TMA;DL - Too Much Audio, Didn't Listen

tmadl is a meetings & discussions recordings processor.

It's a personal project trying to make it easier to process meetings and discussion recordings with IA from your terminal/CLI.

The gist of it

The workflow that enable TMA;DL is essentially:

1. Record a meeting or discussion — no matter if it's through the native GMeet feature, a local ffmpeg command (soon to be wrapped in tmadl record), a recording made on your phone, etc.
2. Upload the resulting audio file to S3 through tmadl upload
3. An AWS StepFunction (deployed through AWS SAM as provided in the tmadl-transcription-audio-to-clean-txt directory) automatically kicks in, transcribe the audio into text through AWS Transcribe and a couple lambdas
4. Summarize/interrogate your transcrips using tmadl ask that queries AWS Bedrock (Anthropic Claude), providing meaningful context and the transcript in order to summarize it the way you want.

The CLI

$ ./target/debug/tmadl --help

TMA;DL (Too Much Audio, Didn't Listen) — A terminal/CLI meetings & discussions recordings processor using AI

Usage: tmadl [COMMAND]

Commands:
  record  Record a new meeting/conversation # Through ffmpeg. To be implemented
  upload  Upload a local recording to S3
  list    List all recordings (and their transcription statuses)
  ask     Interrogate a transcription
  help    Print this message or the help of the given subcommand(s)

Options:
  -h, --help     Print help
  -V, --version  Print version

The StepFunction

Internals

• The awesome clap Rust library is used to handle the CLI processing (commands/subcommands and their arguments/parameters)
• AWS S3 to store the meeting recordings (audio files)
• AWS Transcribe to do the audio transcription to text
• AWS Bedrock to do the transcriptions summarization on-demand

TODO later add:

• Sam for reaction on S3
• Ratatui or other for the dynamic termui?

https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_InvokeModel.html

• Uses the InvokeModel API from the Bedrock Runtime - not the Converse API or something else
• Claude message API format https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages.html
• High number of tokens + timeout to support long meetings

magicalrecord: aliased to ffmpeg -f pulse -i virtual_output_for_recording_purpose.monitor -f pulse -i alsa_input.pci-0000_07_00.6.HiFi__Mic1__source -filter_complex "[0:a]aresample=sample_rate=44100,volume=1[a0];[1:a]aresample=sample_rate=44100,volume=3[a1];[a0][a1]amix=inputs=2"  -ac 2

pactl load-module module-null-sink sink_name=virtual_output_for_recording_purpose
pactl load-module module-loopback source=virtual_output_for_recording_purpose.monitor

• id+date must be unique
• ID can be duplicate
• Files will be name as date_id.whateverextension