Robust speech recognition using speech enhancement

Automatic Speech Recognition (ASR) has matured into a technology which is becoming more common in our everyday lives, and is emerging as a necessity to minimise driver distraction when operating in-car systems such as navigation and infotainment. In “noise-free” environments, word recognition performance of these systems has been shown to approach 100%, however this performance degrades rapidly as the level of background noise is increased. Speech enhancement is a popular method for making ASR systems more ro- bust. Single-channel spectral subtraction was originally designed to improve hu- man speech intelligibility and many attempts have been made to optimise this algorithm in terms of signal-based metrics such as maximised Signal-to-Noise Ratio (SNR) or minimised speech distortion. Such metrics are used to assess en- hancement performance for intelligibility not speech recognition, therefore mak- ing them sub-optimal ASR applications. This research investigates two methods for closely coupling subtractive-type enhancement algorithms with ASR: (a) a computationally-efficient Mel-filterbank noise subtraction technique based on likelihood-maximisation (LIMA), and (b) in- troducing phase spectrum information to enable spectral subtraction in the com- plex frequency domain. Likelihood-maximisation uses gradient-descent to optimise parameters of the enhancement algorithm to best fit the acoustic speech model given a word se- quence known a priori. Whilst this technique is shown to improve the ASR word accuracy performance, it is also identified to be particularly sensitive to non-noise mismatches between the training and testing data. Phase information has long been ignored in spectral subtraction as it is deemed to have little effect on human intelligibility. In this work it is shown that phase information is important in obtaining highly accurate estimates of clean speech magnitudes which are typically used in ASR feature extraction. Phase Estimation via Delay Projection is proposed based on the stationarity of sinusoidal signals, and demonstrates the potential to produce improvements in ASR word accuracy in a wide range of SNR. Throughout the dissertation, consideration is given to practical implemen- tation in vehicular environments which resulted in two novel contributions – a LIMA framework which takes advantage of the grounding procedure common to speech dialogue systems, and a resource-saving formulation of frequency-domain spectral subtraction for realisation in field-programmable gate array hardware. The techniques proposed in this dissertation were evaluated using the Aus- tralian English In-Car Speech Corpus which was collected as part of this work. This database is the first of its kind within Australia and captures real in-car speech of 50 native Australian speakers in seven driving conditions common to Australian environments.

An evolutionary computation approach to three-dimensional path planning for unmanned aerial vehicles with tactical and kinematic constraints

This paper presents a novel evolutionary computation approach to three-dimensional path planning for unmanned aerial vehicles (UAVs) with tactical and kinematic constraints. A genetic algorithm (GA) is modified and extended for path planning. Two GAs are seeded at the initial and final positions with a common objective to minimise their distance apart under given UAV constraints. This is accomplished by the synchronous optimisation of subsequent control vectors. The proposed evolutionary computation approach is called synchronous genetic algorithm (SGA). The sequence of control vectors generated by the SGA constitutes to a near-optimal path plan. The resulting path plan exhibits no discontinuity when transitioning from curve to straight trajectories. Experiments and results show that the paths generated by the SGA are within 2% of the optimal solution. Such a path planner when implemented on a hardware accelerator, such as field programmable gate array chips, can be used in the UAV as on-board replanner, as well as in ground station systems for assisting in high precision planning and modelling of mission scenarios.

Design methodologies and architectures of hardware-based evolutionary algorithms for aerospace optimisation applications on FPGAS

This thesis is a study of new design methods for allowing evolutionary algorithms to be more effectively utilised in aerospace optimisation applications where computation needs are high and computation platform space may be restrictive. It examines the applicability of special hardware computational platforms known as field programmable gate arrays and shows that with the right implementation methods they can offer significant benefits. This research is a step forward towards the advancement of efficient and highly automated aircraft systems for meeting compact physical constraints in aerospace platforms and providing effective performance speedups over traditional methods.

Computational experiments involving population size for FPGA-based implementation of a GA for the TSP

The feasibility of using an in-hardware implementation of a genetic algorithm (GA) to solve the computationally expensive travelling salesman problem (TSP) is explored, especially in regard to hardware resource requirements for problem and population sizes. We investigate via numerical experiments whether a small population size might prove sufficient to obtain reasonable quality solutions for the TSP, thereby permitting relatively resource efficient hardware implementation on field programmable gate arrays (FPGAs). Software experiments on two TSP benchmarks involving 48 and 532 cities were used to explore the extent to which population size can be reduced without compromising solution quality, and results show that a GA allowed to run for a large number of generations with a smaller population size can yield solutions of comparable quality to those obtained using a larger population. This finding is then used to investigate feasible problem sizes on a targeted Virtex-7 vx485T-2 FPGA platform via exploration of hardware resource requirements for memory and data flow operations.

Comparison of high level FPGA hardware design for solving tri-diagonal linear systems

Reconfigurable computing devices can increase the performance of compute intensive algorithms by implementing application specific co-processor architectures. The power cost for this performance gain is often an order of magnitude less than that of modern CPUs and GPUs. Exploiting the potential of reconfigurable devices such as Field-Programmable Gate Arrays (FPGAs) is typically a complex and tedious hardware engineering task. Re- cently the major FPGA vendors (Altera, and Xilinx) have released their own high-level design tools, which have great potential for rapid development of FPGA based custom accelerators. In this paper, we will evaluate Altera’s OpenCL Software Development Kit, and Xilinx’s Vivado High Level Sythesis tool. These tools will be compared for their per- formance, logic utilisation, and ease of development for the test case of a Tri-diagonal linear system solver.

Calculation of volume from crank angle using reconfigurable hardware

Fast calculation of quantities such as in-cylinder volume and indicated power is important in internal combustion engine research. Multiple channels of data including crank angle and pressure were collected for this purpose using a fully instrumented diesel engine research facility. Currently, existing methods use software to post-process the data, first calculating volume from crank angle, then calculating the indicated work and indicated power from the area enclosed by the pressure-volume indicator diagram. Instead, this work investigates the feasibility of achieving real-time calculation of volume and power via hardware implementation on Field Programmable Gate Arrays (FPGAs). Alternative hardware implementations were investigated using lookup tables, Taylor series methods or the CORDIC (CoOrdinate Rotation DIgital Computer) algorithm to compute the trigonometric operations in the crank angle to volume calculation, and the CORDIC algorithm was found to use the least amount of resources. Simulation of the hardware based implementation showed that the error in the volume and indicated power is less than 0.1%.

Bit-stream-based space vector modulators

Bit-stream-based control, which uses one bit wide signals to control power electronics applications, is a new approach for controller design in power electronic systems. This study presents a novel family of three-phase space vector modulators, which are based on the bit-stream technique and suitable for three-phase inverter systems. Each of the proposed modulators simultaneously converts a two-phase reference to the three-phase domain and reduces switching frequencies to reasonable levels. The modulators do not require carrier oscillators, trigonometric functions or, in some cases, sector detectors. A complete three-phase modulator can be implemented in as few as 102 logic elements. The performance of the proposed modulators is compared through simulation and experimental testing of a 6 kW, three-phase DC-to-AC inverter. Subject to limits on the modulation index, the proposed modulators deliver spread-spectrum output currents with total harmonic distortion comparable to a standard carrier-based space vector pulse width modulator.

Logic-gate devices based on printed polymer semiconducting nanostripes

The applications of organic semiconductors in complex circuitry such as printed CMOS-like logic circuits demand miniaturization of the active structures to the submicrometric and nanoscale level while enhancing or at least preserving the charge transport properties upon processing. Here, we addressed this issue by using a wet lithographic technique, which exploits and enhances the molecular order in polymers by spatial confinement, to fabricate ambipolar organic field effect transistors and inverter circuits based on nanostructured single component ambipolar polymeric semiconductor. In our devices, the current flows through a precisely defined array of nanostripes made of a highly ordered diketopyrrolopyrrole-benzothiadiazole copolymer with high charge carrier mobility (1.45 cm2 V-1 s-1 for electrons and 0.70 cm2 V-1 s-1 for holes). Finally, we demonstrated the functionality of the ambipolar nanostripe transistors by assembling them into an inverter circuit that exhibits a gain (105) comparable to inverters based on single crystal semiconductors.

Solution-processed dual-gate polymer field-effect transistors for display applications

We describe the advantages of dual-gate thin-film transistors (TFTs) for display applications. We show that in TFTs with active semiconductor layers composed of diketopyrrolopyrrole-naphthalene copolymer, the on-current is increased, the off-current is reduced, and the sub-threshold swing is improved compared to single-gate devices. Charge transport measurements in steady-state and under non-quasi-static conditions reveal the reasons for this improved performance. We show that in dual-gate devices, a much smaller fraction of charge carriers move in slow trap states. We also compare the activation energies for charge transport in the top-gate and bottom-gate configurations.

Microphone array beamforming approach to blind speech separation

In this paper, we present a microphone array beamforming approach to blind speech separation. Unlike previous beamforming approaches, our system does not require a-priori knowledge of the microphone placement and speaker location, making the system directly comparable other blind source separation methods which require no prior knowledge of recording conditions. Microphone location is automatically estimated using an assumed noise field model, and speaker locations are estimated using cross correlation based methods. The system is evaluated on the data provided for the PASCAL Speech Separation Challenge 2 (SSC2), achieving a word error rate of 58% on the evaluation set.

Frame-rate stereopsis using non-parametric transforms and programmable logic

A frame-rate stereo vision system, based on non-parametric matching metrics, is described. Traditional metrics, such as normalized cross-correlation, are expensive in terms of logic. Non-parametric measures require only simple, parallelizable, functions such as comparators, counters and exclusive-or, and are thus very well suited to implementation in reprogrammable logic.