz. This is to avoid distributing waveforms that may have been over-filtered. Internally within the app, it is still possible to run Bird Detection on heavily filtered waveforms.

Concatentation

Concatenation processing can be found by right-clicking on the Species Name in the directory tree (3rd level). A window will pop-up showing all the WAV file that exist under this species, and you select the files and order for concatenation.

Concatination cannot be applied from the Year-Month-Day directory tree. It can only be applied from the Species level of the Order-Group-Species directory tree.

Note that for Concatenation, the data is concatentated "as is", meaning that any prior normalization that was applied is not removed before concatenating. As an example, if you concatenate different processed versions of the same signal, then they will probably have different relative amplitudes in the concatentated version.

Re-normalization and Clip

This process can be useful if the recording has a voice announcement at the end, or if using a pre-recording buffer that contains clicks or walking noise. After loading these types of file, these higher amplitude events may be dominant, and you can't see or hear the bird songs. Renormalization and Clip allows you to select a time window inside the recording around the bird song, and the normalization is re-computed over this window, and applied to the entire recording. This will boost all amplitudes and any amplitudes that exceeds 1.0 are clipped to 1.0, which effectively destroys the undesired higher amplitude parts of the recording.

This process will have produced a recording with some corrupted amplitudes, so this file is locked and it is not possible to save or do further processing on it. View and listen to this clipped recording, and use it to decide how to further process (crop, mute and/or filter) on the original version.

Spectral Processing.

Full Waveform Spectrum

The full waveform spectrum can be displayed by right-clicking on the WAV file in the directory tree and selecting "Show Waveform Spectrum". This spectrum computed from a 2048 sliding window, where the results of each sliding window is summed as it is advanced along the waveform. This gives an average spectrum of the entire waveform. At 48 kHz sample rate, the 2048 point window is 43 millisec long.

Spectrograms

Spectrograms are also computed using a sliding window, but now the results are kept as individual spectrums for each sliding window position. The spectrogram parameters can be adjusted; the default values are window length is 1024 samples (= 21 msec at 48 KHz) and hop size of 256 samples. The hop size is how far the window is advanced along the trace for each new computation.

These values are OK and keep the computation time low. But for bird songs that have rapidly change tone (such as Dark-necked Tailorbird), a 1024 point window smears the spectrogram in the time domain, and shorter values (window=512) can be considered. However, a shorter window gives a lower resolution in the frequency domain .... A much better alternative is to record the data at 96 kHz and use a 1024 point window. !