Scripts for downloading YouTube audio and transcribing to text with OpenAI Whisper.

• Python 3.10+
• FFmpeg installed on system
• CUDA GPU (optional, for faster processing)

pip install yt-dlp openai-whisper librosa scikit-learn

# 1. Download audio from YouTube
python download_audio.py

# 2. Transcribe with speaker diarization
python audio_to_text_with_speakers.py

# 3. Clean transcript with Gemini
python clean_transcript_with_gemini.py

# 4. Replace speaker names
python replace_speaker_names.py

Downloads audio from YouTube videos and converts to MP3 format. Handles Turkish character sanitization in filenames.

Basic audio transcription using Whisper. Generates full text and timestamped output files. Supports GPU acceleration via CUDA.

Outputs:

• {filename}_full.txt - Full transcription
• {filename}_timestamps.txt - Transcription with timestamps

Advanced transcription with speaker diarization. Uses MFCC and spectral features with K-Means clustering to identify different speakers.

Outputs:

• {filename}_full.txt - Full transcription
• {filename}_with_speakers.txt - Timestamped with speaker labels
• {filename}_speakers.txt - Grouped by speaker

Uses Gemini AI to clean and correct transcription errors, fix grammar, and improve readability.

Replaces generic speaker labels with actual names:

• [Speaker 1] → [Enes Fehmi Manan]
• [Speaker 2] → [Guest Name]

• All scripts optimized for Turkish language
• GPU acceleration automatic when CUDA available
• Speaker diarization works best with 2-3 speakers