Voice recognition on mobile devices

Recently, speech recognition applications on mobile platforms have become very popular. There are Siri and Google Voice Search abroad, and domestic speech input and control functions such as the web browser digging finance and UC are available. Today, let's try it out. I feel that this type of technology has reached the stage of large-scale application.

Previously, the mobile phone also had a function similar to voice control. Remember that Nokia 6630 was the first to use a button for specialized voice Instruction Input. However, this type of application is very simple and only supports limited command matching. Now, thanks to the addition of the cloud, voice recognition is very accurate and has amazing results. People who do not believe it can try to dig for financial speech recognition. Therefore, the current speech recognition applications are basically completed in the remote cloud, and mobile devices are only used to collect and send voice data.

So much, I am interested in mining such applications is how to achieve such a high recognition rate, this discovery, it is based on xunfei (http://www.voicecloud.cn) to complete speech recognition, it sends voice data based on the APIS provided by xunfei, and xunfei returns the recognized text. In this case, we can also make some speech recognition applications based on xunfei. Xunfei is such a platform company that provides interfaces and has good support for developers. In the afternoon, it took less than an hour to create a good speech recognition application on mobile phones.

1) apply for a developer account

Apply for an account number on http://dev.voicecloud.cn/and you will be given an app ID. This approval process is very fast and will take less than 10 minutes.

2) download the SDK

Http://dev.voicecloud.cn/download.php? Vt = 1

I downloaded the android SDK.

3) Compile and debug the demo

The downloaded SDK has a sample directory, and the following is an application of mscdemo, which is directly imported into eclipse.

If there is an error during compilation, delete the line with the error.

I have to debug it on the real machine because it has not solved the problem of voice input failure in the android simulator.

4) run the demo

After a simple trial, we found that the results were indeed good.

Voice Input prompt:

Sentence recognition:

Escape some bad terms (I say "fuck "):

It is quite interesting. It can be seen that xunfei's background has a powerful corpus with certain AI capabilities.

However, no information has been found on how xunfei charges fees. If it is necessary to make a commercial use, you should take a good look. If you have time, you can study how the speech recognition cloud is implemented and try to build a small cloud service on your own.