Non-uniform quantization

The next step after the voice is non-uniform quantization. The basic ideal of this quantization is coarsely quantize the large input and finely quantize the smaller input as ear is less sensitive to change at higher amplitude. 

This is the plot of the data sampled by the recorder:

In order to compare non-uniform quantization with uniform one, first we applied uniform quantization to the signal:

From the picture above we can see that the shape of the wave generally kept.

Then we applied nonuniform quantization:

The picture above shows the result of Mu quantization. The shape this really different from the original data. However, there is nothing wrong here. Nonuniform quantization, different from uniform quantization, need an extra step, that is expansion. The expanded data are showed in the picture below and this time it looks much more like the original data.

The picture below shows the distortion of the two quantization method. The latter one represent the nonuniform quantization's distortion. We can see that the non-uniform one will have a less distortion.