Briefly talk about voice evaluation (voice evaluation)

        Regarding voice evaluation and voice evaluation, it may not be a new concept, but judging from the popularity of the current preschool language education market, this is another topic that has to be mentioned. Although it is an old topic, there are so many application products on the market, which also shows how much people pay attention to it.

        The earliest speech evaluation technology is to use human voice and standard original sound to compare waveforms. Although the real-time rate is good, the accuracy rate is not satisfactory, and no matter what kind of words, the original sound is required, and the flexibility is very poor. Later, using speech recognition technology, a method similar to confidence calculation was used to analyze the degree of distinction between pronunciation and the given text answer, and the accuracy was naturally higher. At present, with the popularization of the application of neural networks, the technical level of voice evaluation has also been improved, and the accuracy has been improved.

        At present, among the various voice evaluation engines on the market, I personally prefer the online voice evaluation of iFLYTEK. It has both Chinese and English, and has high accuracy. The compatibility for children's pronunciation is also good, but it cannot be used offline. It means that it will not work if the network is disconnected. This is inappropriate in some applications, such as when children are doing homework, some parents are reluctant to provide the Internet to distract their children. Of course, there are also some environments where you must not be able to access the Internet. It is also inconvenient to carry out some studies and tests.

        It’s useless to talk about it, but I should also share the voice evaluation tools I use so far. Everyone knows the download path of iFLYTEK’s online voice cloud, so I won’t go into details. As for offline voice evaluation, I have used some experimental tools, but the performance is not good. It's great, but it can still handle some simple applications at present. Here is an Android SDK for offline voice evaluation. There is also an IOS version, but it is not used at present. This SDK seems to have an offline speech recognition function in it, but I haven't used it much, so please try it yourself.