I am currently a Technical Project Officer (Media) at the IT Services of the University of Oxford.
I worked for the JISC-funded SPINDLE project where we explored the use of automatic speech recognition to generate automatically better cataloguing of media content (audio and video podcasts). We reported our findings using the Open Spires blog.
The main goal was to generate automatically a set of keywords using natural language processing techniques from the automatic speech-to-text transcriptions of the university podcasts to increase the open educational audio and video resources discoverability. These automatic transcriptions are generated using state-of-the-art Large Vocabulary Continuous Speech Recognition Software.
For example, for one of our most succesful podcasts, The nature of human beings and the question of their ultimate origin, we transcribe it automatically obtaining a transcription that is 56.29% accurate (out of 100 words around 56 are correct). Then, we obtain automatically the most relevant keywords using the Log-likelihood measure. A word cloud (using Wordle) of the resulting 100 most significant keywords can be found in the following figure.