Each of the factors discussed in the previous sections are important in the choice of speech recognition engine, and picking the right balance is largely dependent on the target application.
For a dictation application, the mode of operation has no choice but to be at least continuous speech recognition, and since the vocabulary is required to be relatively large, and the recognition rate high, the engine will almost certainly need to use a speaker dependent speech model.
If on the other hand the engine is required for an application involved in speaker verification, then a small vocabulary and connected word recognition will likely be sufficient, although speaker dependent mode is a necessity for obvious reasons.
The other end of the scale sees a simple device such as a television that includes speech recognition in order to turn on and off. This would require a tiny lexicon, resulting in high recognition rates with even a speaker independent voice model. Clearly for just two words, isolated word recognition would be adequate.