► We propose parylene C coating serves as surface anti-adhesion layer in PDMS double casting technique. ► Microstructure copies with different aspect ratios and sharp angles can be fabricated from ...master by using this new method. ► This new method is environment friendly. ► A single coating of parylene C can keep anti-adhesive property on its substrate for long lifetime regardless of the number of replica molding cycles.
As a simple method to fabricate a high quality copy of master, PDMS double casting technique has been more and more popular in microfluidics chips and bioMEMS application. In this work, the method in which conformal coating of parylene C serves as a demolding anti-adhesion layer in PDMS double casting technique was proposed. First casting was carried out onto master mold to generate negative PDMS mold and second casting was done onto negative PDMS mold likewise to generate positive PDMS replica with the same structure as master mold. Microstructures with aspect ratio from 4:1 to 20:1 and sharp angle from 5° to 40° were successfully obtained by using this new method. Experiments show replicas remain high fidelity to their masters. This new method of surface anti-adhesive treatment is environment friendly. Moreover, a single coating of parylene C can make the treated mold keep its anti-adhesive property for long lifecycle regardless of the number of replica molding cycles.
Platinum black coating can effectively improve the performance of MEAs (microelectrodes array) in neural signal transduction, though its lack of adhesion strength and durability tampers its usage in ...long term experiments. Here a new method of composite electrodeposition provides highly adhesive platinum black coating that enables MEAs for a month's long task and repeatable utilization. The new method was compared with present techniques on multiple aspects, e.g. actual surface area, surface morphology, interfacial impedance, durability and real application tests. Results show that the new composite coating provides greatly improved durability without compromising its performances. Neural cells were cultured on these MEAs for 40 days in vitro and spontaneous action potentials with high signal/noise ratio were recorded. Theoretical model and simulation provided preliminary understanding on the mechanism of this strengthened platinum black coating.
A novel linear microprobe array (LMPA) has been developed by a conventional microfabrication method from silicon. The LMPA leverages the properties of conventional microwire with additional features ...of naturally formed regular spacing. With the help of periodic microprobe arrays and double-side V-grooves fabricated in advance between each pair of the two micro- probes' rear ends, the number of microprobe units for assembly in one array can be flexibly chosen by cleavage fracture from the LMPA. The fabrication method was demonstrated and the prototype device was assessed by electrochemical impedance spectroscopy (EIS) and in vivo test. The SNR of the spikes recorded was 6.
Multi-electrode array is an important tool in the study of neural-network, cognition, remembrance, as well as brain-computer-interface, etc. Fork-like 32-site microelectrodes are developed with ...silicon. By use of integrated circuit technology, the length of the electrodes, the area of the recording sites, as well as the spaces between the sites are closely controlled. SiO2/SiNx/SiO2 composite dielectric membrane and Pt black are introduced to improve the characteristics of the electrodes. The whole thickness of the thin-film probe was 21 p.m. By combining the modifying process with the micro-fabrication method, this kind of silicon based microelectrode satisfies high-density recording and the performance characterization is evaluated by test in vitro and in vivo.
Self-supervised learning (SSL) achieves great success in speech recognition, while limited exploration has been attempted for other speech processing tasks. As speech signal contains multi-faceted ...information including speaker identity, paralinguistics, spoken content, etc., learning universal representations for all speech tasks is challenging. To tackle the problem, we propose a new pre-trained model, WavLM, to solve full-stack downstream speech tasks. WavLM jointly learns masked speech prediction and denoising in pre-training. By this means, WavLM does not only keep the speech content modeling capability by the masked speech prediction, but also improves the potential to non-ASR tasks by the speech denoising. In addition, WavLM employs gated relative position bias for the Transformer structure to better capture the sequence ordering of input speech. We also scale up the training dataset from 60 k hours to 94 k hours. WavLM Large achieves state-of-the-art performance on the SUPERB benchmark, and brings significant improvements for various speech processing tasks on their representative benchmarks.
The speech representations learned from large-scale unlabeled data have shown better generalizability than those from supervised learning and thus attract a lot of interest to be applied for various ...downstream tasks. In this paper, we explore the limits of speech representations learned by different self-supervised objectives and datasets for automatic speaker verification (ASV), especially with a well-recognized SOTA ASV model, ECAPA-TDNN 1, as a downstream model. The representations from all hidden layers of the pre-trained model are firstly averaged with learnable weights and then fed into the ECAPA-TDNN as input features. The experimental results on Voxceleb dataset show that the weighted average representation is significantly superior to FBank, a conventional handcrafted feature for ASV. Our best single system achieves 0.537%, 0.569%, and 1.180% equal error rate (EER) on the three official trials of VoxCeleb1, separately. Accordingly, the ensemble system with three pre-trained models can further improve the EER to 0.479%, 0.536% and 1.023%. Among the three evaluation trials, our best system outperforms the winner system 2 of the VoxCeleb Speaker Recognition Challenge 2021 (VoxSRC2021) on the VoxCeleb1-E trial.
Continuous Speech Separation with Conformer Chen, Sanyuan; Wu, Yu; Chen, Zhuo ...
ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP),
2021-June-6
Conference Proceeding
Odprti dostop
Continuous speech separation was recently proposed to deal with the overlapped speech in natural conversations. While it was shown to significantly improve the speech recognition performance for ...multichannel conversation transcription, its effectiveness has yet to be proven for a single-channel recording scenario. This paper examines the use of Conformer architecture in lieu of recurrent neural networks for the separation model. Conformer allows the separation model to efficiently capture both local and global context information, which is helpful for speech separation. Experimental results using the LibriCSS dataset show that the Conformer separation model achieves the state of the art results for both single-channel and multi-channel settings. Results for real meeting recordings are also presented, showing significant performance gains in both word error rate (WER) and speaker-attributed WER.
Silicon-based planar neuroprobes are composed of silicon substrate, conducting layer, and insulation layers of SiO2 or SiN membrane. The insulation layer is very important because it affects many key ...parameters of neuprobes, like impedance, SNR (signal noise ratio), reliability, etc. Monolayer membrane of SiO2 or SiN are not good choices for insulation layer, since defects and residual stress in these layers can induce bad passivation. In this paper a composite insulation structure is studied, with thermal SiO2 as the lower insulation layer and with multilayer membrane composed of PECVD SiO2 and SiN as the upper insulation layer. This structure not only solves the problem of residual stress but also ensures a good probe passivation. So it's a good choice for insulation layer of neuroprobes.
High-performance electrode materials play a crucial role at the interface of implantable neural electrode. To realize bidirectional transduction between neural tissue and neural microelectrodes, the ...electrode material must satisfy the function of stimulating and recording. As the number and density of electrode increase, tiny electrodes with high performance are needed in future bioengineering study. In this study, a method of electrochemically co-deposited poly(3,4-ethylenedioxythiophene)/multi-walled carbon nanotube (PEDOT/MWCNT) onto microelectrode arrays with 8 channels was investigated. After modification, the impedance, charge transfer ability and frequency response characteristic were improved simultaneously. Compared with bare golden electrode, the coated microelectrodes with a surface area of 615μm2 exhibited a particularly high safe charge injection limit of 7.74mC/cm2 and low impedance of 12kΩ at 1kHz. In vivo inferior colliculus implantation of rats revealed that the composite film coated microelectrodes showed higher signal to noise ratio recordings >15dB compared to 6dB SNR of bare gold microelectrodes.
With its strong modeling capacity that comes from a multi-head and multi-layer structure, Transformer is a very powerful model for learning a sequential representation and has been successfully ...applied to speech separation recently. However, multi-channel speech separation sometimes does not necessarily need such a heavy structure for all time frames especially when the cross-talker challenge happens only occasionally. For example, in conversation scenarios, most regions contain only a single active speaker, where the separation task downgrades to a single speaker enhancement problem. It turns out that using a very deep network structure for dealing with signals with a low overlap ratio not only negatively affects the inference efficiency but also hurts the separation performance. To deal with this problem, we propose an early exit mechanism, which enables the Transformer model to handle different cases with adaptive depth. Experimental results indicate that not only does the early exit mechanism accelerate the inference, but it also improves the accuracy.