Python is my Love Language

A Valentine’s day gift for Laura

5 min readFeb 12, 2025

Hey,

This year for Valentine’s day instead of getting you some flowers that will die a few days later (I may have gotten you some at the last minute, but that is a problem for future Kyle) I have decided to express my love in the most natural way for me.

In a short-form technical document that merges Python coding and 3D printing, my other two loves.

Now, if you are reading this I hope you were able to access it via the NFC tag embedded in the Lovogram 3D print that I have hopefully given to you by now. Fingers crossed I followed through. As for this article, I wanted to chronicle my creation to show you it is indeed a labour of love. So without further ado, and in all likeliness you will skip over the most of it, here we go.

Step 1: Saying I Love You

To kick this off I had to get me and kids to record a sweet “I Love You” message and have it available online. You may listed to it here:

The next step was to take this recording and use python code to convert the .mp3 file to a spectrogram. A spectrogram is a 2 dimensional representation of a Fourier transform over time. You love Fourier transforms, I know, but settle down please. #LoveOnTheSpectrum

Part 2: Sound to Picture

What this means is that each column of data is a moment in time, and each row is a frequency. The darker the region (or higher the print) the more the frequency is binned into this section. You do that for each moment in time and get a 2d heatmap. Here is the actual spectrogram from our love message:

Woosh! That is super cool! Look at that! Love in visual form! Of course I know you want to see the code also!

import librosa
import librosa.display
import matplotlib.pyplot as plt
import numpy as np
import sys
import pandas as pd

def mp3_to_spectrogram(mp3_file, output_image='spectrogram.png', output_csv='spectrogram.csv'):
    # Load the audio file
    y, sr = librosa.load(mp3_file, sr=None)
    
    # Compute the spectrogram
    S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=60, fmax=8000)
    S_db = librosa.power_to_db(S, ref=np.max)
    
    # Normalize amplitude values to range [0, 1]
    S_db_norm = 5*(S_db - S_db.min()) / (S_db.max() - S_db.min())
    
    # Resample time dimension to 60 steps
    # S_db_resampled = librosa.util.fix_length(S_db_norm, size=60, axis=1)

    # # Get time and frequency bins
    # times = np.linspace(0, len(y) / sr, num=60)  # 60 evenly spaced time steps
    # freqs = librosa.mel_frequencies(n_mels=60, fmax=8000)
    
    # Get time and frequency bins
    times = librosa.times_like(S_db, sr=sr)
    freqs = librosa.mel_frequencies(n_mels=60, fmax=8000)
    
    # Save spectrogram data as CSV with integer time and frequency values
    data = []
    for i, t in enumerate(times):
        for j, f in enumerate(freqs):
            # Round time and frequency to the nearest integer
            t_int = int(round(t))  # Round time to nearest integer
            f_int = int(round(f))  # Round frequency to nearest integer
            data.append([i, j, S_db_norm[j, i]])
    
    df = pd.DataFrame(data, columns=['Time', 'Frequency', 'Amplitude'])
    df.to_csv(output_csv, index=False)
    print(f"Spectrogram data saved as {output_csv}")
    
    # Plot the spectrogram
    plt.figure(figsize=(10, 4))
    librosa.display.specshow(S_db_norm, sr=sr, x_axis='time', y_axis='mel')
    plt.colorbar(format='%+2.0f')
    plt.title('We Love You Mommy')
    plt.xlabel('Time')
    plt.ylabel('Frequency')
    plt.tight_layout()
    
    # Save the spectrogram as an image
    plt.savefig(output_image)
    plt.close()
    print(f"Spectrogram saved as {output_image}")

# Example call
mp3_to_spectrogram('C:\\Bitbucket\\Lovogram\\weloveumommy.mp3')

But Kyle! Where do you find the time!

With love Laura. I find the time with love.

Part 3: Picture to 3D Model

A picture is just data. In this case it has an x and y location of the pixels and the colour intensity is the z (height). All I did in the above was make a csv that could be read in. In Fusion 360 (the modelling software) you can use a “Add-In” and write python code. Here it is:

"""This file acts as the main module for this script."""

import adsk.core
import adsk.fusion
import csv
import os
import traceback

# Initialize the global variables for the Application and UserInterface objects.
app = adsk.core.Application.get()
ui  = app.userInterface
design = app.activeProduct
rootComp = design.rootComponent
sketches = rootComp.sketches
design.designType = adsk.fusion.DesignTypes.DirectDesignType

def run(_context: str):
    """This function is called by Fusion when the script is run."""

    try:
        # Your code goes here.
        xyPlane = rootComp.xYConstructionPlane
       
        extrudes = rootComp.features.extrudeFeatures

        # Open and read the CSV file
        with open('C:\\Bitbucket\\Lovogram\\spectrogram.csv', mode='r', newline='') as file:
            csv_reader = csv.reader(file)
            data = [row for row in csv_reader]  # Convert CSV to a list of rows


        #10-25
        #25-40
        for i,row in enumerate(data):
            if int(row[0])<100:
                continue
            if int(row[0])>150:
                return

            sketch = sketches.add(xyPlane)
            sketch.name = "TheGrid"  # Set the name of the sketch
            lines = sketch.sketchCurves.sketchLines
            recLines = lines.addTwoPointRectangle(adsk.core.Point3D.create(int(row[0]), int(row[1]), 0), adsk.core.Point3D.create(int(row[0])+1, int(row[1])+1, 0))
            # Now get the laest profile

            profile = sketch.profiles[len(sketch.profiles)-1]
            # Extrude the profile
            extrusion_height = float(row[2])+1

            extrude = extrudes.addSimple(profile, adsk.core.ValueInput.createByString(str(abs(extrusion_height))), adsk.fusion.FeatureOperations.NewBodyFeatureOperation)
            # extrude.bodies[0].name = 'John'
            # Get the whole set of bodies

            sketch_name = "TheGrid"  # Change to your sketch name

            for sketch in rootComp.sketches:
                if sketch.name == sketch_name:
                    sketch.deleteMe()
                    # ui.messageBox(f"{sketch_name} has been deleted.", "Success")
                    break


    except:  #pylint:disable=bare-except
        # Write the error message to the TEXT COMMANDS window.
        app.log(f'Failed:\n{traceback.format_exc()}')