Generative Sound with Python Code and GPT | by Pablo (Apes Ascendance) | Jan, 2024

I want to create sounds using a WaveNet Neural Network created with a Custom GPT using WaveNet white paper as context (Knowledge) for GPT-4.

I want to make a Lion Roar (I have a Lions .wav 5-second dataset with 258 samples) so I’m using a WaveNet architecture created using a custom GPT for creating WaveNet architectures for generating the sound.

The GPT

Instructions: Create a full working Python Wavenet Code architecture that has:
Create a Neural Network with WaveNet architecture using TensorFlow and Keras
1) load_and_preprocess_audio method padding for 5 seconds at 22050 sample rate
2) Create a Generator
3) Create a Discriminator
4) Training for 10 epochs for Generator
5) saving the model
6) showing loss and training data

By custom Knowledge the Wavenet Paper

The sound I’m generating core points

✅I have a small 258 lion roaring 5- seconds each sample rate 22050

✅I have 32 GB of RAM, RTX 3070 Nvidia GPU, AMD Ryzen 7 CPU

First GPT Prompt (Optional)

Prompt: Give me a good Wavenet Architecture for a small dataset of 258 5-second samples of sounds, you are an expert Google DeepMind neural network architecture expert, give me the generator and discriminator with an optimized WaveNet architecture for a small dataset

Second GPT Prompt (Optional)

Give me a FULL working Python Code that includes: a) A generator optimized for WaveNet small dataset b) a discriminator c) pre_process_load 258 audio samples 5 seconds at 22050 sample rate d) train for 10 epochs and save the model, show loss and training info

My GPU wasn’t being detected and TensorFlow was running on the CPU on my Windows (I learned that this is the Preset) so…