Automatically transcribe Tuple call recordings using whisper-cpp and generate a Markdown summary with a timeline of events and dialogue.

Fires on the call-recording-complete event. Reads Opus audio files and events.jsonl from the call artifacts directory, transcribes each audio segment, and writes a Summary.md file.

Reach out to us at support@tuple.app if you want us to enable local call recording for you or your team.

Prerequisites

1. whisper-cpp — install via Homebrew:

   brew install whisper-cpp
   

2. ffmpeg — install via Homebrew:

   brew install ffmpeg
   

3. A GGML model file — download one to the default location:

   mkdir -p ~/.local/share/whisper-cpp/models
   curl -L -o ~/.local/share/whisper-cpp/models/ggml-large-v3.bin \
     https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3.bin
   

Configuration

The trigger works out of the box with no configuration. Two optional environment variables are available:

Variable	Description
WHISPER_MODEL	Path to a GGML model file (overrides the default ~/.local/share/whisper-cpp/models/ggml-large-v3.bin)
TRANSCRIPTION_OUTPUT_PATH	Directory to write Summary.md (defaults to the call artifacts directory)

Output

The generated Summary.md includes:

• Call date and time
• List of participants
• A timeline interleaving call events (joins, leaves) with transcribed speech, attributed to each speaker