No description available.

This report describes the design and implementation of a fixed audio equalizer based on a scheme where parts of the signal spectrum are downsampled and treated differently for the purpose of reducing the computational complexity and memory requirements. The primary focus has been on finding a way of taking an equalizer based on a simple minimum-phase FIR filter and transform it to the new type of equalizer. To achieve this, a number of undesireable effects such as aliasing distortion and upsampling imaging had to be considered and dealt with. In order to achieve a good amplitude response of the system, optimization procedures were used. As part of the thesis, a cost-effective implementation of the filter has been made for an FPGA, in order to verify that the scheme is indeed usable for equalizing an audio signal.

Latency in the live audio processing chain has become a concern for audio engineers and system designers because significant delays can be perceived and may affect synchronisation of signals, limit interactivity, degrade sound quality and cause acoustic feedback. In recent years, latency problems have become more severe since audio processing has become digitised, high-resolution ADCs and DACs are used, complex processing is performed, and data communication networks are used for audio signal transmission in conjunction with other traffic types. In many live audio applications, latency thresholds are bounded by human perceptions. The applications such as music ensembles and live monitoring require low delay and predictable latency. Current digital audio systems either have difficulties to achieve or have to trade-off latency with other important audio processing functionalities. This thesis investigated the fundamental causes of the latency in a modern digital audio processing system: group delay, buffering delay, and physical propagation delay and their associated system components. By studying the time-critical path of a general audio system, we focus on three main functional blocks that have the significant impact on overall latency; the high-resolution digital filters in sigma-delta based ADC/DAC, the operating system to process low latency audio streams, and the audio networking to transmit audio with flexibility and convergence. In this work, we formed new theory and methods to reduce latency and accurately predict latency for group delay. We proposed new scheduling algorithms for the operating system that is suitable for low latency audio processing. We designed a new system architecture and new protocols to produce deterministic networking components that can contribute the overall timing assurance and predictability of live audio processing. The results are validated by simulations and experimental tests. Also, this bottom-up approach is aligned with the methodology that could solve the timing problem of general cyber-physical systems that require the integration of communication, software and human interactions.

One particular application domain of interest for sensor networks is in the real-time processing of audio information for ecological research questions such as species identification. Real-time audio processing generally involves sophisticated signal processing algorithms and requires substantial computational power. As FPGAs increase in capacity and speed but decrease in cost and power consumption, they are now able to provide low-cost, high performance, energy efficient, flexible, and convenient implementations for a wide range of digital systems. This thesis uses the computational power and single-chip solution capabilities of FPGAs to implement a typical audio processing application for sensor networks onto an FPGA using software / hardware co-design approach, and then evaluate the usefulness of this approach. Some background on sensor networks, audio recognition, FPGAs, MicroBlaze and hardware / software co-design is firstly introduced. A few widely adopted feature extraction and pattern matching algorithms are also presented and compared. Several digital signal processing applications based on FPGAs are then reviewed and analyzed. Software / hardware co-design method is then employed to implement an example system. A bird call recognition system based on linear predictive cepstral coefficients and dynamic time warping algorithm is developed and verified on a PC. Then, a software-only solution for this bird call recognition system is implemented on an FPGA with embedded MicroBlaze processor in a Xilinx development board. By means of code profiling, the performance bottlenecks of the software-only solution are identified. Taking the profiling results and the complexity of the recognition algorithm into account, the dynamic time warping algorithm was mapped into custom FPGA hardware. Fast Simplex Links, which are intended specially for high-speed uni-directional transfers to and from the processor, were used to attach the custom hardware to MicroBlaze and pre-defined driver functions supplied by EDK enabled the communications between software and the custom hardware. The software-hardware implementation was then built after substituting custom hardware for software counterparts. The influence of memory assignments for performance is also investigated. External memory access is identified as a major bottleneck. By moving all code from external DRAM into internal BRAM, the system performance is increased by a factor of about 10. From the analysis and comparison of execution time, logic area, and energy consumption of various implementations, it is shown that the software-hardware implementation can speed up a software-only FPGA implementation up to 528 times, and achieves of the order of 20 times “time-area efficiency” and 40 times energy efficiency. Compared with the PC-based C implementation running with a 40 times faster clock rate, the improved software-hardware system runs only about 7 times slower and its performance can meet the real-time requirement to complete a recognition in under one second. In addition, the software / hardware co-design also significantly reduces the energy consumption associated with individual computations.

In the present day, the music becomes more popular due to the following three reasons: (1) the evolution of the MP3 compression technology, (2) the growth of the public platform, and (3) the development of the MP3 portable discs. Most people follow the music to hum or follow the rhythm to tap sometimes. The meanings of a music style may be various if it is explained or felt by different people. Therefore we cannot obtain a very explicit answer if the notation of the music cannot be exactly made. We need some techniques and methods to analyze the music, and obtain some of its embedded information. Tempo and beats are very important elements in the perceptual music. Therefore, tempo estimation and beat tracking are fundamental techniques in automatic audio processing, which are crucial to multimedia applications. In this thesis, we first develop an artificial neural network to classify the music excerpts into the evaluation preference. And then, with the preference classification, we can obtain accurate estimation for tempo and beats, by either Ellis's method or Dixon's method. We test our method with a mixed data set which contains ten music genres extracted from the "ballroom dancer" database. Our experimental results show that the accuracy of our method is higher than that of only one individual Ellis's method or Dixon's method.

Extraction of singing voice from music is one of the ongoing research topics in the field of speech recognition and audio analysis. In particular, this topic finds many applications in the music field, such as in determining music structure, lyrics recognition, and singer recognition. Although many studies have been conducted for the separation of voice from the background, there has been less study on singing voice in particular. In this study, efforts were made to design a new methodology to improve the separation of vocal and non-vocal components in audio clips using REPET [14]. In the newly designed method, we tried to rectify the issues encountered in the REPET method, while designing an improved repeating mask which is used to extract the non-vocal component in audio. The main reason why the REPET method was preferred over previous methods for this study is its independent nature. More specifically, the majority of existing methods for the separation of singing voice from music were constructed explicitly based on one or more assumptions.

The advances in music technology and cinematography in recent years has granted a higher level of importance to the film music. There exists a gap in the academic study of film music as it relates to music technology, as no appropriate methodology exists that can accurately measure the contribution that music technology makes towards the music as it exists in film. This study aims to contribute towards existing methodologies for analysing film music, but from the perspective of music technology, and more specifically how audio processing in the domains of dynamic, spectral, spatial and temporal processing contribute towards the music in the film. This is achieved by building on the proposed methodologies of the study of film music as proposed by Kassabian (2009) and Altman (2000). This new method can be utilized to create a reference list of contributions that audio processing can make towards the soundtrack of a film by isolating the particular contribution that every moment of music is contributing to the film, and then finding how audio processing adds to this. / Mini Dissertation (MMus)--University of Pretoria, 2017. / Music / MMus / Unrestricted

We begin with an overview of some signal processing terms and topics relevant to music analysis including facts about human sound perception. We then discuss common objectives of music analysis and existing methods for accomplishing them. We conclude with an introduction to a new method of automatically transcribing a piece of music from a digital audio signal.

Existem inúmeras técnicas de síntese de áudio utilizadas atualmente em instrumentos musicais profissionais, dentre as quais as mais fundamentais são a síntese aditiva e a síntese subtrativa. A síntese subtrativa se tornou popular e foi muito explorada entre as décadas de 60 e 70 com a criação de módulos analógicos de hardware que podiam ser interconectados, criando o conceito de sintetizador analógico modular. Apesar do uso deste tipo de sintetizador ter diminuído durante as décadas subsequentes, nos últimos anos sua utilização voltou a crescer e diversos modelos deste tipo de instrumento são vendidos atualmente, porém em geral a preços elevados. Sintetizadores digitais também disponibilizam a técnica de síntese subtrativa utilizando componentes eletrônicos customizados e desenvolvidos pelos fabricantes de sintetizadores com o intuito de utilizar avançadas técnicas de processamento de sinais, o que ainda mantém seus preços elevados. Neste trabalho investigamos a hipótese de que é possível desenvolver um instrumento musical funcional e de qualidade com recursos limitados de processamento, e exploramos essa hipótese implementando síntese subtrativa em uma plataforma acessível e de baixo custo. O desenvolvimento é baseado em linguagem orientada a objetos para criação de módulos de software replicando as características dos módulos encontrados em sintetizadores analógicos modulares. Com esta abordagem, obtemos um software modular que pode ser facilmente modificado baseado nas preferências do programador. A implementação foi testada na plataforma Arduino Due, que é uma plataforma de baixo custo e contém um processador 32-bits ARM 84 MHz. Foi possível adicionar osciladores com algoritmo anti-aliasing, filtros, geradores de envelope, módulo de efeito, uma interface MIDI e um teclado externo, obtendo assim um sintetizador subtrativo completo. Além disto, incluímos no desenvolvimento a implementação de um órgão baseado em síntese aditiva, com polifonia completa e inspirado na arquitetura de órgãos clássicos, mostrando a possibilidade de possuir dois importantes e poderosos métodos de síntese em uma plataforma acessível e de baixo custo. Com esta implementação aberta e pública, buscamos contribuir com o movimento maker e faça-você-mesmo, incentivando novos desenvolvimentos nesta área, em especial na computação e engenharia, aumentando o uso e acesso a instrumentos musicais eletrônicos e a criatividade musical. / Subtractive and additive synthesis are two powerful sound synthesis techniques that caused a revolution when the first electronic and electro mechanic music instruments started to appear some decades ago. Subtractive synthesis became very popular during the 60s and 70s after the creation of analog hardware modules that could be interconnected, creating the concept of the modular synthesizers. After the initial impact, for some years these instruments faced a slow-down in its usage, a tendency that was reverted on the past decade. Nevertheless, the prices of these instruments are often high. Digital synthesizers also offer the subtractive synthesis technique, by using customized electronic components designed and developed by the synthesizers vendors in order to use the most up-to-date technologies and signal processing techniques, which also leads to high prices. In this project, we investigate the hypothesis that it is possible to design and develop a good quality music instrument with low budget electronic components and limited processing capabilities, by implementing this on a low budget and easy to use platform. The development is based on object oriented design, creating software modules that replicates the functionalities of analog synthesizer hardware modules. With this approach, we have a modular software that can be easily changed based on programmers’ preferences. The implementation was tested on the Arduino Due board, which is a cheap, easy to use and widely available platform and powered by a 32-bits ARM 84Mhz processor. We were able to add oscillators with anti-aliasing algorithms, filters, envelope generators, delay effects, a MIDI interface and a keybed, making a complete synthesizer. In addition to this, we included an additive synthesis organ design with full polyphony based on classic organs design, demonstrating the possibility of having two powerful synthesis methods on a cheap and widely available platform. With this design, suitable for low cost platforms, we intend to contribute to the maker movement and encourage new implementations in this area, especially in the computing and engineering fields, increasing the usage and access to (electronic) musical instruments and musical creativity.

This thesis discusses the future of smart business applications on mobile phones and the integration of voice interface across several business applications. It proposes a framework that provides speech processing support for business applications on mobile phones. The framework uses Gaussian Mixture Models (GMM) for low-enrollment speaker recognition and limited vocabulary speech recognition. Algorithms are presented for pre-processing of audio signals into different categories and for start and end point detection. A method is proposed for speech processing that uses Mel Frequency Cepstral Coeffcients (MFCC) as primary feature for extraction. In addition, optimization schemes are developed to improve performance, and overcome constraints of a mobile phone. Experimental results are presented for some prototype applications that evaluate the performance of computationally expensive algorithms on constrained hardware. The thesis concludes by discussing the scope for improvement for the work done in this thesis and future directions in which this work could possibly be extended.