Watson Speech to Text Integration

Available since version 3.3.0

This feature is AEM 6.2 ONLY!

Purpose

Use IBM’s Watson Speech to Text transcription service to extract transcriptions from Assets.

How to Use

The way this works is that when a video or audio file is run through the TranscriptionProcess workflow process, it first generates a FLAC file of just the audio, then passes that file to the Watson web service, getting a Job ID back. The workflow process then polls to see if the job is complete, finally saving the rest of the process into a rendition of the asset named transcription.txt.

To use this, first create a configuration for com.adobe.acs.commons.http.impl.HttpClientFactoryImpl (with some unique name) with factory.name property set to watson-speech-to-text, the credentials and hostname for Watson.

For example, on the hosted BlueMix platform, this might look like:

{
  "jcr:primaryType":"sling:OsgiConfig",
  "factory.name":"watson-speech-to-text",
  "hostname":"stream.watsonplatform.net",
  "username":"1aa30b41-a015-4a6c-a335-aad71b6ec8fd",
  "use.ssl":true,
  "port":443,
  "password":"XXXX"
}

Then, add the Transcription Process to a workflow (either the DAM Update Asset Workflow or a separate workflow).

Watson automatically chunks the audio based on pauses and attaches a timecode to each chunk. These time-codes are used in the transcription.txt rendition like this:

[0.66s]: hello hello hello
[3.27s]: hello hello hello hello

FFMPEG Configuration

In order for this process to work, FFmpeg must be installed and must be on your path. See the AEM documentation for general instructions for installing FFmpeg.

FFmpeg must also be capable of creating FLAC file. To confirm, this you can run ffmpeg -formats and confirm that flac is listed like this:

 DE flac            raw FLAC

You can also go to the flacmono Video Profile page in AEM (i.e. http://localhost:4502/etc/dam/video/flacmono.html) and upload a test file.

Please report any issues with the Feature or documentation on the ACS AEM Commons GitHub Issue tracker.
comments powered by Disqus