Effects of Quantization on MP3 Audio

The MP3 hybrid filter bank utilizes the psychoacoustic model to allocate bits in the encoding process in such a way that the quantization error is imperceptible to the human ear. While the MP3 model is generally successful in this task and enjoys widespread support in industry, its most basic implementation allows for unwanted artifacts in the reconstructed process when subjected to specific input signals. I characterize these artifacts for the inputs of castanets, wideband male and female speech, and modern indie music. I show that harmful auditory effects like pre-echo, reverb, and flanging can all arise from quantization error, but their presence is dependent on the type of input signal being encoded. Moreover, I show that the effects of quantization error can be reduced by adjusting the length of the MP3 MDCT windows or the magnitude of the psychoacoustic masking threshold. In these cases, I will show that bitrate is sacrificed for performance.