0
votes

I want to use CMU Sphinx 5 prealpha for realtime speech recognition (for detecting some hot words) and use this command:

pocketsphinx_continuous -inmic yes\
                        -hmm zero_ru.cd_cont_4000\
                        -dict ru-dict-min.dic\
                        -lm ru-cmds-min.lm\
                        -samprate 16000

I am satisfied with the accuracy, but I noticed a lag. For example, I started recognition and say "hello", Sphinx prints "hello" after 1-3 seconds (great!), I wait a few minutes and say "hello" again and Sphinx prints "hello" after 10-18 seconds. I wait a few minutes and say "hello" again and Sphinx prints "hello" after 3-7 seconds. Task manager shows that Sphinx uses 2-17% of CPU. How can I fix it? What's the problem?

I use very small files: ru-dict-min.dic has 28 words, ru-min.lm has 24 ngrams.

1

1 Answers

0
votes

Voice activity detector picks too much noise, that is why sphinx spends too much time on processing.