After having done speech vs. song segmentation of the audio podcasts of Radio Lockdown (see here), I could start playing a bit with these intervals. Therefore I used the fantastic SpeechRecognition python package to perform speech to text of the detected speech intervals (after having created my own google cloud API account, necessary because I needed to be able to perform more queries than the ones offered without an account). Once I obtained the transcription of all the speech segments of a given podcast, I used the WordCloud package to generate a visual summary of the most frequent words as a proxy of the episode’s main topics. Here a couple of examples of some of the resulting wordclouds!