I recently purchased a pair of high performance headphones. Not high performance by brand but by specifications. Yes, I know how to read them. They are rated for 10Hz to 30kHz and they are relatively flat, with variable tuning plugs to change their response curve acoustically. I decided to give them a little test run using various material in my home system. I was surprised by the results. Now first allow me to explain and indicate that this is by no means a double blind test. Yes, there are lots of things I could have done to improve it but I was still surprised with the results. I am purposely trying to leave product and brand names out of this post as the desire is to talk about the signal flow and process. It is very easy to get into the debate of is Brand A or Brand B the better product. Instead I am just talking about the signal chain and process. I am also going to not share the number thresholds I found, each person’s needs and opinions will be different. The entire point of this is to not let the bigger number be the better number, just because it is bigger.
The basic signal flow was the following:
- Source – 44.1kHz 16bit WAV files (1411kbps), MP3 153kbps (VBR) files, MP3 320kbps files
- Playback Engine – iMac based, Digital Audio Workstation Software (Adobe Audition CC 2014 & Audacity)
- Output Device – USB connected Digital to Analog converter running at 44.1kHz locked to computer sample clock. (D/A (24 Bit) 106 dB typical, A-weighted, 20Hz – 20kHz via headphone output
- Headphone Output – 1/4″ TRS for stereo converted to 3.5mm TRS via passive adapter
- Headphones – 10Hz to 30kHz passive devices in ear style with acoustic tuning plug at flat
So I tried a few different sample tracks. Ones that I had extracted as WAV files, same file extracted as MP3, original WAV files purchased directly from the artists, MP3 encoded by the artists. They were all well produced tracks ranging from full band rock and roll to acoustic pieces. The artists ranged for well established musicians (Peter Gabriel, Nine Inch Nails, and Robert Fripp) to less well known musicians (Jonathan Coulton, Marian Call, and California Guitar Trio). These were all tracks I am familiar with. What I would do was import the files into the audio editing software as stereo tracks, both the WAV files and MP3 files. I would place them on adjacent tracks that I could exclusively solo (screen captures at end of post). The sample rate of the project was set at 44.1kHz and 16bit to minimize coloring by the audio software resampling. This configuration allowed me to play both the MP3 and WAV track simultaneously and switch between them easily. The switches were typically very fast and with little artifacts. I found that the numbers of bits flowing had less of an impact than I expected at higher rates. I really like that many musicians are providing uncompressed formats, but Marian Call gets a gold star for providing me WAV files. (If you listen to her stuff, the typewriters are not sound effects they record them as part of the process.)
I was able to tell that there were differences between the compressed and uncompressed formats, no matter what the bit rates were for the MP3’s. However what I was more surprised was how subtle the differences were between each step or file in the process. I then took it a step further, I took the same WAV file I extracted from a CD as well as a purchased WAV file and created different MP3 streams. ranging from 320kbps to 32kbps. I used a batch converter, I did not go in and tweak each encoding as can be done with better audio editing software. I then loaded up all the files into both editing software packages and once again went through and used the exclusive or simple solo feature. I saw surprised at how far down the sample rate could be set before I found the music quality objectionable. This value changed based on the material I used. Let me say that again, the minimum bit rate value I found acceptable changed based on the material being used. Multiple points of diminishing returns were found. Yes, I understand what the numbers mean and how more data is typically better. But at the higher bit rates with good converters the differences were smaller than expected. As soon as I crossed a threshold, it was a point of no return. The number was lower than I expected as I had been applying my knowledge of the encoding processes previously, now I was just listening as objectively as possible. it also varied by the material as I indicated.
If I am listening to a podcast, does it need to be 44.1kHz, 16 bit, stereo for the human voice? I don’t believe so. Especially as most podcasts are just the human voice. Voice over IP studies have found most vocal information is in the 3,500Hz and under range. Transmitting at the higher sample rate is just wasting bandwidth and storage for the listener typically. But that is a discussion for a different time.
I still find and believe uncompressed audio files to be the best. Especially if one is tuning and adjusting an audio system. There were definitely shifts in the tonal and temporal qualities of the music. However for listening while traveling or as background sound, perhaps the lower data rates are the proper solutions. I do know that for my travel selection, encoding down to a more reasonable file size makes sense. I can place lots of music on the portable player. I am listening in an environment, especially when on an airplane, that is less than ideal. Yes, I am still keeping my music library as WAV files, yes those are my preferred format. However when I want to load up 8,561 songs for a ten day business trip onto my music player or laptop I am sated (not satiated – yes, there is a difference) with downconverting to MP3’s. I will still travel with WAV files for critical listening as well, often on an optical media.
So I encourage you to try this yourself. There is open source software such as Audacity that I used so that you can do your own tests. However like me, I believe that you will find that picking quality by the numbers of bits flowing is not always providing a full or simple answer.