Joe Frank Player

Three-Stage Pipeline

Each episode is created through an automated three-stage process:

1

Script Generation

Using OpenAI GPT-4o, the system generates:

Episode outline and narrative structure
5-6 story segments (800-1000 words each)
Unique voice assignments per segment
Text optimized for text-to-speech

Output: YAML file with structured episode data

2

Voice Narration

Using Bark TTS (Suno AI), the system:

Converts script text to natural speech
Applies different voice presets per segment
Adds optional telephone/broadcast effects
Normalizes audio to prevent clipping

Output: Individual WAV files (24kHz, 16-bit)

3

Music Generation

Using Facebook MusicGen, the system:

Generates atmospheric background loops
Uses curated prompts (jazz, ambient, trip-hop)
Evaluates quality with dynamic scoring (0-10)
Regenerates clips that score below 3.0

Output: Background music loops with spectral analysis

Key Technologies

GPT-4o Script generation & storytelling

Bark TTS Natural voice synthesis

MusicGen AI music composition

PyTorch Deep learning framework

Generation Time

A complete episode typically takes:

Script: 5-10 minutes
Voice: 10-30 minutes (GPU-accelerated)
Music: 30-120 minutes (8 quality-checked loops)

Total: ~1-2.5 hours per episode

Playback System

The player uses a sophisticated real-time mixing system built with Tone.js:

🎵 Intelligent Music Looping

Each music loop plays 5-10 times before transitioning
5-10 second pause between music changes
Smooth fade-in (6s) and fade-out (10s) transitions
Random loop selection to avoid repetition

🎙️ Story Narration

Random story selection without repeats
30-60 second pauses between story segments
75% chance of telephone effect per segment
NPR broadcast-style audio processing chain

🔊 Audio Processing

Music Ducking: Automatically reduces music to 50% volume during narration
Broadcast Chain: High-pass filter → Gate → Compressor → EQ → Room reverb
Telephone Effect: Bandpass filter (1700Hz) + distortion for vintage sound
Master Processing: Compression + limiting to prevent clipping

⏱️ Episode Management

30-minute episodes with auto-advance to next episode
Real-time progress tracking and visualization
FFT spectrum analyzer for visual feedback
Volume settings saved in browser cookies