voice-recognition - 如何在语音识别中使用arpa文件

Question

我使用 CMU SLM 工具包从文本文件创建了一个 ARPA 文件。

目前我不知道如何在我的项目中使用生成的 ARPA 文件而不是.lm和.dic文件。

如果有人知道这件事，请告诉我。

Answer 1

you use the probability of the language model when considering the "cost" of a word-transition in the search. :-) but that's probably not what you wanted to hear.

Your question is too open-ended.. what is your specific problem?

The dictionary and the language model are two separate items -- you can not convert one into the other.

The dictionary is used to tell the search what the valid words are and how they relate to phonemes / the phonetic transcription.

The language model is used during the recognition of an utterance, by using the probability of a uni-gram, bi-gram, n-gram .. when the search algorithm is considering a word-transition.

Edit:

check:

http://www-speech.sri.com/projects/srilm/manpages/ngram-format.5.html

http://www.ee.ucla.edu/~weichu/htkbook/node243_ct.html

http://www.ling.ohio-state.edu/~bromberg/ngramcount/ngram2fsm.html