Hello,
I was interested in hearing the audio files from the dementia bank. I noticed that some of them have no sound present for example file name abgk.mp3. Was the features calculated off that particular audio file for that participant? There are other files that are silient.
I also found some files where the assistant is talking and giving instructions to the participant, not the participant for example acqd.mp3.
I know the use of the audio files are optional, but it would be interesting to know if these were actually the files used.
Cecil
@Kingseso Hi Cecil! Yes, the acoustic features were generated from the exact same audio files shared via DementiaBank.
Silent audios — I double checked file abgk.mp3 , and there should be sound present in that audio. Is it possible that it did not download correctly from the TalkBank website? The metadata includes hashes for each individual audio file, which you can use to confirm the integrity of your downloaded file. I’d recommend using this to investigate whether there was an issue with the download. If not, is it possible that sound exists but is not being detected with the method you are using to listen to files?
Instructions in audios — You are correct, there are some audio files where the interviewer can be heard giving instructions to the participant. This is a result of the fact that diarization is not perfect, so there is sometimes another person besides the participant in the recordings. From our calculations, this issue is not widespread, and the participant is still speaking for the vast majority of the audios provided.
I hope this information is helpful, good luck!