1. Why "Tom's Diner," a Voice-Only Song, Was Good for Testing MP3
The song used to test MP3 was "Tom's Diner" by Suzanne Vega. It is an acapella track, meaning it features only the human voice with no piano or guitar accompaniment [1]. The reasons given for choosing this song are that it consists almost entirely of a single voice and has a wide range of sounds (from high to low notes) [1].
If there is only one voice, any distortion caused by compression does not get hidden by other instruments. Because the song covers a broad range of frequencies, errors from the compression process are easier to hear. But what is compression? To understand the "distortion" that was tested, we must look at how sound data is handled.
2. How MP3 Shrinks Music Data to About 1/11th of a CD
Sound on a music CD is measured 44,100 times per second. Each measurement uses 16 bits, recorded for two channels (left and right). Calculated together, this equals about 1,411,200 bits per second (or 1,411.2 kbit/s) [1]. MP3 at 128 kbit/s reduces this amount to about one-eleventh of the original size [1].
Let us compare a 3-minute song. On a CD, it takes up about 32 megabytes. As an MP3 at 128 kbit/s, it takes up only about 2.9 megabytes. However, this ratio applies specifically to the 128 kbit/s setting; changing the settings changes the size [1]. It is thought that even with this reduction, the sound quality remains good enough for normal listening [1].
So what is being removed? MP3 uses the human ear’s tendency to "miss" certain sounds. It discards components that are believed to be inaudible to most people [1]. The main example of this missing information is called auditory masking.
3. Why We Can't Hear Quiet Sounds Next to Loud Ones (Auditory Masking)
In a noisy classroom, you might not hear a child whispering next to you. When a loud sound is nearby, quieter sounds become hard to detect. This is auditory masking: a phenomenon where one sound interferes with the recognition of another [3].
There are two main types. One is frequency masking, where a loud sound hides a quiet sound that has a similar pitch (frequency). The other is time masking, where sounds are hidden in the short moments before or after a loud noise [3].
The important point is that the hidden sound still exists in the air and reaches your ear. You just do not notice it. If you cannot hear it, there is little value in recording it in fine detail. This idea is the starting point for MP3 [1]. But how does the technology handle these unheard sounds?
4. How MP3 Lowers Sound Precision Instead of Deleting Sounds
Rather than simply cutting out inaudible parts, MP3 adjusts how finely the sound is recorded. The encoder divides sound into several frequency bands (called critical bands) and estimates how much noise can be hidden in each band [2]. It then assigns fewer bits to bands where sounds are easily masked, and more bits to bands where they are not [2].
Fewer bits mean a coarser record, which creates small noise (quantization noise). However, if this noise stays below the level of masking, the ear does not notice it. So instead of "deleting inaudible sounds," the method is "lowering recording precision within a range where noise is unnoticeable." Some educational materials explain this by dividing bands into about 24 groups, but the number varies depending on the design [2].
Here lies a weakness in this method. If the estimate of what is hidden does not match how ears actually work, noise that should have been masked will still reach your ear.
5. How Distortion in "Tom's Diner" Tests Led to Fixes in MP3
The team developing MP3, including Karlheinz Brandenburg, reportedly tested with "Tom's Diner" around 1988 [4]. When they compressed this song using the MP3 method, the voice became severely distorted. The psychoacoustic model (the computer’s guess about how we hear) was not yet accurate enough [4].
Because there was no accompaniment, there were no other sounds to hide the errors. The inaccuracy of the model showed up directly in the voice. The team rebuilt the model, and one developer has said he listened to this song many times [4]. Note that sources differ on the exact year of testing; some say 1988, others suggest different years.
The refined format was published as standard ISO/IEC 11172-3 in 1993. The file extension ".mp3" was decided on July 14, 1995 [5]. It is important to remember that MP3 was not tuned only for this one song; it was just one of the tests used.
6. Why MP3's Lost Sound Detail Cannot Be Restored (Lossy Compression)
Methods like MP3, which reduce size by lowering precision, cannot return exactly to the original sound. This is called lossy compression. Once information is discarded, it cannot be recovered later.
There are also methods that can restore the original perfectly, such as FLAC, known as lossless compression. However, for music, this typically reduces data size by only about half, varying by song. It does not shrink to 1/11th like MP3 [1].
These two methods have different goals. MP3 aims to "make it very small within limits the ear cannot detect," while lossless compression chooses to "keep the original intact, even if the size reduction is modest." The limit of using what we miss is that individual hearing differences and listening environments change what is hidden.
7. An At-Home Experiment: Hearing Auditory Masking Yourself
You can check how ears miss sounds at home. Play music at a low volume. Listen to the same part of the song in a quiet room, and then in a room with the sound of a fan or ventilation system [3]. If you find places where small instrument sounds or breaths become hard to hear because of other noises, you have experienced masking. Not noticing it is not a failure.
Do not turn the volume up too high, and tell an adult before you start. When using machine noise, do not put it close to your ears. Changing the song or the room might change where sounds get hidden [3].