Signals

Digital signal processing provides us with many of the basic tools we need for analysing or generating speech signals.
Log in

The first step in digital signal processing is to get the signal into digital form. This involves converting an analogue (continuous) value into a digital (discrete) one. Discretisation in time is called sampling and discretisation in amplitude is called quantisation.

The second step is usually to analyse the signal in some way, and for signals that change over time, that requires short-term analysis.

Start with these blog posts

then watch the videos.

When we digitise a natural audio signal, we are converting from the continuous analogue domain into the discrete digital domain.
Log in if you want to mark this as completed
Each sample must be represented by a number, with fixed precision: this is called quantisation.
Log in if you want to mark this as completed
When a signal changes over time, as speech does, we need to analyse short sections of the signal one-by-one, not the whole signal at once.
Log in if you want to mark this as completed

Reading

When we extract short sections out of a longer, continuous signal, we need to be careful. The very act of extracting (or “cutting out”) a short section can create artefacts (specifically, discontinuities) at the beginning and end of the frame that is extracted.

An artefact is something created by the processing that is not part of the original signal being analysed. It’s impossible to avoid artefacts, but we can minimise their effect by applying a tapered window, which is basically a “fade in and fade out”.

(Note: “artefact” is the most common modern British spelling and “artifact” is more common in North America. Both are correct.)