Wavelet-CWT

Python helper for computing and visualising the Continuous Wavelet Transform (CWT) on 1-D signals such as audio waveforms.
And for computing the Discrete Wavelet Transform in regards to Neural Network Compression and inference after compression.

Powered by NumPy · SciPy · PyWavelets · Matplotlib · Torch · Transformers.

---

Features

• Robust path-handling (file or directory) & WAV loading (int/float, stereo→mono)
• Fast CWT via PyWavelets, plus scale→frequency conversion in Hz
• scalogram plot with log-frequency axis and colour-bar
• One-file turnkey CLI or import as a library
• Pure-Python, no compiled extensions

---

Quick start

# 1 – clone the repository
git clone https://github.com/your-org/Wavelets.git
cd wavelet-cwt

# 2 – create & activate a fresh venv  (optional but recommended)
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

# 3 – install dependencies
pip install -r requirements.txt

# 4 – run the demo cwt scalogram of audio file
python cwt_scalogram_audiofile.py samples/Katydid.wav --wavelet cmor1.5-1.0 --seconds 5 --fft

Comments

To run: python cwt_scalogram_audiofile.py samples/Katydid.wav --wavelet cmor1.5-1.0 --seconds 5 --fft

Set up the .venv (as seen under "Quick start") OR install the dependencies under requirements.txt with pip on your own version of python. You'll then need to open a terminal run: cd "path/to/Wavelets" (replace with the actual path, then run the above command.

You can adjust the following:

• samples/Katydid.wav <=> Choose between samples/Katydid.wav, samples/RedeyedVireo.wav, samples/Gibbon.wav OR add more sample files to folder 'samples'

• for --wavelet "name of wavelet" you can choose between: ['cgau1', 'cgau2', 'cgau3', 'cgau4', 'cgau5', 'cgau6', 'cgau7', 'cgau8', 'cmor', 'fbsp', 'gaus1', 'gaus2', 'gaus3', 'gaus4', 'gaus5', 'gaus6', 'gaus7', 'gaus8', 'mexh', 'morl', 'shan']

  Complex Gaussian: cgau1 … cgau8 ~ Analytic derivatives of a Gaussian.

  Complex Morlet: cmor, cmorB-C ~ Plain cmor uses the library’s default bandwidth/centre; you can dial your own, e.g. cmor1.5-1.0.

  Frequency B-spline: fbsp, fbspM-FB-FC ~ Default fbsp; or specify order M, bandwidth FB, centre FC: fbsp2-1-1.5.

  Gaussian (real): gaus1 … gaus8 ~ nth derivative of a Gaussian, real-valued.

  Mexican hat: mexh ~ Also called the Ricker wavelet.

  Morlet (real): morl ~ Classic real Morlet.

  Shannon: shan, shanB-C ~ Default shan; or give bandwidth B and centre C, e.g. shan1-1.5.

• for --seconds "number of seconds" ~ Should never exceed the sound files duration and doesn't have to equal it either. Ideally short durations for less computation (e.g. 5 seconds)

• finally, for --fft ~ this is an optional boolean flag, meaning if you want an fft on top of the scalogram you can add this part to the end of the command, else, exclude it, like such: python cwt_scalogram_audiofile.py samples/Katydid.wav --wavelet cmor1.5-1.0 --seconds 5