Voice-Blog-Creator

Automated agent workflow to convert voice recordings into polished blog posts using Gemini AI.

Overview

This project provides a complete pipeline to:

1. Preprocess raw audio files for optimal speech-to-text performance
2. Transcribe audio using Gemini 2.5 Flash with light redaction (removes filler words, adds paragraphs)
3. Generate formatted blog posts optimized for web presentation

Workflow Steps

input/audio-file/{folder}/raw.mp3
    ↓ [Step 1: Preprocess]
output/{folder}/processed.mp3
    ↓ [Step 2: Transcribe]
output/{folder}/transcript.txt
    ↓ [Step 3: Generate Blog]
output/{folder}/blog_post.md

Step 1: Audio Preprocessing

• Converts stereo to mono
• Removes silence while keeping natural pauses
• Reduces background noise
• Normalizes audio levels
• Applies dynamic range compression
• Optimizes for STT (16kHz sample rate)

Step 2: Transcription (Gemini 2.5 Flash)

• Sends processed audio to Gemini API
• Light redaction:
  • Removes filler words (um, uh, like, you know, etc.)
  • Organizes into paragraphs based on topic changes
  • Adds proper spacing
  • Maintains original meaning and speaker's voice

Step 3: Blog Post Generation (Gemini 2.5 Flash)

• Converts transcript to formatted blog post
• Creates compelling title
• Adds introduction and conclusion
• Organizes with subheadings (## and ###)
• Optimizes for web readability
• Uses proper markdown formatting

Quick Start

1. Setup

./setup.sh

This will:

• Create virtual environment
• Install dependencies
• Create .env file template
• Set up directory structure

2. Configure API Key

Edit .env and add your Gemini API key:

GEMINI_API_KEY=your_api_key_here

3. Add Audio

Place your raw audio file:

input/audio-file/1/raw.mp3

4. Run Workflow

./run.sh 1

This will generate:

• output/1/processed.mp3 - Preprocessed audio
• output/1/transcript.txt - Lightly redacted transcript
• output/1/blog_post.md - Formatted blog post

Directory Structure

Voice-Blog-Creator/
├── input/                    # Input files
│   └── audio-file/
│       └── {folder}/
│           └── raw.mp3       # Your original audio
│
├── output/                   # Output files
│   └── {folder}/
│       ├── processed.mp3     # Preprocessed audio
│       ├── transcript.txt    # Lightly redacted transcript
│       └── blog_post.md      # Formatted blog post
│
├── scripts/                  # Processing scripts
│   ├── preprocess_audio.py
│   ├── gemini_transcribe.py
│   ├── gemini_blog_post.py
│   └── workflow.py           # Orchestration
│
├── setup.sh                  # Setup script
├── run.sh                    # Run workflow
├── requirements.txt          # Python dependencies
└── .env                      # API keys (not in git)

Usage

Complete Workflow

./run.sh 1                    # Process folder 1
./run.sh 1 --force            # Force regenerate all files
./run.sh 1 --verbose          # Show detailed processing
./run.sh 1 --steps 2 3        # Run only steps 2 and 3

Individual Steps (Advanced)

# Step 1: Preprocess audio
.venv/bin/python scripts/preprocess_audio.py \
  --input input/audio-file/1/raw.mp3 \
  --output output/1/processed.mp3

# Step 2: Transcribe
.venv/bin/python scripts/gemini_transcribe.py \
  --input output/1/processed.mp3 \
  --output output/1/transcript.txt

# Step 3: Generate blog
.venv/bin/python scripts/gemini_blog_post.py \
  --input output/1/transcript.txt \
  --output output/1/blog_post.md

Requirements

• Python 3.12+
• UV package manager
• Gemini API key
• ffmpeg (for audio processing)

Features

• Clear Separation: Input and output files are clearly separated
• Smart Caching: Skips steps if output already exists (use --force to override)
• Flexible Pipeline: Run individual steps or full workflow
• Verbose Logging: Debug with -v flag
• Quality Optimization: Each step optimized for its specific purpose

API Costs

• Gemini 2.5 Flash is used for both transcription and blog generation
• Estimated cost: ~$0.01-0.05 per hour of audio (varies by length and complexity)

Troubleshooting

Virtual environment not found

Run ./setup.sh to create the environment and install dependencies.

API key errors

Make sure .env contains your valid Gemini API key:

GEMINI_API_KEY=your_actual_api_key_here

Audio file not found

Ensure your audio file is placed at:

input/audio-file/{folder_number}/raw.mp3