This project lets you transcribe your audio input to Google Docs using Google's Cloud Speech API.

The Cloud Speech API is not free, so you will need to provide your own credentials to use this project.

As with any node project, npm install to install dependencies.

You will also need to install sox and have it available on your $PATH. On MacOS you can install it with Homebrew as brew install sox.

Create a Google Cloud project, enable the Speech-To-Text API for that project, create a service account, and aquire private key for that service account. You can do that all in one step by clicking the "Set up a project" button behind this link. Then copy the JSON with the private key for the service account to a file named speech-service-account-key.json in this directory.

Enable the Google Docs API for your project. You can do this from the APIs & Services page on the Cloud console. Make sure you're in the right project at the top, then click "Enable APIs and Services", search for the Google Docs API, and click Enable.

Set up the app to allow users to grant it permission to edit documents on their behalf. From the APIs & Services page, click "OAuth Consent Screen" in the left menu, click "External", and press "Create". Provide a name and email to be displayed to users and a developer contact, then click Save and Continue. On the Scopes page, select the ".../auth/documents" scope ("View and manage your Google Docs documents"). Ignore the warning about verification for now. Click Save and Continue until you get to the Summary page. Don't worry about registration for now, just click Back to Dashboard.

Finally, acquire the OAuth secret for your application. From the APIs & Services, select "Credentials", click Create Credentials at the top, select "OAuth Client ID", set the type to "Desktop app", give it a name, and click "create". It will tell you the ID and secret, but it's easier to click through and the click the download icon to the right of the newly created credential. Copy that JSON to a file named gdocs-client-oauth-secret.json in this directory.

node run.js

This will ask for the ID of a Google Doc to transcribe to. This is the 1fOSJp9hZLoR2BhiE0da87bLtoBBgcvGdi063IU1wqW8 part of doc's URL, as in https://docs.google.com/document/d/1fOSJp9hZLoR2BhiE0da87bLtoBBgcvGdi063IU1wqW8/edit. Make sure it's a document you have permission to edit.

If this is the first time running it, you'll then be asked to authorize an account. (The resulting credential is stored in GENERATED_TOKEN.json, so you should only need to do this once.) Go the URL it provides and grant permissions. You'll need to click through the big "This app isn't verified" warning (under "advanced"), unless you've verified your application. Ultimately you should get a string you will need to copy back to the terminal.

That's it! It will then run forever, transcribing your microphone to the doc. Kill it with control-C when you're done.

Keep in mind that this costs money. Specifically, since this is configured to use the fancy "video" model (which seemed to work best in my experience), it will cost $0.009 every 15 seconds, or about $2.16 an hour.

This records audio in, not audio out. So it won't directly help with, e.g., transcribing a stream you're viewing on your computer. You'll need to use a tool like JACK or BlackHole (MacOS only) to route your audio out to your audio in.