I have made two different assignments. Of course you may do them both but one is enough.
Install Whisper on your own computer or, if that’s not possible, on a server at your own institute. Describe the installation procedure very carefully and do it in such a way that any somewhat experienced colleague can easily copy the procedure.
Once installed, run the test recordings. These can be found on AV-files (will be available before 29 jan)
Compare the recognition results with the Golden Transcripts provided and show what goes right and what goes wrong. Come up with suggestions on what can be improved and show in at least 3 cases that your suggestions actually work.
Finally, provide the transcription of each file in English, Dutch and (eventually) in your own language.
The paper should be written in such a way that it is reasonably easy for us to install it as you have done it. It means that we will try to copy what you have done.
Describe the development over the last 20 years of automatic speech recognition for your own language. Make clear how the different (no longer) existing "versions" worked, and describe the advantages and disadvantages of each version. Also state what is still going right and wrong today, and indicate whether that is something that is more-or-less permanent or likely to be fixed soon.
Run 5 AV files of ±5 minutes through the best version of the recogniser of your own language and show what can/should be improved. The transcripts should be delivered in your own language and in English.
The length of the paper (including images and other non-text items) is maximal ±16 pages. Moreover, the different transcripts need to be added separately as a zip-file.
For questions: mail me!!