🎙️ PPTX Text and Audio Transcriber

Extract text and transcribe audio from PowerPoint presentations using OpenAI Whisper.

Features

• 📝 Text Extraction: Extracts all text content from PowerPoint slides
• 🎤 Audio Transcription: Uses OpenAI Whisper to transcribe embedded audio files (WAV, MP3, M4A)
• ⚡ GPU Acceleration: Automatic CUDA detection with CPU fallback
• 🎯 Multiple Models: Supports various Whisper model sizes (tiny, base, small, medium, large)
• 📊 Progress Tracking: Real-time progress bars during processing
• 🔧 Configurable: Easy-to-modify settings for performance and quality tuning

Requirements

• Python 3.8 or higher
• CUDA-compatible GPU (optional, but recommended for faster processing)

Installation

1. Clone or Download the Project

git clone https://github.com/sankeer28/pptx-text-audio-transcriber.git
cd pptx-text-audio-transcriber

2. Create Virtual Environment (Recommended)

python -m venv venv
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

3. Install Python Dependencies

pip install -r requirements.txt

4. CUDA Setup (Optional but Recommended)

For GPU acceleration, install CUDA and compatible PyTorch:

Windows:

1. Install CUDA Toolkit:

   • Download from NVIDIA CUDA Toolkit
   • Choose version 11.8 or 12.1 (recommended)
   • Follow the installer instructions

2. Install cuDNN:

   • Download from NVIDIA cuDNN
   • Extract and copy files to CUDA installation directory

3. Install CUDA-enabled PyTorch:

   pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118

macOS/Linux:

# For CUDA 11.8
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118

# For CUDA 12.1
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu121

5. Verify Installation

Test CUDA availability:

import torch
print(f"CUDA available: {torch.cuda.is_available()}")
if torch.cuda.is_available():
    print(f"GPU: {torch.cuda.get_device_name(0)}")

Usage

1. Prepare Your Files

• Create a presentations folder in the project directory
• Place your PowerPoint files (.pptx) in the presentations folder

2. Run the Extractor

python main.py

3. Check Results

• Extracted content will be saved in the output folder
• Each PowerPoint file generates a corresponding .txt file with:
  • All slide text content
  • Transcribed audio content
  • Processing metadata

Configuration

Edit the configuration settings at the top of main.py:

Folder Settings

PPTX_FOLDER = "presentations"   # Input folder
OUTPUT_FOLDER = "output"        # Output folder

Whisper Model Settings

WHISPER_MODEL = "base"        # Options: "tiny", "base", "small", "medium", "large"
FORCE_LANGUAGE = "en"         # Force language ("en", "es", "fr", etc.) or None for auto-detect

Performance Settings

FORCE_DEVICE = None           # Options: None (auto), "cpu", "cuda"
USE_HALF_PRECISION = False    # Enable fp16 for 30-50% speed boost (GPU only)
GPU_BEST_OF = 3              # Higher = more accurate, slower
GPU_BEAM_SIZE = 3            # Beam search size

Quality Settings

TEMPERATURE = 0.0             # 0.0 = deterministic, 0.1-1.0 = more creative
ENABLE_WORD_TIMESTAMPS = True # Get word-level timing data

Model Size Guide

Model	Size	Speed	Quality	VRAM Usage
tiny	39MB	Fastest	Good	~1GB
base	74MB	Fast	Better	~1GB
small	244MB	Medium	Good	~2GB
medium	769MB	Slow	Very Good	~5GB
large	1550MB	Slowest	Best	~10GB

Troubleshooting

Common Issues

1. CUDA Out of Memory

• Use a smaller Whisper model (tiny or base)
• Set USE_HALF_PRECISION = True
• Reduce GPU_BEST_OF and GPU_BEAM_SIZE

2. No Audio Files Found

• Ensure audio is embedded in PowerPoint (not linked)
• Supported formats: WAV, MP3, M4A

3. Installation Issues

• Ensure Python 3.8+ is installed
• Try installing dependencies one by one
• Use virtual environment to avoid conflicts

4. Poor Transcription Quality

• Use larger Whisper model (medium or large)
• Set correct language with FORCE_LANGUAGE
• Increase GPU_BEST_OF for better accuracy

Performance Tips

• GPU Users: Use base or small models for best speed/quality balance
• CPU Users: Stick with tiny or base models
• Large Files: Process in batches to avoid memory issues
• Quality Focus: Use medium or large models with higher beam size