garystafford

Introduction

AI-generated voices have become remarkably convincing. Services like ElevenLabs, Amazon Polly Neural, and dozens of other text-to-speech platforms can now clone voices with startling accuracy. While these technologies enable amazing applications—from audiobooks to accessibility tools—they also raise serious concerns about misinformation, fraud, and identity theft.

In this post, we'll build a production-ready deepfake audio detection system that can distinguish human speech from AI-generated audio in real-time. We'll cover the complete machine learning lifecycle: creating a high-quality dataset with sophisticated audio preprocessing, fine-tuning a Wav2Vec2 transformer model using transfer learning, and deploying it to Amazon SageMaker with streaming inference support.

By the end, you'll understand how to:

• Implement a two-pass audio splitting algorithm with Voice Activity Detection for optimal training data
• Leverage transfer learning to achieve 99%+ accuracy with minimal training time

garystafford

#!/usr/bin/env python3
"""
Strands Agent using Ollama and IBM Granite model for math calculations and LaTeX formatting.

This script sets up a Gradio interface to interact with a Strands Agent powered by
an Ollama-served IBM Granite model. The agent can perform mathematical calculations
and format responses using LaTeX.
"""

import logging

garystafford

# Author: Gary A. Stafford
# Purpose: Delete all SageMaker Models
# Date: 2024-05-08
# License: MIT License
# Based on https://gist.github.com/Nxtra/49cde5999de20023e2607433046ca6cf


import boto3

client = boto3.client("sagemaker")

garystafford

# Author: Gary A. Stafford
# Purpose: Delete all SageMaker Endpoint Configs
# Date: 2024-05-08
# License: MIT License
# Based on https://gist.github.com/Nxtra/49cde5999de20023e2607433046ca6cf

import boto3

client = boto3.client("sagemaker")

garystafford

# Install ComfyUI on Ubuntu
apt install python3-pip
sudo apt install python3-pip

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI/
pip install -r requirements.txt

pip install torchvision # error message in terminal

garystafford

# For use with AUTOMATIC1111 on Ubuntu

# current install 2024-06-10:
# version: v1.8.0
# python: 3.10.12
# torch: 2.3.0+cu121
# xformers: 0.0.26.post1
# gradio: 3.41.2
# checkpoint: e6bb9ea85b

garystafford

import torch

print(torch.__version__) # e.g., 2.0.0 (at the time of the post)
print(torch.cuda.get_device_name(0)) # e.g., NVIDIA A10G

garystafford

pipeline = DiffusionPipeline.from_pretrained(
    model_name_base,
    torch_dtype=torch.float16,
).to(device)

# new LoRA weights from fine-tuning process
pipeline.load_lora_weights(
    project_name, 
    weight_name="pytorch_lora_weights.safetensors"
)

garystafford

subject_prompt = """oue, photo of a oue electric scooter in a brightly colored 
neon-lite city at night, sleek design, smooth curves, colorful, nighttime, 
urban environment, futuristic cityscape"""

subject_negative_prompt = """person, people, human, rider, floating objects, daytime, 
sunlight, text, words, writing, letters, phrases, trademark, watermark, icon, logo, 
banner, signature, username, monochrome, cropped, cut-off, patterned background"""

refiner_prompt = """ultra-high-definition, photorealistic, 8k uhd, high-quality, 
ultra sharp detail"""

garystafford

subject_prompt = """oue, photo of a oue electric scooter, sleek, smooth curves, colorful, 
daytime, urban, futuristic cityscape"""

subject_negative_prompt = """person, people, human, rider, floating objects, text, 
words, writing, letters, phrases, trademark, watermark, icon, logo, banner, signature, 
username, monochrome, cropped, cut-off, patterned background"""

refiner_prompt = """ultra-high-definition, photorealistic, 8k uhd, high-quality, 
ultra sharp detail"""