Skip to content

Latest commit

 

History

History
26 lines (20 loc) · 2.19 KB

readme.md

File metadata and controls

26 lines (20 loc) · 2.19 KB
Raw

Article to Audio Google Cloud Function

This is a Google Cloud Function I hacked together that takes a url to an article on the web, and generates an audio file of it using Google's new Cloud Text-To-Speech API which has been updated with access to DeepMind's WaveNet voices.

I created it as part of a project to generate a personal podcast of articles I want to consume. To get the full thing working see my other repository with the Cloud Function that generates the Podcast RSS.

Sketch of how it works

  • The function accepts a POST request with json in the body.
    • E.g. {"url": "http://example.com/somearticle"}
  • It then uses the free Mercury Web Parser API to get the body of the article and some metadata.
  • Since the body is returned as HTML it then converts it to plain text. I also add some of the metadata at the top of the article, since I wanted this in the audio.
  • Then it slits up the body into chunks of no larger then 5,000 characters, since that's the limit on what the TTS API can handle per request.
  • From there is then sends each chunk of text to Google's TTS API which returns the audio encoded as MP3, and writes them to a temporary location.
  • Since having multiple files for parts of the article is annoying, it then uses FFMPEG to concatenate the audio chunks into one file.
  • Finally, it stores the audio file as and object in a Google Cloud Storage bucket, along with some of the metadata.

Configuration details

To get this working you need a Google Cloud Project with a Cloud Storage bucket setup, and the Cloud Text-To-Speech API enabled.

You'll then need to create a new Cloud Function (see configuration details below), and replace the undefined global constants in the code, gcpProjectID, gcpBucketName, and mercuryApiKey, with the appropriate values.

Cloud Function configuration

  • Trigger type: HTTP trigger
  • Memory allocated: 256 MB
  • Timeout: 240s
    • I had to extend this from default of 60s.