Highly sensitive, biomimetic flexible piezoelectric acoustic sensor for flexible user interface system

Abstract

PNZT) was adopted with the optimized material/dimension design. Deep learning-based multi-channel NPAS showed the outstanding improvement in speech processing including speaker recognition and speech enhancement under noisy condition, compared to a conventional microphone.

Chapter 3. For the personalized Artificial Intelligence of Things (AIoT) services, the electronics should accurately respond to the command via the voice recognition of deep learning (DL) algorithms. In addition, to stimulate the auditory nerve of patients with the deafness, the hearing aids should be developed for amplifying the mechanical vibrations over high-frequency range. The detected voice intensity in the range of 6 – 8 kHz is crucial for the accurate speech recognition. The fundamental resonances should be designed with the additional dimension variation of f-PAS, since the most informative voice data such as the voiced consonants of /s/, /z/, /$\theta$/, /$\eth$/, /f/, and /v/ is located above 6 kHz range. Since the electromagnetic interference (EMI) noise is a critically undesirable signal to affect intrinsic speech information, such as the fundamental frequency, harmonics, formant, and spectrum, the inherent less noise-interfered voice signals of hardware sensor should be acquired. Herein, we have demonstrated a thickness-changed (gradual and step type) piezoelectric acoustic sensor (GPAS and SPAS) with flexible PMN-PNN-PT membrane by mimicking the thickness structure of human basilar membrane for full-phonetic spectrum coverage. While the gradual PI substrates were fabricated vi the molding process of vanish solution, the UV laser ablation method was adopted to form the 7 step-structured film. In addition, to compensate the high bending rigidity of thick PI parts, the PMN-PNN-PZT was optimized to have almost 2 times higher piezoelectric properties than PZT. In addition, the differential electrode design was successfully demonstrated to show the opposite piezoelectric signs of splitted IDEs. Via the differential signal processing, high-frequency electrical noise was removed. Finally, the recorded signals of GPAS and SPAS were utilized to process the speaker recognition, and transmit far-distance voices for hearing aids, respectively.

Chapter 4. Since the VUI systems are applied in social communications, business meeting, and court trials, the acoustic sensors should have the capability of far-distant voice recording and multi-speaker identification. Therefore, the accurate speech separation is required to improve the speech recognition rate by producing the clean signals of target voices without the undesirable interferences. The human ear can easily detect what each people said simultaneously even in the noisy condition, due to the frequency-selectivity. Recent research reported the sensitivity and frequency-selectivity are attributed to tectorial membrane (TM). However, the high frequency-selectivity (low damping) induces the narrow bandwidth (high quality factor), which should be compensated by the high piezoelectric properties for full-phonetic coverage. Furthermore, for the real-life VUI applications, multiple f-PAS with the uniform properties should be fabricated based on the large-scale process. Herein, we successfully demonstrated the large-scale fabrication of tectorial membrane-inspired flexible piezoelectric acoustic sensors (TM-PAS) via the parylene-C deposition on La-doped PZT (PLZT) surface. The optimized PLZT film showed the 1.5 times higher piezoelectric voltage coefficient compared to PZT, which indicated the capability of detecting more far-distant voices. To obtain the uniformly detached PLZT membrane from a rigid 4-inch wafer, the coupling agent was treated on the surface before the parylene coating. The analysis result of finite element method (FEM) simulation suggested the higher sensitivity and frequency-selectivity, which was verified by the measured TM-PAS properties. At last, the speech signals of TM-PAS were successfully separated from the mixtures with undesirable data to each user voice.

Chapter 1. With the advent of hyper-connected society, the researches of smart flexible sensor systems have been reported to demonstrate the sensing/recognition interface. The voice user interface (VUI) has been spotlighted as the most intuitive human–machine interface, replacing conventional touch-based electronic systems. Due to extremely convenient and bilateral communication, smart acoustic sensors are the core technology of internet of things (IoT) and artificial intelligence (AI) for speaker recognition, biometrics, personalized AI secretary, and smart home appliances. Recently, flexible piezoelectric acoustic sensors (f-PAS) have attracted significant attention to improve the sensitivity and recognition rate by mimicking the resonance mechanism of basilar membrane in human cochlea. Narrow part (base) intensively responds to high frequency sound, while low voice frequencies can vibrate the membrane at wide region (apex). The flexible piezoelectric membrane of trapezoidal shape can detect minute sound from far distance by generate the electrical voltage signal via extreme resonant vibration according to voice wave. The self-powered f-PAS shows higher sensitivity over human utterance frequency range compared to commercial condenser microphone, because this capacitive sensors belongs to non-resonant type designed to resonate above voice spectrum for flat frequency response. In addition, analogous to 3,500 inner hair cells, the f-PAS can produce abundant voice information for speech processing depending on channel width, due to multiple IDE channels. The highly sensitive piezoelectric membrane can consistently exhibit above merits without degradation over humidity and heat, via the durability of inorganic thin film materials. In this thesis, the advanced biomimetic flexible piezoelectric acoustic sensors were demonstrated by considering the material effect (piezoelectric coefficient, modulus, and density) and mechanical design (width, thickness, and pores).

Chapter 2. In the era of artificial intelligence of things (AIoT), enabling the signal detection and recognition/prediction simultaneously, machine learning-base flexible piezoelectric acoustic sensors (f-PAS) have attracted intensive interest as a potential key technology of voice user interfaces (VUI). However, the most challenging obstacle for real-life application of highly sensitive biomimetic f-PAS is signal distortion issue caused by the fundamental difference compared with the commercial microphones. In addition, previous f-PAS exhibited narrow frequency response up to 4 kHz, which is insufficient to cover the full phonetic region of human voices (0.1 ~ 8 kHz). Here, a noise-robust flexible piezoelectric acoustic sensor (NPAS) was fabricated by locating the multi-resonant bands outside the noise dominant spectrum. To broaden the voice coverage up to 8 kHz, an advanced piezoelectric membrane (Nb-doped PZT