Hello, welcome to Qooco Chinese Learning Website! Since it is your first time to be here, let me show you around!
|
and we will help you as soon as possible.| About ASR |
 What is ASR? Why do we use ASR instead of waveform comparison? Why do we use ASR? What does the ASR difficulty level mean? What can you do to help us build more accurate acoustic models? |
| What is ASR? | |
| ASR: Automatic Speech Recognition Automatic speech recognition is the process by which a computer maps speech signals to text. A speech recognizer's acoustical model is created by taking several thousand people's standard speech patterns and then compiling the patterns into a database i.e., it summarizes several thousand different kinds of correctly pronounced waveforms and uses a generalized model to give a pretty fair and accurate judgment regarding the user's pronunciation. Its result is not limited to one person's or one teacher's pronunciation. Because speech recognition uses a precise mathematical probability model named as the acoustical model, it can give a high degree of recognition accuracy, not only with sentences, but also with words and even phonemes. |
|
|
|
| Why do we use ASR instead of waveform comparison? | |
| The waveform comparison is carried on the user's speech waveform and a teacher's standard speech waveform. Its only merit is that it is easy to see. Its shortcoming: It is very inaccurate and highly misleading. First, many unrelated sounds can appear to make similar waveforms. Second, although different teachers' speech waveforms vary, their pronunciations are still correct. When only one teacher's standard speech is given, it is unable to cover all possible correct pronunciations. Even if a user's pronunciation is correct, but just because if it is different from some teacher's speech waveform, it will be judged as wrong. Simply put, a waveform visualization is an inappropriate tool to evaluate pronunciation.
When ASR judges the user's pronunciation, it is not based on pre-recorded audio; it is based on the precise mathematics probability acoustical model to judge the degree of quality of the user’s pronunciation. |
|
|
|
| Why do we use ASR? | |
| Using a Speech Recognizer provides significant value to many more learners at a fraction of the cost of live instructors. Currently, qualified live instructors are in scarce supply everywhere, and in many cases they are simply unavailable, which leads us into our next problem: most people attempting to learn a language do not have the environment in which to speak that language. Our Speech Recognizer plays an invaluable role in our application in that it forces the student to speak. When a student speaks aloud, the Speech Recognizer uses its existing data to try to understand not only what was said, but how it was pronounced. The only way you can learn to speak any language is by actually speaking-speaking with our animated characters to simulate a real life conversational environment will only improve your spoken language skills. | |
|
|
| What does the ASR difficulty level mean? | |
| You can adjust the difficulty level of the ASR to make the speech recognition tougher or more forgiving. We suggest you start from easy which will allow you to say words imperfectly yet pass, the purpose behind that is to build confidence. When you have more confidence speaking the words then switch to intermediate or hard to make your speaking drills more demanding and accurate. | |
 |
|
| You can select ASR level setting from the pull-down menu on the top page as shown in the following picture: | |
 |
|
|
|
| What can you do to help us build more accurate acoustic models? | |
| When you find some words or sentences that are hard to recognize no matter how hard you try, please send those voice files to us for analysis in order to help us build more accurate acoustic models. We will continually update our acoustic models based on more voice samples. To send us your voice files, please follow these 4 steps:
1) Go to C:\Program Files\Qooco\Chinese\enginea or the location you installed the Qooco client; 2) Create a zip file of the entire "yourvoice" folder, which should include some wave files and a text file; 3) Send the zip file to voice@qooco.com; 4) We will continually update our acoustic models to make our system better with your help. IMPORTANT! Remember, only your most recent completed exercise gets saved in "yourvoice" folder, so if you notice a problem in a exercise, be sure to create the zip file right away before you move on to another exercise. |
|
|