A Voci V-Blaze Natural Language Processor

• Python 2.7
• NLTK 2.0 (with data)

pip install blazor

NLTK data packages are required as well. For installation instructions see http://www.nltk.org/data.html.

The transcript output of the V-Blaze web service is parsed and processed as follows:

This is where the transcript generated from the Voci V-Blaze is loaded, tokenized into full sentences and cleaned from vocal tags (<s>, ++BRTH++, ++AH++, etc.). Call is divided into Utterances, each utterance contains its respective properties (id, confidence, start, end, channel, sentiment, gender, breath_count, silence_count, and sentences).

A Sentiment Analyzer is trained on movie reviews and then utilized to classify (using NaiveBayesClassifier) each utterance sentence to one of pos or neg classes. Positive and Negative fractions are then stored for later indexing.