_Like my articles? Feel free to vote for me_ as ML Writer of the year here_._

It takes no political analysis expertise to tell that public opinion is much influenced by what and how the politicians speak in public. In this article we have used 30 videos from the two politicians, relatively uniformly spread from January 2020 until today (September 2020), to demonstrate how speech analytics can be used to extract valuable conclusions from such data.

Our goal is to illustrate WHAT and HOW the two candidates have been speaking about during the last year

Method

• Data: 35 videos from speeches from both candidates (around 2 speeches per month on average) have been used to extract our analytics.
• Text data were retrieved through Automatic Speech Recognition (ASR). This information will help us to visualize what the two candidates talked about. ASR, of course is not totally error-free, the ASR model used in this particular use-case have a word-error-rate around 25% on the particular data. However, in this article we have used simple aggregates to analyze our final texts (such as word clouds), and so the error of this final representation should be much lower.
• The raw audio data from the speeches, were analyzed in terms of the speakers emotions and behaviors using Behavioral Signals Oliver API. In particular, we focused on the speech emotional strength (arousal) and the speech emotional positivity (also known as valence). This second level of analysis will provide a metric on how the two candidates have spoken during that period.

#machine-learning #speech-recognition #text-mining #us-election #donald-trump #joe-biden