Applied NLP and voice experiences
PDF-to-Audiobook
I designed an AI-assisted PDF-to-audiobook experience that treats narration as a context and characterization problem, not just text-to-speech.
- Role
- Designer and engineer
- Team
- Solo, personal project
- Status
- Personal project
- Stack
- Machine Learning, NLP, LLMs, Document processing
Outcome
The project demonstrates product-minded AI engineering: a system that combines NLP and voice experiences to make documents more accessible, personal, and engaging.
- NER + corefkeeps characters consistent across a book
- Dialoguespeaker attribution before voice rendering
- Resumablebatch pipeline for long-form books
What I owned
Built a workflow that turns source documents into narration-ready content with structure, speaker context, and listening flow in mind.
Applied ML, AI, and NLP techniques to personalize voice and tone rather than using one flat reading style for every passage.
Focused on an experience that can preserve meaning while making long-form content easier and more engaging to consume.
System at a glance
A document ingestion step extracts text and keeps useful structural context.
Key decisions
- Treated document structure and context as first-class inputs because raw text alone loses much of what makes narration understandable.
- Designed personalization around listening intent, tone, and character rather than a single generic voice.
- Kept the system focused on a clear human outcome: making dense written content more natural to listen to.
How I approached reliability
- Separated extraction, analysis, narration planning, and voice rendering so the system can be understood and improved one layer at a time.
- Preserved document context through the workflow to reduce abrupt tone shifts and disconnected narration.
- Used explicit intermediate stages so quality can be reviewed before content reaches the listener.
Technology focus
- Machine Learning
- NLP
- LLMs
- Document processing
- Text-to-speech
- Voice design
- Context-aware generation