Since the main purpose of voice wake-up is to activate the device to enter the interactive working state, the wake-up rate becomes extra important. In theory, the best state is when I say the wake word once and the device responds immediately. However, due to various factors such as the distance between the speaker and the device, special database pronunciation standards, noise, etc., it is difficult to achieve the perfect working state of the sub-response in actual work. Generally speaking, in the quiet scene at close range, the technology on the market can achieve a wake-up rate of more than 95%. However, the actual usage scenarios of consumers are more complex.
Therefore, it is necessary to simulate different application scenarios for actual measurement to obtain the final wake-up rate indicator. 2. False Wake Rate Refers to the probability that the device is woken up by non-wake words. The higher the false wake-up rate, special database the worse the effect. Everyone may often encounter in their daily life, the smart speaker at home, inexplicably answering "I'm here" at night, scaring people to death; and if the smart TV used at home wakes up the mobile phone voice assistant, Instead of the same wake-up word, the TV also pops up with an "I'm here.
This is a typical false wake-up scenario, which is generally measured by how many times it is falsely woken up in 24 hours. 3. Response time It mainly refers to the time it takes for the device to give feedback after the user finishes speaking the wake-up word. Generally, the shorter the better. The form of feedback is more diverse, which can be responded special database by voice response, or by bright screen or flashing indicator light. According to statistics, if it is a bright screen or indicator light flashing response, the user experience is better if it is controlled within 200ms; if it is a voice response response, it will be bad for the experience if it is too fast. Generally, it is better to control it at about 500ms.