🧠 Speech-to-Text Analysis & Semantic Chunking

This Jupyter Notebook explores advanced speech-to-text processing through two major tasks:

🎬 Part 1: Semantic Chunking of YouTube Audio

• Downloaded and extracted audio from a YouTube video using yt-dlp.
• Transcribed speech to text using OpenAI Whisper.
• Performed time-aligned transcription and speaker diarization using PyAnnote.
• Applied semantic chunking based on sentence structure, speaker turns, and conjunctions for meaningful text segmentation.

📖 Part 2: Exploratory Analysis of New Testament Audio

• Aligned audio and transcript at the word and phoneme level.
• Conducted detailed word-level speech analysis, including misalignment, pauses, and anomalies.
• Detected audio anomalies (like silence, distortion) using waveform and spectral analysis.
• Performed text bias and linguistic analysis to examine potential content skew.
• Analyzed audio quality, duration trends, and phoneme usage.

🔧 Tech Stack

• Python, Jupyter
• OpenAI Whisper, PyAnnote, SpaCy, NLTK, Librosa, YouTube-DLP
• Matplotlib, Seaborn for visualization

📌 Future Work

• Automate semantic segmentation for long-form educational videos.
• Extend phoneme-based alignment to multilingual corpora.