Automatically transcribe your Evernote voice notes

I am a big believer in taking notes and working on my personal infrastructure. I find joy in polishing the “Artur OS” and removing little pockets of friction in my setup. Have a look at my automation philosophy:

Why do you have so many bots?

Recently, we adopted a dog and I have quite a bit of time spent walking. What if I could use it for some deep thinking?

So I wrote a bot.

How does it work?

A: I record an audio note on my phone:

B: I run my magical code

C: A transcription shows underneath

D: Profit!

There are other ways to solve this problem. Particularly, Otter.AI is a great service to transcribe your notes. However, it requires extra manual steps to open the app, export recordings, etc.

If you don’t have an elaborate setup behind your Evernote account, I recommend you check out Otter.

What do you need:

1. You need to set up the PHP (Yes) SDK for Evernote
2. You need an API key to Google Cloud Speech API. You don’t need a client library! Just follow these steps:
   1. Enable the API for your project here
   2. Create a new “API Interface Key” here

This code assumes you already search for a note and pass it in. I tag a note with `Tools`. A cron job periodically checks the label and does certain magical things on the notes tagged with it.

Code

It finds the audio recording in your Evernote note and transcribes it:

	<?php
	// https://piszek.com/2020/05/27/evernote-transcriber/
	function transcribe_audio_file_in_a_note( $evernoteClient, $note ) {
	// This will find the place where file is embedded, so we can display the transcription underneath.
	if( preg_match( '#<en-media hash="([a-z0-9]+)"[^>]+>#is', $note->content, $res ) ) {
	$id = hex2bin($res[1] );
	$resources = array_filter( $note->resources, function( $resource ) use ( $id ) {
	return $resource->data->bodyHash === $id;
	} );
	if( $resources ) {
	$resource = array_shift( $resources );
	}
	}
	if ( isset( $resource->mime ) && $resource->mime === 'audio/x-m4a' ) {
	// Only audio files
	if ( isset( $resource->attributes->applicationData->keysOnly['transcribed'] ) ) {
	$this->log( LOG_INFO, 'This resource is already transcribed.' );
	return;
	}
	// Set for the future
	$evernoteClient->client->getNoteStore()->setResourceApplicationDataEntry( $resource->guid, 'transcribed', 'true' );
	// this is your Google Speech API token
	$token = "tsrtrastr8astars8tras8t";
	$in = tempnam(sys_get_temp_dir(), 'evernote_transcript') . '.mp4';
	$out = tempnam(sys_get_temp_dir(), 'evernote_transcript') . '.wav';
	$data = $evernoteClient->client->getNoteStore()->getResourceData( $resource->guid );
	file_put_contents( $in, $data );
	// Because Google Speech API is crap and cannot deal with other formats, we have to recode it.
	system( "ffmpeg -i $in $out" );
	$data = file_get_contents( $out );
	$payload = array(
	"audio" => array( "content" => base64_encode( $data ) ),
	"config" => array(
	"languageCode" => "en-US",
	"alternativeLanguageCodes" => [ "pl-PL" ], // I only use English or Polish. Your mileage may vary.
	"encoding" => "LINEAR16",
	"sampleRateHertz" => 44100,
	"maxAlternatives" => 1,
	"enableAutomaticPunctuation" => true
	)
	);
	$payload = json_encode( $payload );
	$context = stream_context_create( array(
	'http' => array(
	'ignore_errors' => true,
	'header' => "Content-Type: application/json\r\n",
	'method' => 'POST',
	'content' => $payload
	)
	) );
	// Wondering about the v1p1beta1 here? You have to use this version to have alternativeLanguageCodes. This of course is not in documentation.
	$result = file_get_contents( "https://speech.googleapis.com/v1p1beta1/speech:recognize?fields=results&key=$token&quot;, false, $context );
	$result = json_decode( $result, true );
	if ( ! isset ( $result['results'][0]['alternatives'][0]['transcript'] ) ) {
	$this->log( LOG_WARNING, 'Empty transcript. ' . print_r( $result, true ) );
	return;
	}
	$text = $result['results'][0]['alternatives'][0]['transcript'];
	$this->log( LOG_INFO, 'Transcript OK: ' . $text );
	$new_body = str_replace( $res[0], $res[0] . "<div style='font-style: italic'>$text</div>",$note->content );
	$note->content = $new_body;
	$evernoteClient->client->getNoteStore()->updateNote( $note );
	}
	return $note->content;
	}

view raw transcribe_audio_file_in_an_evernote_note.php hosted with ❤ by GitHub

More of my adventures in automation:

I covered my automation philosophy in https://t.co/ieo9sGmMLz

The main goal is to reduce cognitive load and let me do more stuff with less time.
This is a thread 🧵with ongoing improvements to my personal infrastructure 👇.

— Artur Piszek (@artpi) January 10, 2020