Transcription and Summarization Pipeline

This project provides a pipeline to transcribe audio files (.m4a format) using AssemblyAI, summarize the transcriptions, and extract action points with OpenAI's GPT-4 model. It includes support for asynchronous transcription, intermediary storage, and reusable consolidated transcripts for consistent results.

Features

• Converts .m4a files to .wav for transcription.
• Asynchronous transcription with intermediary file storage.
• Summarizes transcriptions and extracts action points with GPT-4.
• Stores consolidated transcriptions to avoid redundant work.
• Date-time prefixed output files for organized storage.

Setup

1. Clone the Repository

git clone git@github.com:doingandlearning/voicenotes-to-actions.git
cd voicenotes-to-actions

2. Install Dependencies

pip install openai assemblyai pydub python-dotenv

Additionally, ensure ffmpeg is installed, as pydub requires it to handle .m4a files:

macOS

brew install ffmpeg

Ubuntu/Debian Linux

sudo apt update
sudo apt install ffmpeg

Windows

Download the ffmpeg binary from FFmpeg's official website and add it to your system PATH.

3. Environment Variables

Create a .env file in the root directory of the project. Add the following environment variables:

OPENAI_API_KEY=your_openai_api_key
ASSEMBLYAI_API_KEY=your_assemblyai_api_key

# Directory paths (default values shown)
INCOMING_BUCKET=./incoming_audio
INTERMEDIATE_TRANSCRIPTS=./intermediate_transcripts
OUTPUT_FOLDER=./output

• Replace your_openai_api_key and your_assemblyai_api_key with your actual API keys.

4. Directory Structure

The following directories are required for the pipeline to work as expected:

• incoming_audio/: Place .m4a audio files here for processing.
• intermediate_transcripts/: Stores individual transcription files for each audio file.
• output/: Stores the final transcripts.md and summary_action.md files, with a date-time prefix for unique filenames.

You can customize the directory paths in the .env file.

Usage

To run the pipeline, execute:

python main.py

Workflow

1. Transcription: The script converts .m4a files to .wav and transcribes them using AssemblyAI. Transcriptions are saved in intermediate_transcripts/.
2. Consolidation: Transcriptions are combined and saved as total_transcript.md in intermediate_transcript/.
3. Summarization: If total_transcript.md exists, it is used directly for summarization with OpenAI GPT-4 to create summary_action.md.
4. Output: The consolidated transcript and summary files are moved to output/, with the current date-time prefixed to each filename.

Example File Structure

After running the script, before clean up, you might have a structure like this:

project_root/
├── incoming_audio/
│   ├── audio1.m4a
│   └── audio2.m4a
├── intermediate_transcripts/
│   ├── audio1.wav.md
│   ├── audio2.wav.md
│   └── total_transcript.md
├── output/
│   ├── 2023-10-10-14-30-transcripts.md
│   └── 2023-10-10-14-30-summary_action.md
└── main.py

Notes

• Ensure ffmpeg is correctly installed and accessible in your system PATH for pydub to process .m4a files.
• Run the script in an environment with internet access, as it relies on external APIs (AssemblyAI and OpenAI).

License

This project is licensed under the MIT License.