As part the project outreach, MICO is exhibiting at IBC in Amsterdam, Europe’s largest conference for broadcasters. We demonstrate the system and illustrate its versatility though a range of showcase: MICO is used by Zooniverse to add machine support and analysis to their citizen-science platform; by Fraunhofer to power the forensic software AudioTrust+, by INsideOUT10 for their publication platforms HelixWare and Shoof!, and by Smart Video to turn marketing videos into online stores. Visitors are able to see the MICO system in action and try their hands at creating workflows and querying the derived data.
Automatic metadata extraction is a clear theme at this year’s edition of the conference and many stop to talk about MICO. As Patrick Aichhtroth, Fraunhofer IDMT, observed, the most looked-for extractors are speech-to-text, NER, and face detection & recognition. We receive particularly many questions about the extensibility of the speech-to-text extractor based on Kaldi. The happy answer is that it is open source, and can be traind for new languages. What is needed is a parallel corpus with spoken and written text, a good deal of processing time, and some understanding of how to work Kaldi’s parameters.
If you are interested to know more, stop by for a demo in the Fraunhofer booth, Room 8, number 80.