Time-frequency filtering

In this chapter, we continue the discussion of joint time-frequency energy representations for speech signals. Here we shall make stronger assumptions about the form of the signals. We will introduce a particular model of the time-varying vocal tract, and define its ‘transfer function’, H(t,ω). We will show that time-frequency filtering can be used to estimate |H (t,ω)|2, a technique that is essentially a two-dimensional generalization of straight-forward, stationary methods. Further, we will see that |H(t, ω)|2 is closely related to the time-frequency representations of the previous chapter.