Chapter 16: Exercises

  1. (*) Dataset collection for audio fingerprinting:
    1. Create corpus for AFP
      • At least 20 music files in mp3, with file name "songtitle_singerName.mp3". (You can use ffmpeg to convert music files into mp3 format.)
      • At least 10 noisy recordings of the music files via your smartphone, with file name "songTitle_singerName_noisy.wav".
        • You should start the recording after playing the music, and end the recording before the music stops. That is, you should try your best not to record any silence or non-music parts during your recording.
        • Try to make the recording noisy (by talking, yelling, clapping, etc), but the music should be recognizable by human.
        • Again, you can use ffmpeg to convert your recordings into wav format.)
    2. Test your collection
      • Download the AFP toolbox.
      • Add necessary toolboxes to the search path by changing the statements containing "addpath" in afpOptSet.m.
      • Test your collection using "afpPerfEval.m" in AFP toolbox. Try "help afpPerfEval" to see the online instruction.
      • Put the result into a readme.txt file like this.
    3. Please upload the rar file to the server provided by TA. The top folder should named after your student ID, with two sub-folders named "music4db" and "music4query", and one file "readme.txt".

Audio Signal Processing and Recognition (音訊處理與辨識)