Transcribe Podcast Audio Testing

This is a repository demonstrating transcribing podcast audio with a few different services: IBM Watson Speech to Text, Google Cloud Speech, and CMU Sphinx.

Getting Started

First run fetch_podcast.sh to pull some content down for converting. The example from my blog post is used, though you can replace the URL if you like to try with other content.

Running CMU Sphinx

Ensure that sphinx is installed. On Ubuntu this is done with

sudo apt-get install pocketsphinx pocketsphinx-en-us

Then run transcribe_with_sphinx.sh. It will produce sphinx-transcription.log as output.

Running Watson

Copy user/password into auth.cfg in the [watson] section.

Install python prereqs

pip install watson-developer-cloud

Then run it with

watson-transcribe.py podcast.flac

Optionally specify a pretrained customized language id

watson-transcribe.py --customization a13780b0-52b7-1fe7-fbb7-77471c70c949 podcast.flac

Note

It will take 30 - 45 minutes to run, it processes slightly faster than realtime.

Running Google

Sign up for Google Cloud and create a project and a service key. Go through all the setup around authentication - https://googlecloudplatform.github.io/google-cloud-python/stable/google-cloud-auth.html

Create a key for the service, put that json in auth.json in this directory.