Abstract

Digital audio synthesizers are frameworks for generating digital audio, and are the backbone for creating synthesized digital music. Some of the existing audio synthesizer engines are quite popular due to the following reasons: economy, flexibility and convenience for the user in synthesizing music/audio, compatibility with commodity hardware and software platforms. One such audio synthesis engine is the ALSA (Advanced Linux Sound Architecture) Modular Synthesizer, which is an emulation of an analogue modular synthesizer. Until recently, audio programming software has been inherently sequential. There have been some attempts to parallelize a few of these engines with mixed results. The goal of parallelization is not only to obtain speedup but also to increase throughput so that more complex synthesizers can be built to enhance quality and/or complexity of the sound generated. By design, audio synthesizers have soft real-time requirements. This can mean that many of the techniques that are normally used to parallelize a program can in certain situations be too expensive to offer any real performance gain. As a consequence, a naïve parallelization technique for an audio synthesizer can in fact be too expensive due to added overheads, and hence is of no benefit. This paper discusses our methodologies and experiences on parallelizing the ALSA Modular Synthesizer on a multicore environment, and elaborates the experimental results highlighting the advantages of parallelization.