Acoustic Systems Research
One of my principal research interests are acoustic systems that can be seamlessly integrated into our everyday environment, enabling collaborative spaces that enhance interpersonal interactions. As my main Ph.D. project, I led a five student team to demonstrate the first large-area flexible microphone array with a wallpaper format, involving novel algorithms, instrumentation circuits and transducers. This system is targeted for deployment in noisy and reverberant rooms, where multiple humans are speaking simultaneously. Using the spatially-distributed microphones, individual voice commands can be separated to enable collaborative human-computer interfaces.
AlGoRiTHMS: Blind Source Separation with
LArge-Area MicROPHONE Arrays
My algorithm accomplishes high-quality source separation in a practical reverberant room with practical speakers. Interfering sources are cancelled using a delay-sum beamformer stage with frequenecy-dependent time delays, followed by a binary mask stage. The algorithm is fully "blind", since it requires no prior information about the location of the speakers or microphones. This is enabled by a time delay estimator stage, which applies cluster analysis techniques from machine learning, to extract time delays for each microphone-source pair on the fly from the sound mixture of simultaneous sources. In this way the algorithm can adapt to the unique acoustic properties of each room (e.g., size, reverberation time, placement of objects) and to changes in location of the sources.
This work has been accepted to ICASSP 2016. Here is the manuscript. Below you can hear results for a 16 microphone array (width > 2 m) with four simultaneous sources. You can hear the initial raw recording of the 4 sources playing simultaneously, and the output of the algorithm once each source has been separated.
Simultaneous Sources (Recording before algorithm) | |
---|---|
Algorithm Output (Separated Source) | Source Playing in Isolation (For Comparision) |
Transducer Design: FlexiBLE, PIEZOELECTRIC Microphones
I have developed a flexible microphone based on a diaphragm formed from 1.5 cm (width) × 1.0 cm (length) PVDF (Polyvinylidene fluoride), a piezoelectric thin-film polymer. The film is clamped to posts, which standoff 1 mm from the sheet.To leverage the inherent translucency of the PVDF film, transparent electrodes are applied to both faces of the film by spray-coating silver nanowires, resulting in a clear, unobtrusive microphone.We have tuned the tension and dimensions of the PVDF diaphragm to design the resonant peaks to match human speech, which is concentrated from 500 - 3000 Hz. More details can be found in my invited JSSC 2016 paper.
Hybrid SYstems: Combining LARGE-Area Electronics and CMOS
We have developed a prototype system of a microphone array for source separation, based on a hybrid architecture. It combines large-area electronics (LAE), which enables a physically-expansive microphone array, and a CMOS IC, which provides superior transistors for readout and signal processing.Each channel consists of a thin-film transducer formed from PVDF, a piezopolymer, and a localized amplifier composed of amorphous silicon (a-Si) thin-film transistors (TFTs). More details can be found in my invited JSSC 2016 paper.
For each channel I developed a localized, two-stage differential amplifier. It is based on amorphous silicon (a-Si), thin-film transistors (TFTs). This material was chosen because it is processed at low temperatures (< 200 °C), allowing us to deposit it on flexible plastic substrates. I refined the Spice TFT models, designed and layed out the circuit, and manufactured it in our own cleanroom. Although the TFTs have a higher intrinsic noise than their CMOS counterparts, from an overall noise perspective they are still beneficial when compared to conventional amplifiers on the distant CMOS IC. This is because the localized amplifiers provide immunity to stray noise coupling, which the long LAE interconnects are susceptible to (e.g. 60 Hz).