Sample records for vaino helje kaarma
-
Li, Kan; Príncipe, José C.
2018-01-01
This paper presents a novel real-time dynamic framework for quantifying time-series structure in spoken words using spikes. Audio signals are converted into multi-channel spike trains using a biologically-inspired leaky integrate-and-fire (LIF) spike generator. These spike trains are mapped into a function space of infinite dimension, i.e., a Reproducing Kernel Hilbert Space (RKHS) using point-process kernels, where a state-space model learns the dynamics of the multidimensional spike input using gradient descent learning. This kernelized recurrent system is very parsimonious and achieves the necessary memory depth via feedback of its internal states when trained discriminatively, utilizing the full context of the phoneme sequence. A main advantage of modeling nonlinear dynamics using state-space trajectories in the RKHS is that it imposes no restriction on the relationship between the exogenous input and its internal state. We are free to choose the input representation with an appropriate kernel, and changing the kernel does not impact the system nor the learning algorithm. Moreover, we show that this novel framework can outperform both traditional hidden Markov model (HMM) speech processing as well as neuromorphic implementations based on spiking neural network (SNN), yielding accurate and ultra-low power word spotters. As a proof of concept, we demonstrate its capabilities using the benchmark TI-46 digit corpus for isolated-word automatic speech recognition (ASR) or keyword spotting. Compared to HMM using Mel-frequency cepstral coefficient (MFCC) front-end without time-derivatives, our MFCC-KAARMA offered improved performance. For spike-train front-end, spike-KAARMA also outperformed state-of-the-art SNN solutions. Furthermore, compared to MFCCs, spike trains provided enhanced noise robustness in certain low signal-to-noise ratio (SNR) regime. PMID:29666568
-
Li, Kan; Príncipe, José C
2018-01-01
This paper presents a novel real-time dynamic framework for quantifying time-series structure in spoken words using spikes. Audio signals are converted into multi-channel spike trains using a biologically-inspired leaky integrate-and-fire (LIF) spike generator. These spike trains are mapped into a function space of infinite dimension, i.e., a Reproducing Kernel Hilbert Space (RKHS) using point-process kernels, where a state-space model learns the dynamics of the multidimensional spike input using gradient descent learning. This kernelized recurrent system is very parsimonious and achieves the necessary memory depth via feedback of its internal states when trained discriminatively, utilizing the full context of the phoneme sequence. A main advantage of modeling nonlinear dynamics using state-space trajectories in the RKHS is that it imposes no restriction on the relationship between the exogenous input and its internal state. We are free to choose the input representation with an appropriate kernel, and changing the kernel does not impact the system nor the learning algorithm. Moreover, we show that this novel framework can outperform both traditional hidden Markov model (HMM) speech processing as well as neuromorphic implementations based on spiking neural network (SNN), yielding accurate and ultra-low power word spotters. As a proof of concept, we demonstrate its capabilities using the benchmark TI-46 digit corpus for isolated-word automatic speech recognition (ASR) or keyword spotting. Compared to HMM using Mel-frequency cepstral coefficient (MFCC) front-end without time-derivatives, our MFCC-KAARMA offered improved performance. For spike-train front-end, spike-KAARMA also outperformed state-of-the-art SNN solutions. Furthermore, compared to MFCCs, spike trains provided enhanced noise robustness in certain low signal-to-noise ratio (SNR) regime.
-
NASA Astrophysics Data System (ADS)
Catalano, M.
2009-05-01
Geonics Limited has a very rich and varied history. This talk will provide a historical perspective about how a few key individuals shaped the development of some of the world's most useful electromagnetic (EM) geophysical instrumentation. A brief review of these systems, including the science behind them, will showcase the evolution of each to the market place and emphasize how a combination of business savvy and a constant investment to research is what lead to a successful line of instrumentation. In 1950 a company called Aeromagnetic Surveys Ltd. was established that was considered "the largest and most diversified air- survey firm in the world" (FLIGHT, 1954), for its time. It employed Vaino Ronka and Alex Herz, young engineers, who patented several new EM technologies including an in-phase and quadrature towed bird helicopter EM system (the first commercial transistorized instrument). The two also set new standards for ground based horizontal loop EM systems and won several mining Blue Ribbon Awards. By the end of 1958, Mr. Ronka began offering independent design services for geophysical instruments and it became inevitable that one day he would form his own company. Geonics Limited was incorporated in 1962 by Vaino Ronka and Alex Herz and the EM-16 VLF receiver, first sold in 1965, became the first successful instrument. It's considered the best selling electrical geophysical tool of all-time and is still sold today by the same model name 44 years later. In 1974, the company was purchased by James Duncan McNeill, the former chief engineering physicist of Barringer Research Ltd. During his time as president of Geonics he was responsible for an explosion of new instruments from the 70's, 80's and into the 90's that permanently placed Geonics instruments in virtually every government environmental lab and consulting firm active in near-surface geophysics. His ability to foresee new problem areas and to define new roles that geophysical methods could play in a