Generative models
Time for a conceptual leap: we will assume that the speech to be recognised really was generated by our model.
The model
Once we have an initial conceptual understanding of generative modelling, we can develop this simple generative model of variable-duration sequences.
Properties of HMMs
The special properties of HMMs make them very simple to use. Later, this will help us create elegant algorithms for doing computations with HMMs.
Algorithms for recognition
The model is only half the story. Now we need to perform computations with it. We'll start with recognising a test observation sequence.
If you’d like to read a little further (beyond the scope of this course), here are some suggested readings.
Reading
Holmes & Holmes – Chapter 11 – Improving Speech Recognition Performance
We mitigate the over-simplifications of the model using ever-more-complex algorithms.
Furui et al: Fundamental Technologies in Modern Speech Recognition
A complete issue of IEEE Signal Processing Magazine. Although a few years old, this is still a very useful survey of current techniques.
