October 12, 2020
In this this article all data will be referenced in bits using decimal prefix aka 1 kilobit (kb)= 1000 bits and 1 byte=8bits
In music you hear the debate compressed versus lossless music. Audiophiles will say that you can hear the difference between compressed and uncompressed music. Then someone runs a listening test such as ABX or others outlined in standards such as ITU P.800 (it is a free download), which quantitatively generate no difference to minimal difference in the metrics between compressed and uncompressed media. I am going to start with the basics what is compressed music?
There are three ways to store media content uncompressed, losslessly compressed and lossly compressed. First, uncompressed is the simplest way to store media. It is the exact captured content stored in a file unaltered. Some example of uncompressed audio format include WAV and AIFF. Second, lossless compression is a format that is reduced in sized but, when uncompressed the media is numerically identical and indistinguishable (aka each sample is exactly the same across the entire audio file). Examples of a lossless audio format are FLAC and ALAC. The third and final type is lossy compression. Lossy compression causes data to be lost and is not identical to the uncompressed audio format Examples of a lossy audio format is MP3 and Ogg.
|Codec Name||WAVE & AIFF||FLAC & ALAC||Ogg,MP3 & AAC|
|Common File Ending||.wav, .aiff or .aif||.flac & .m4a||.ogg, .mp3 & .m4a*|
*.m4a is a container and can hold an audio file encoded in losslessly or lossyly
The quality of a lossy audio format is determined by a comparing how indistinguishable the sound or metrics of the compressed audio to the uncompressed audio in a listening test or listening test derived metric. It is important to know, this is relatively complex to truly achieve. Agata Rogowska investigated how using different musical instrument samples (stimuli) where distinguishable based on LAME-MP3 (the MP3 encoder used by audacity) encoding. He found the low bitrate (32kbps) or highly compressed MP3s of horns from the Sound Quality Assessment Material recordings for subjective tests (EBU-TECH 3253) was indistinguishable from the uncompressed horn sounds.
However, a viola sound could be distinguished almost 95% of the time . Even when moving to the generally recommended bitrate of 128kbps  Viola was still distinguishable 65% of the time. This is where 50% distinguishable would be a random guess. Almost two years after this investigation, the creator of MP3 would recommend that people stop using MP3 .
The question that is overlooked in this discussion —Why are we even using compressed audio formats? Internet speeds having drastically increased since the MP3 codec was introduced in 1993 . Storage cost have drastically decreased. For instance, according Matthew Komorowski article, “A History of Storage Cost (Update)” it cost $250 per Gigabits ($2000 per GB) and $0.00375 per Gigabits ($0.03 per GB) in 1993 and 2015 respectively. Let’s examine storage cost of a CD, a CD at max can hold 80 minutes of audio. It cost a little bit less than $0.03 dollars to store a CDs worth of audio and under a $0.01 to store a CD as a 128kbps mp3. Now, how much does it cost to transfer the content of a CD over the internet? That can be estimated two ways. The first way is by using the data cap for Xfinity home of 9.6 per Terabits (1.2 TB ); then find the minimum internet speed to be able to hit the datacap in an average length month. which is about 4 Mbps. The slowest internet speed that Xfinity provides is up to 25Mbps for $49.95 per month (retrieved August 30th 2020). That means at maximum it cost $0.0005 per Megabit to transfer the content of a CD over the internet. The second way to determine the cost per megabit is to determine the maximum data that could be possibly be transferred to a home in an average month, then determine based on the monthly cost how much it would cost. As you know, for an additional $30 per month you can remove the data cap . The fastest plan Xfinity provides is a 2 Gbps plan for $299.95 (retrieved August 30th 2020) plus the additional $30 per month. Using this calculation, it comes out to $0.00000000006 per megabit to transfer the content of a CD over the internet. Continually focusing on cost, the average active Spotify user listened to 25 hours of music per month in 2017  which would cost (internet transfer cost can be neglected) $0.86-0.07 per month depending on the quality settings in storage alone. If Spotify premium were to stream in CD quality uncompressed it would cost $3.81 in storage. The average active user only listen a short time relative to the hours in an average month. Therefore, dedicate storage per user can be reduced by the ratio average hours listened to per month since it can be reused across users. This brings the uncompressed audio cost down to $0.13 per month, which is 1 percent of monthly revenue.
Media content is also streamed on Youtube, Youtube has stopped reporting streams overtime. The most recent data for peak plays in a week is Despacito by Luis Fonsi featuring Daddy Yankee is 157.307 million views in the 28th week of upload (this was captured by Engauge graph digitizer from a drawing of the plot). By manipulating the numbers, you can find that the peak plays per minute were approximately 15,600. Using youtube-dl to estimate the size (I did not download and would never condone acts of piracy) of all the □.mp4 video resolution, I summed the storage space for all supported video only tracks” and added the □”.m4a audio for each supported resolution. Youtube pre-encodes all supported resolutions , so in total Youtube needs 398 Mb per viewer. The social media stats website Social Blade estimates at the time of writing (september 9th 2020 noon) 7 billion views generated between $3.5-$27.8 million dollars . Taking the middle of these revenue generated estimates a view is estimated to be worth $0.002. Assuming that the plays per minute all count as views. The peak plays per minute generated is $3,120 per minute in revenue and it cost $23.28 per minute (using internet bandwidth and storage) to serve the videos. Performing the same analysis with Spotify and adding raw CD quality audio (this could be shrunk using a lossless codec like FLAC), it would only take 121 Mb at a cost of $7.08 to serve versus $2.11 without lossless support. Or simply stated, it is a cost difference of 0.15% more of the revenue generated during peak playback. After examing cost, I ask the question again, why are we even using compressed audio formats? I am simply asking Streaming Services to let customers choose to listen to lossless quality audio for less than 0.2% of revenue.
Why Should I the Consumer Care?
As a consumer you should care because, you are not getting a discount on the reduced bandwidth usage or not having to ship the product. For instance searching on Amazon for Luis Fonsi’s album Vida I find that the CD cost $10.99 and the MP3 download cost $9.49. In more extreme cases ScHoolboy Q’s album CrasH Talk the CD cost $7.58 and the MP3 Download cost $9.49. In the most extreme case presto classical was hosting a sale on High Resolution Music (lossless music recorded at a higher sample rate and higher bit depth than CD) where the 96kHz 24bit was less expensive than the 320kbps mp3. Remember the lossless audio requires over 14 times (a little less when encoded in FLAC 7-8 times ) more storage than the MP3. The price difference between lossless and lossy codecs is relatively arbitrary. In many cases, a consumer has to buy a CD to obtain lossless audio even when they would prefer a lossless download. Why don’t music retailers and labels provide this to customers who want lossless audio?
The consumer is still bearing many hidden costs for being forced to use compressed music with non-FOSS (Free and open-source software) every device that can play these propriety Codecs is required to pay a licensing fee which is then passed on to the consumer. Compressed audio also trades storage space for power usage, as it requires more computation power to playback compressed audio.
Another hidden cost that many streaming devices and Bluetooth headphones is the native wireless protocols use lossy compression. This means that your lossly compressed audio could have sounded fine with one stage of compression but, could be drastically affected by a second round of audio compression. Also, many of the artifacts heard on internet calls are caused by lossy audio compression. Using lossy codecs increases latency by having to wait for the audio to encode and decode. This encoding time changes depending on the content which can cause video sync issues. One of the biggest shames is that we by licensing fee or government grants are still developing new lossy codecs that in general we do not need or use. Notice that most music stores sell you mainly MP3 even when the creators of the MP3 prefers that you use their new lossy audio creation . What other problems could be solved instead of research new audio compression techniques for minimal gains?
- “Audibility of lossy compressed musical” by Agata Rogowska
- “Which MP3 bit rate should I use?” by Nate Lanxon
- “The MP3 Is Officially Dead, According To Its Creators” by Andrew Flanagan
- Matthew Komorowski personal blog “a history of storage cost (update)”
- https://dataplan.xfinity.com/faq/ (retrieved August 30th 2020)
- “Average monthly time spent listening to Spotify content among monthly active users worldwide from 1st quarter 2015 to 4th quarter 2017” by Amy Watson
- “How YouTube Works – Computerphile” on the Computerphile youtube channel.
- Social Blades statistical information on Luisfonsivevo channel
- “FLAC compression level comparison” by Nathan Zachary