As mentioned in my previous post the audio quality perceived in each of the clips composited in my third artefact differ to a large extent. There are two main audio codecs at work here, they are ‘MPEG AAC Audio’ (mp4a) and ‘Vorbis Audio’ (vorb).
MPEG AAC Audio is the successor to the MP3 format and allows for higher quality compression at lower bitrates (Brandenburg, 1999). Due to patent holders of the MPEG format encoders insisting on royalties and other payments for use of their encoders (Robertson, 1998) open source companies began developing their own encoders that would be free for use by all. Vorbis has emerged from one of these projects and now allows pieces of work to be compressed and more widely distributed through platforms such as YouTube. This is where we see that through official music channels on video sharing platforms mp4a is the codec used and unofficial videos often use vorb.
When looking at the waveform for the audio of my artefact you can clearly see drops in loudness levels from around 0:32 to 0:36 and 0:39 to 0:42. All three of the songs that were playing at these points were encoded using MPEG AAC Audio. As there were other songs using mp4a that did not drop in loudness we cannot assume that this codec is the cause. However no settings were changed when importing each of the video files from YouTube into Blender so there must either be these significant differences in sound level in the source videos themselves or during the download or importing of the video files some sound level change was automatically made. As Muldoon (2015) points out different video uploaders set their peak levels to different dBFS. There is a range of standards used from -12dBFS to 0dBFS. Muldoon states that YouTube are looking at normalising music videos across the board to account for these different audio levels but it appears that change has not materialised yet. For now when compositing multiple videos the audio will have to be checked and normalised individually to ensure a continuous loudness level throughout.
Brandenburg, K. (1999) MP3 and AAC Explained. Paper presented at AES 17th International Conference on High Quality Audio Coding. Retrieved from https://www.iis.fraunhofer.de/content/dam/iis/de/doc/ame/conference/AES-17-Conference_mp3-and-AAC-explained_AES17.pdf
Robertson, M. (1998) Fraunhofer Lowers Patent Boom On MP3 Software Developers. Retrieved 13 December, 2016, from https://web.archive.org/web/20000816002547/http://www.mp3.com/news/095.html
Muldoon, K. (2015, 24 April) The Right Audio Levels for YouTube. [Weblog]. Retrieved from http://www.kevinmuldoon.com/audio-levels-youtube/
Xiph (2016) About Xiph. Retrieved 13 December, 2016, from https://www.xiph.org/about/