I'm developing an App that compare a voice speech with a text, sometimes the text can be a long phrase with 20 words. I take all extra.results obtained and compare this with the phrase. Different people with different accents will make the speech so I need to receive as many reults as possible. I have seen that always I receive 5 results as a maximum, I should get up to 10 results to avoid errors. I have tried with different phrases and with different sized phrases. Thanks!
Custom task require custom speech engine, in your case with a custom engine you can get much more accurate matching by configuring the expected words.
10 variants will not help you. If you have just 2 decoding variants in the sentence of 10 words, you need to retrieve 1024 different decoding results. You'd better have them in more compact form of confusion network or lattice.