The Secret of speech recognition (SR)

Speech Recognition ( SR function is the standard feature of the foreign operating system, and the domestic operating system does not have this trait, and the national team has no relevant subjective motive force, to develop the actual available speech recognition system. Compared with foreign countries, the domestic operating system behind a large sections son, blame who?

How to make the machine recognize people's voice? In fact, the big truth is not complicated, first, let the machine equipment understand the vowel, and then understand the consonants, the two synthesized into words, and then do further processing. But, actually, it's complicated. What do you do?

Establish a "corpus" ( Corpus ) is the acoustic model library for building voice. For example, the pronunciation of the word "cat" must let the machine remember the "cat" pronunciation, try to let the machine hear the "cat" corresponding to the "sound" (so-called "biological signal") know that the owner said to it is "cat." Corpus is the establishment of the corresponding relationship between words and speech. This is the infrastructure, need to invest a lot of manpower, material resources, there is no other way.

any voice has personal characteristics, although - individual pronunciation is basically the same, however, there are some differences. So the machine's hearing always has the "error rate", which is inevitable. Reducing the machine recognition error rate is necessary, which is a basic understanding that the machine can not be too demanding. The machine's ears never compare to the human ear.

Microsoft, Google, and Baidu's voice-recognition giants all have their own tens of thousands of-hour corpus, but we don't, Linux There's no community . GPL Available Corpora ( Voxforge do not play with us). What to do? There is no automatic drop of pie in the sky. It is important to know that private companies have a huge input cost to build their own large corpus of copyright. And some of us like to drop pies in the sky, sitting in the yard all day with big mouths open to the sky and waiting for the pie to fall.

Shimen 7 Month - Day

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

The Secret of speech recognition (SR)