Application of car voice technology in car electronic systems

In the era of intelligent cars, voice control has gradually become a standard feature. However, when using the air conditioner or opening the car window, voice control is often not smooth due to the influence of surrounding noise. Recently, NEC announced that it has developed a noise removal technology that can greatly improve the accuracy of voice control of the in-vehicle system.

It is understood that the technology uses two microphones to absorb the sound source, and then goes through two steps to process the noise, and enables the sound model to make it easier for the device to recognize. Through this technology, correct voice operations can be completed even in a noisy environment that is 5 times noisier than before.

The specific processing steps of this technology are as follows: first, by analyzing the location of the unique noise source in the car, such as the air conditioner, and the way the sound propagates to the microphone in the car, two microphones are arranged in the best position in the car with a small spacing, front and back. In this way, the noise from the front of the car and the voice of the speaker can be accurately identified, thereby improving the efficiency of noise removal.

Secondly, the detected speech and noise are processed twice. First, the waveform similarity of the noise entering the two microphones is used to accurately capture the waveform of the noise that changes over time and extract the main noise components. Then, the speech correlation in the two microphones is used to further eliminate the noise. Through this processing, most of the noise is removed.

Finally, since there may be speech distortion during the denoising process, which makes it difficult for the in-vehicle system to recognize it, NEC has also developed two speech models for adjustment. Specifically, the speech components and their changes are first modeled, and then the distorted speech is automatically selected from the two models, and finally adjusted to an easily recognizable speech.