What should be my training set for a word-spotting program? I wanted to create a simple voice recognition program. It should be able to distinguish my voice from other sounds. Aside from my voice, what should I add to my training set? Also, I want to extend my program to be a word recognition program.
Am I right in thinking that voice recognition is a type of classification problem?
EDIT:
For now I want to create a word spotting program. Example, I want to open google chrome if I have the input word "chrome".
 A: Gathering a speech corpus can be lots of work and depends on the exact sort of task you're doing, how large a vocabulary, how many speakers, languages, accents, different acoustics. The main reason Google ran their free GOOG-411 phone service between 2007-2010 was to gather such a corpus.

It should be able to distinguish my voice from other sounds.

If you mean it should recognize any voice in general but not background noise or other sounds, that's voice recognition (with background noise).
If you mean it should recognize your voice and not Alice or Bob, that's called speaker identification.

It should be able to distinguish my voice from other sounds. Aside from my voice, what should I add to my training set?

You tell us. What other sounds does it need to reject? Phone rings? Alarms? Background conversations? Wind? Car noise? Train noise?

Also, I want to extend my program to be a word recognition program.
  Am I right in thinking that voice recognition is a type of classification problem?

As in discrete word-spotting? Or must it understand every word in a sentence?
For example, if you say "I'm going out to a movie with friends, I'll be home at 8, turn on the heating", should it recognize every word, or just the words/phrase "turn on the heating"?
Yes, each of these is a classification problem, but the accuracy, computational complexity and the feature vector you use can vary hugely.
Again, gathering a corpus depends on what exactly you're trying to do.

EDIT: For now I want to create a word spotting program. Example, I want to open google chrome if I have the input word "chrome".

Then you need a corpus with you (and other people, preferably) saying words like 'chrome', 'home', 'time', 'comb', 'core', 'grow', 'crime', 'crumb', 'run' i.e. positive and negative exemplars.
