Wrong folder structure, fix it later

FLAC Audio Chunking & Whisper Transcription

This project chunks FLAC audio files into 20-second segments and transcribes them using OpenAI's Whisper LLM via HuggingFace Inference API.

Features

• Audio Chunking: Splits FLAC files into configurable 20-second segments using PyDub
• Whisper Integration: Sends each chunk to Whisper LLM for transcription
• Results Management: Writes all transcriptions to a formatted text file
• File Protection: Results file is set to read-only after creation

Prerequisites

1. Python 3.8+

2. FFmpeg: Required by PyDub for audio processing

   • macOS: brew install ffmpeg
   • Linux: apt-get install ffmpeg
   • Windows: Download from https://ffmpeg.org/download.html

3. HuggingFace Credentials:

   • Create account at https://huggingface.co
   • Get HF Token from https://huggingface.co/settings/tokens
   • Set up HF Inference endpoint for Whisper

Setup

1. Install Dependencies:

   pip install -r requirements.txt

2. Configure Environment Variables in .env file:

   HF_TOKEN=your_huggingface_token_here
   HF_INFERENCE_ENDPOINT=your_inference_endpoint_url_here
   

3. Prepare Audio Files:

   • Place FLAC files in the audio/ folder
   • Ensure files are in FLAC format (.flac extension)

Usage

Run the main script:

python chunk_and_transcribe.py

What the Script Does

1. Scans the audio/ folder for FLAC files
2. For each FLAC file:
   • Creates chunks of 20 seconds
   • Saves chunks to chunks/{filename_without_extension}/ folder
3. Processes all chunks through Whisper LLM API
4. Writes results to transcription_results.txt (read-only)

Output

• Chunks: Stored in chunks/{original_filename}/ directory
• Transcriptions: Saved to transcription_results.txt
  • Formatted with chunk names and transcription results
  • Timestamp of generation included
  • File is automatically set to read-only

File Structure

chunk-script/
├── audio/                          # Place FLAC files here
├── chunks/                         # Generated chunk files
├── sciprt2.py                      # Original script (reference)
├── chunk_and_transcribe.py         # Main processing script
├── transcription_results.txt       # Output file (read-only)
├── requirements.txt                # Python dependencies
├── .env                            # Environment variables (not in repo)
└── README.md                       # This file

Configuration

Edit chunk_and_transcribe.py to customize:

• Chunk Length: Change chunk_length = 20 (in seconds)
• Audio Folder: Change audio_folder = "audio"
• Chunks Folder: Change chunks_folder = "chunks"
• Output File: Change output_text_file = "transcription_results.txt"

Troubleshooting

"FFmpeg not found"

• Install FFmpeg (see Prerequisites section)

"No FLAC files found"

• Verify files are in audio/ folder
• Check file extension is .flac

API Errors

• Verify HF_TOKEN is correct in .env
• Check HF_INFERENCE_ENDPOINT URL is valid
• Ensure you have API credits/access

Permission Denied on Output File

• The file is intentionally read-only. Change permissions if needed:

  chmod u+w transcription_results.txt

Original References

• Original script: sciprt2.py
• Reference chunking code provided
• Built with PyDub and Whisper LLM integration

Notes

• Each chunk is processed sequentially for reliable API handling
• Results include both successful transcriptions and any errors
• File timestamps help track when transcriptions were created
• Read-only status prevents accidental modification of results