In recent years, domestic artificial intelligence technology has been changing rapidly, but some front-end technologies have few breakthroughs. With the popularity of products such as AI intelligent voice, people's human-computer interaction is gradually changing from physical interaction to biological interaction. And the first step of interaction-pickup, the effect is not satisfactory. For smart homes, their pickup range is mostly concentrated in the near-field environment, about 2-3 meters, and the pickup effect is often poor, requiring multiple wake-ups.
If you compare the camera to the "eyes", the pickups are the "ears". The combination of the image seen by the eyes and the sound heard by the ears constitutes a basic audio-visual recording system. However, this flexible ear will be affected by many factors such as sound pickup distance, indoor reverberation, environmental noise, etc., which is not a small challenge for sound pickup.
How to truly "listen to all directions"? IFLYTEK, which has been deeply cultivating the field of intelligent voice and artificial intelligence for many years, recently launched a new pickup brand-DiTing, which is striving for a place in ultra-low volume pickup and noise reduction. In fact, the products such as the iFLYTEK voice recorder, smart mouse, and Alpha Egg that iFLYTEK launched earlier all involve voice interaction. Based on the accumulated technology of product application, it is gratifying to work hard in the field of pickup this time.
It is understood that the iFlytech listening series of HKUST is equipped with 32 microphones, featuring fully automatic sound source localization, adaptive beamforming and reverberation suppression technology, as well as noise suppression and voice automatic gain adjustment algorithms based on deep learning, which can realize automatic indoor speaker Positioning, noise and reverberation suppression, automatic volume adjustment and other functions to achieve the purpose of precise sound pickup.
Automatically track the sound source, accurately pick up 30 decibels ultra-low volume
Recently, a technology blogger's evaluation video on pickups has attracted attention. In the video, iFLYTEK’s iFLYTEK series products and similar products in Sennheiser of Germany and Shure of the United States "compete on the same stage", iFLYTEK shows great success.
In an environment that simulates an ultra-low volume of 30 decibels that cannot be heard by human ears, Sennheiser’s sound pickup is stable and the content is clear. The sound picked up by Shure is small and it is difficult to distinguish the content of the speech. The content picked up by iFLYTEK is clear and the sound quality Hearing better.
This is mainly due to the use of self-developed automatic sound source localization technology by iFlytek. As long as there is a slight sound, it quickly locates the sound source like a spotlight and suppresses reverberation and noise from other directions. In practical applications, an array of 32 microphones can achieve 7×24 hours of all-weather, omni-directional, no dead-angle pickup, and accurately pick up an ultra-small volume as low as 30 decibels.
As we all know, sound will be attenuated during the propagation process, and different azimuths of sound sources will cause large differences in the volume and effect of the picked up voice. Fully automatic sound source positioning and adaptive beamforming technology also enable iFLYTEK to pick up moving sound sources. Outstanding performance. The beam can automatically "aim" the sound source direction of the movement just like a gunman shooting a prey. This means that iFLYTEK is a breakthrough for those devices that still need preset and restricted areas to pick up sound. Not only that, by automatically adjusting different volume levels, iFLYTEK makes the picked up sound more in line with the human hearing effect.
Overcome technical difficulties, DiTing has amazing noise reduction capabilities
The acoustic environment is more complicated than imagined. Noises such as environmental noise, interference noise, current noise and voice signals often overlap each other in time and frequency spectrum, plus the effects of echo and reverberation, and want to capture relatively pure voice very difficult. In the evaluation video, the evaluator simulated the environmental noise of 70 decibels and 90 decibels respectively, and the results showed that even in the extreme noise environment of 90 decibels, Di Listening suppressed the noise and the content of the conversation was still clear.
Faced with the challenge of noise, iFLYTEK can effectively enhance the voice and significantly suppress the impact of noise on the target voice based on information in the time, frequency and space domains. It first picks up the voice through sound location technology, performs voice enhancement, and achieves a preliminary noise reduction effect. Then through beamforming and deep learning-based speech enhancement algorithms and suppression of non-directional and directional noise, at the final output, the volume is automatically increased and optimized according to the auditory characteristics of the human ear to make the sound more full.
Core voice technology drives development, enabling multiple scenarios in the future
The era of intelligent connection of all things has come. AI-enabled IOT will inspire unlimited possibilities. The level of sound quality picked up by the front end will undoubtedly affect the level of later speech processing.
DiTing series products are the embodiment of HKUST iFLYTEK’s 21 years of firm core technology independent innovation. Relying on the belief that "Chinese speech technology should be the best by the Chinese", since 2018, HKUST iFLYTEK has won 30 international artificial intelligence competitions, covering speech recognition, speech synthesis, machine reading comprehension, gesture recognition, and images. Identification and many other fields. With the breakthrough of a key technology, HKUST also provided strong technical support for the implementation of diverse application scenarios of pickups. The previous sound pickup equipment has high cost, poor sound quality, and strong directivity, which cannot be promoted in a large area. The introduction of DiTing may break this phenomenon.
It is reported that Diting series products can be widely used in key places and key parts such as security, transportation, high-quality conferences and so on in the future. In terms of practical applications, it can be said to be promising. Taking public places as an example, most of the previous videos could not accurately pick up sound. The effective combination of audio and video, omni-directional collection of audio-visual solves the image blind spot of pure video, which is beneficial to prevent the occurrence of mass and violations and meet more realistic needs demand.
AI empowers all walks of life, further promotes the overall rise of social productivity, and profoundly reconstructs the economy and society. Liu Qingfeng, Chairman of the University of Science and Technology iFLYTEK, who has insisted on the "top indomitable" since its inception, once said, "Only by occupying the high point of core technology can we win the initiative in industrial development and have the right to speak in international competition." iFLYTE listened to it. Excellent noise reduction and ultra-low volume pickup are comparable to international first-class, and the prospect of domestic pickup technology is worth looking forward to.