Question

I’m quite curious about this.

In a broad way, how does one go about doing the following:

• Detection of word separations.
• Detection of syllables.
• Compensate for normal speech word connections.

Answer 1

This is subject of extensive research until these days. One usually starts building a model based on linguistic analysis of the language you’ll do recognition in and detect all the cases for word separations and syllables. Then recognition is mainly done using Hidden Markov Models over the signal.

Here are some references that might give you some better ideas:

http://lands.let.kun.nl/literature/eric.2004.2.pdf

http://www.asel.udel.edu/icslp/cdrom/vol4/778/a778.pdf

http://en.wikipedia.org/wiki/Speech_segmentation