Automating Voice Memos in Notion

•

At this point everyone has their voice to notion system. So, here is mine. Why you should read this thread? (1) On device whisper (2) Uploads the recording to notion too (3) Does some cool processing to add to database Which gives me the Transcribed Dump Database.

How do I do it? Well, a combination of stuff. 1) Shortcuts from Apple 2) Shortery for recurring shortcut runs on MacOS 3) Aiko for on device transcription (it is slow at the moment) 4) Terminal 5) Python packages: notional, openai [gpt3.5-turbo API], notion

I record stuff using "Say something". That shortcut works everywhere (watch, iphone, macos). The process shortcut only works on macos. It runs every 20 minutes using Shortery because I never need immediate processing.

Why not process on both iphone and macos? Because while I can run any python program on iphone using a-shell mini, I am using a unofficial notion api v3 library that is git dependent. Why? Because I need it to upload recording to notion.

Why are you running a python script instead of calling everything through shortcuts? Because shortcuts is buggy af and I want to move back to Android soon. Why not use pipedrive, zapier etc? Because I am broke, and cannot afford that or consider them worth the cost.

Alright, what does the Say Something shortcut do? Oh, it just 1. Records stuff 2. Save the recording and 3. Adds the path to a text file. It overwrites the audio file if all paths are processed, otherwise just adds a new one.

And what does the Process shortcut do? 1. Gets all paths to process from the text file 2. Transcribe it (Hides apps that show up because of transcription process) 3. Runs the python script 4. Update the text file to remove the path that was processed

Why not delete the file here? Because I can do it through the python script (but cannot later port it to iphone/android). Or I can use the shortcuts delete file and every single time it runs, it asks me permission. FTW. So, I just overwrite them.

Alright, what does that python script do? 1. It queries the Transcript Dump Settings Database for all settings. The settings database has the prompt, database id, token v2, API key for OpenAI, list of rich text properties, and list of multi-select properties.

Why? Because I want it to be both: (a) Portable across devices later, if I do want to process in mobile as a background process, so do not want to store this information in shortcuts (b) Easily shareable, so do not want to store that information in python script.

2. It then calls OpenAI API with the given prompt. it returns a particularly formatted json. I get a "title", "summary", "emotions", "health", "habit", "medicine", "entities", "people mentions", "date or time mentions", "topics", "food", "categories", "tags"

Based on the prompt, tags for example are free ended, while categories are selected from a pre-existing pool. 3. Then, I just parse the json, process it by keys and check whether it is a rich text or a multi select property, and create a new page with relevant properties.

Andddd...we are done. Use it from iphone lock screen, or menu bar or watch screen, processes when I am on mac and online, and... It creates this useful information database with transcript and audio in page. This is my voice notes setup.