【作者】 张杰；

【导师】 陈永耀；

【作者基本信息】 哈尔滨工程大学 ， 船舶与海洋工程（专业学位）， 2022， 硕士

【摘要】 近年来,水下设备技术的快速发展,已经有各种水下设备应用于水下作业中,对水下摄影、探勘、搜救等水下作业提供了很大的助力。由于海洋环境复杂、水下环境能见度低且存在海浪涌动,潜水员手动操作设备难度较大,因此结合语音识别技术的水下语音控制是一个良好的人机控制方式。现有的语音识别和水下通信设备使用较昂贵的换能器、传感器和高性能处理器等组件使设备价格较高,难以普及。因此开发低成本硬件平台实现水下语音控制,具有较高的实际意义和市场应用前景。本文设计系统的低成本硬件平台由面罩端和接收端组成,总成本不足二百元。面罩端采用具有硬件加速的k210开发板进行语音识别计算,提升了可靠性。通信采用读取预存储指令文件进行解码输出的方式,结合音频解码单元、低功耗的AB类音频放大芯片和改装的蜂鸣器等硬件低成本实现了指令发送功能。接收端采用低功耗且价廉的STM32核心板和MEMS水听器实现了指令信号的接收解调功能。在此硬件基础上,研究了低复杂度语音识别算法和水下通信实现。本文设计了自适应计算阈值的语音段捕获算法可以抵抗呼吸声等噪声影响。分析了面罩内语音信号的混响影响,采用适应该混响条件的梅尔倒谱系数(MFCC)为语音特征。采用低运算量的动态时间弯折算法(DTW)为特征匹配识别算法降低了算法复杂度。论文设计了基于DTW多语音平均模板训练法用于训练语音模板,提升了语音识别稳定性和识别率。水下通信采用汉明编码和2FSK数字调制方法结合进行信号调制。设计了由阈值检测和最佳匹配查找算法组成的低运算量同步头检测算法。设计了基于多帧载频能量检测的判决解调算法。对软件进行运算优化和空间优化降低算法复杂度,实现了稳定可靠的通信功能。最后进行语音识别测试,实验结果表明,识别正确率达98.3%。在消声室进行了空气声通信测试,验证了系统通信模块的硬软件设计的正确性。在消声水池进行了实验测试,分析了蜂鸣器的声源级和频率响应以及传感器的灵敏度。在具有较理想信道环境的信道水池进行水下通信测试,系统在30米通信距离下,指令接收均成功。更多还原

【Abstract】 In recent years,with the rapid development of underwater equipment technology,various underwater equipment has been used in underwater operations,which has provided great assistance to underwater operations such as underwater photography,exploration,and search and rescue.Due to the complex marine environment,low visibility of the underwater environment and the existence of surging waves,it is difficult for divers to manually operate the equipment.Therefore,underwater voice control combined with voice recognition technology is a good human-computer control method.Existing speech recognition and underwater communication equipment uses more expensive components such as transducers,sensors and high-performance processors,making the equipment expensive and difficult to popularize.Therefore,developing a low-cost hardware platform to realize underwater voice control has high practical significance and market application prospects.The hardware platform of the system designed in this paper consists of a mask end and a receiving end,and the total cost is less than 200 yuan.The mask side uses the k210 development board with hardware acceleration for speech recognition calculation,which improves reliability.The communication adopts the method of reading the pre-stored instruction file for decoding and output,and combining with the audio decoding unit,the low-power AB audio amplifier chip and the modified buzzer and other low-cost hardware to realize the instruction sending function.The receiving end adopts the low-power and low-cost STM32 core board and MEMS hydrophone to realize the function of receiving and demodulating the command signal.Based on this low-performance hardware,low-complexity speech recognition algorithms and underwater communication implementations are studied.In this paper,a speech segment capture algorithm with adaptive calculation threshold is designed to resist the influence of noise such as breathing sound.The reverberation effect of speech signals in the mask is analyzed,and the Mel cepstral coefficients(MFCC)adapted to the reverberation conditions are used as speech features.The low-computation dynamic time warping algorithm(DTW)is used as the feature matching and recognition algorithm to reduce the algorithm complexity.This paper designs a multi-voice average template training method based on DTW for training voice templates,which improves the stability and recognition rate of speech recognition.Underwater communication adopts Hamming coding and 2FSK digital modulation method for signal modulation.A low-computation sync-head detection algorithm is designed,which consists of threshold detection and best matching search algorithm.A decision demodulation algorithm based on multi-frame carrier frequency energy detection is designed.The software is optimized for operation and space to reduce the complexity of the algorithm and realize a stable and reliable communication function.Finally,the speech recognition test is carried out.The experimental results show that the recognition accuracy rate is 98.3%.The air acoustic communication test is carried out in the anechoic chamber,which verifies the correctness of the hardware and software design of the system communication module.Experimental tests were carried out in an anechoic tank to analyze the sound source level and frequency response of the buzzer and the sensitivity of the sensor.Underwater communication test is carried out in the channel pool with ideal channel environment,and the command reception of the system is successful under the communication distance of 30 meters.更多还原