基于GSP算法和增强现实技术的景观交互系统设计

Design of the landscape interaction system based on the GSP algorithm and the augmented reality technology

针对城市公园景观存在多人语音识别耗时长、识别准确度低,导致景观交互效果不佳的问题,结合GSP算法和增强现实技术,提出设计一个景观交互系统。首先,将系统分为语音采集与录入、语音分离模块和语音播放及可视化展示模块;然后在语音分离模块中采用一种语音分离与HMM相结合的语音增强算法,通过此算法对含噪语音进行混合状态解码和语音估计。实验证明,在信噪比为0 dB、5 dB、10 dB和15 dB四种不同的高低信噪比下,感知评价指标PESQ得分均有所提升。当语音信号长度为3 s,语音状态数和噪声状态数分别为100和4时,提出算法的平均耗时保持在18.55 s范围内,平均耗时较低。由此说明,提出的算法可有效去除噪声,通过此算法能够降低计算量,提升计算速度和识语音别准确率。系统验证发现,设计的系统功能模块完善,可实现城市公园景观多人语音识别和分离,景观交互效果显著提升。

In view of the problems of long multi-person speech recognition and low recognition accuracy,which lead to poor landscape interaction effect,a landscape interaction system is proposed combined with GSP algorithm and augmented reality technology.Firstly,the system is divided into speech acquisition and input,speech separation module and speech playback and visual display module;then a speech enhancement algorithm combining speech separation and HMM is used for mixed state decoding and speech estimation.The experiment proved that the PESQ scores were improved in four different SNRs:0 dB,5 dB,10 dB and 15 dB.When the speech signal length is 3s,and the number of speech states and noise states is 100 and 4,respectively,the average time consumption of the proposed algorithm remains within 18.55 s,and the average time consumption is low.This shows that the proposed algorithm can effectively remove the noise,through this algorithm can reduce the calculation amount,improve the calculation speed and the accuracy of speech recognition.The system verification found that the designed system functional module is perfect,which can realize the multi-person speech recognition and separation in the urban park landscape,and significantly improve the landscape interaction effect.