Monaural source separation is useful for many real-world applications though it is a challenging problem. In this paper, we study deep learning for monaural speech separation. We propose the joint optimization of the deep learning models (deep neural networks and recurrent neural networks) with an extra masking layer, which enforces a reconstruction constraint. Moreover, we explore a discriminative training criterion for the neural networks to further enhance the separation performance. We evaluate our approaches using the TIMIT speech corpus for a monaural speech separation task. Our proposed models achieve about 3.8⇠4.9 dB SIR gain compared to NMF models, while maintaining better SDRs and SARs.

No abstract

MXenes, an emerging class of two-dimensional (2D) transition metal carbides and nitrides, have potential for application as high-performance, low-cost electrodes in organic field-effect transistors (OFETs) because of their water dispersibility, high conductivity, and work-function tunability. In this study, we successfully fabricated a large-scale, uniform Ti3C2T x MXene electrode array on a flexible plastic substrate for application in high-performance OFETs. The work function of the Ti3C2T x MXene electrodes was also effectively modulated via chemical doping with NH3. The fabricated OFETs with Ti3C2T x MXene electrodes exhibited excellent device performance, such as a maximum carrier mobility of ∼1 cm2·V–1·s–1 and an on–off current ratio of ∼107 for both p-type and n-type OFETs, even though all the electrode and dielectric layers were fabricated on the plastic substrate by solution processing. Furthermore, MXene-electrode-based complementary logic circuits, such as NOT, NAND, and NOR, were fabricated via integration of p-type and n-type OFETs. The proposed approach is expected to expand the application range of MXenes to other OFET-based electronic devices, such as organic light-emitting displays and electronic skins.

Performance enhancement of triboelectric nanogenerators (TENGs) has been largely limited by the relatively low output current density. Thus, extensive research efforts have been made to increase the output current density. In this respect, this work presents a method to effectively increase output current density of TENGs by adopting polarized ferroelectric polymers and MoS 2 composite. Specifically, by compositing bulk MoS 2 flakes with both Nylon-11 and PVDF-TrFE, respectively, charge density of each triboelectric charging surface was significantly increased. In addition, proper polarization of both ferroelectric composite layers has also led to an additional increase in the charge density. A combination of them synergistically increases the surface charge density, generating huge output current and the power output density. By optimizing the fabrication process, the output voltage and current density up to ∼145 V and ∼350 μA/cm 2 are achieved, respectively. Consequently, the TENG exhibits a recordable output power density of ∼50 mW/cm 2 , which is one of the highest output power densities reported to date. The method introduced in this work can greatly increase the output current density of TENGs, facilitating the development of high-performance triboelectric energy harvesting devices.

Ultrafine particulate matter (PM) in indoor air has become a serious concern for public health. Therefore, there is a growing interest in filters that can be installed on the window frames of ordinary homes to improve the indoor air quality by natural passive ventilation without using expensive forced air circulation systems. Thus, these filters require a high filtering efficiency and high air permeability and visibility, which do not compromise the original functionality of the windows. The filters developed for this purpose to date have demonstrated a high filtering efficiency for PM2.5 but a relatively low efficiency for PM1.0. Here, the performance of the ultrathin poly[(vinylidenefluoride‐co‐trifluoroethylene) (PVDF‐TrFE) nanofiber air filter capable of high‐efficiency PM1.0 filtration is reported. To enhance the PM1.0 filtering efficiency, the filter is electrically activated by the polarization of dipoles and triboelectrification using the ferroelectric nature and triboelectrically negative property of the PVDF‐TrFE filter layer. The electrically activated PVDF‐TrFE filter demonstrates a PM1.0 filtering efficiency of over ≈88% after polarization, which is further improved to ≈94% after triboelectrification. In addition, the filter is ultrathin and air‐permeable with 65% light transmittance. The methods introduced in this work can be adopted to develop high performance, highly visible, and air‐permeable filter media.

We present DeepPicar, a low-cost deep neural network based autonomous car platform. DeepPicar is a small scale replication of a real self-driving car called DAVE-2 by NVIDIA. DAVE-2 uses a deep convolutional neural network (CNN), which takes images from a front-facing camera as input and produces car steering angles as output. DeepPicar uses the same network architecture-9 layers, 27 million connections and 250K parameters-and can drive itself in real-time using a web camera and a Raspberry Pi 3 quad-core platform. Using DeepPicar, we analyze the Pi 3's computing capabilities to support end-to-end deep learning based real-time control of autonomous vehicles. We also systematically compare other contemporary embedded computing platforms using the DeepPicar's CNN-based real-time control workload.We find that all tested platforms, including the Pi 3, are capable of supporting the CNN-based real-time control, from 20 Hz up to 100 Hz, depending on hardware platform. However, we find that shared resource contention remains an important issue that must be considered in applying CNN models on shared memory based embedded computing platforms; we observe up to 11.6X execution time increase in the CNN based control loop due to shared resource contention. To protect the CNN workload, we also evaluate state-of-the-art cache partitioning and memory bandwidth throttling techniques on the Pi 3. We find that cache partitioning is ineffective, while memory bandwidth throttling is an effective solution.

No abstract

Monaural source separation is important for many real world applications. It is challenging because, with only a single channel of information available, without any constraints, an infinite number of solutions are possible. In this paper, we explore joint optimization of masking functions and deep recurrent neural networks for monaural source separation tasks, including monaural speech separation, monaural singing voice separation, and speech denoising. The joint optimization of the deep recurrent neural networks with an extra masking layer enforces a reconstruction constraint. Moreover, we explore a discriminative criterion for training neural networks to further enhance the separation performance. We evaluate the proposed system on the TSP, MIR-1K, and TIMIT datasets for speech separation, singing voice separation, and speech denoising tasks, respectively. Our approaches achieve 2.30-4.98 dB SDR gain compared to NMF models in the speech separation task, 2.30-2.48 dB GNSDR gain and 4.32-5.42 dB GSIR gain compared to existing models in the singing voice separation task, and outperform NMF and DNN baselines in the speech denoising task.