Алгоритмы параллельной обработки речевых сигналов
Аннотация
В данной работе рассматриваются реализации параллельных алгоритмов применительно к процедурам акустического этапа обработки речевых сигналов. Проблемы потоковой обработки в задачах анализа речевых сигналов на многоядерных процессорах рассматриваются как новая технология, включающая процедуры очистки от шума, выделение полезных компонент, сегментации и последующего спектрального анализа в выбранной базисной системе.
Маколада нутк сигналларини акустик канта ишлаш жараёнларида параллел алгоритмларининг тадбини куриб чикилади. Шунингдек, купядроли процессорларда параллел (окимли) кайта ишлаш жараёнларини янги технологя сифатида нутк сигналларини тахлил килишда халакитлардан тозалаш, керакли кисмларини ажратиб олиш, сигментлаш ва танланган базис тизимида спектрал тахлил килишни тадбик килишдан иборат.
In this article researches the implementation of parallel algorithms in relation to the procedures of the acoustic stage of processing speech signals. The problems of stream processing in problems of analysis of speech signals on multicore processors are considered as a new technology, including noise removal procedures, allocation of useful components, segmentation and subsequent spectral analysis in the selected base system. In real systems for recording, processing and transmitting speech, increasing the processing speed may be required at the first, acoustic stage of processing. It is at the first stage, before the translation of the speech signal into the parametric representation, that the amount of information is maximal, because this is a phonogram recording from a microphone. At the final processing methods are difficult to parallelize and do not give the expected gain in processing speed. Thus, the greatest effect from the acceleration of computations in real-time systems realizing the tasks of digital processing of speech signals is achieved at the first two stages indicated above. From the point of view of the developer of programs for real-time signal processing systems, the transition to the use of multi-core architectures has a number of fundamental differences from the development of traditional single-core applications, including the distribution of various fragments of code in the kernels, data separation, kernel synchronization, data exchange and overhead between the cores , synchronization cache and others. At the acoustic stage of signal processing, the standard procedures for filtering and noise suppression, amplification, separation of information streams, extraction of useful information for further use are applied in most systems. In general, this is the preliminary processing of speech signals.
Маколада нутк сигналларини акустик канта ишлаш жараёнларида параллел алгоритмларининг тадбини куриб чикилади. Шунингдек, купядроли процессорларда параллел (окимли) кайта ишлаш жараёнларини янги технологя сифатида нутк сигналларини тахлил килишда халакитлардан тозалаш, керакли кисмларини ажратиб олиш, сигментлаш ва танланган базис тизимида спектрал тахлил килишни тадбик килишдан иборат.
In this article researches the implementation of parallel algorithms in relation to the procedures of the acoustic stage of processing speech signals. The problems of stream processing in problems of analysis of speech signals on multicore processors are considered as a new technology, including noise removal procedures, allocation of useful components, segmentation and subsequent spectral analysis in the selected base system. In real systems for recording, processing and transmitting speech, increasing the processing speed may be required at the first, acoustic stage of processing. It is at the first stage, before the translation of the speech signal into the parametric representation, that the amount of information is maximal, because this is a phonogram recording from a microphone. At the final processing methods are difficult to parallelize and do not give the expected gain in processing speed. Thus, the greatest effect from the acceleration of computations in real-time systems realizing the tasks of digital processing of speech signals is achieved at the first two stages indicated above. From the point of view of the developer of programs for real-time signal processing systems, the transition to the use of multi-core architectures has a number of fundamental differences from the development of traditional single-core applications, including the distribution of various fragments of code in the kernels, data separation, kernel synchronization, data exchange and overhead between the cores , synchronization cache and others. At the acoustic stage of signal processing, the standard procedures for filtering and noise suppression, amplification, separation of information streams, extraction of useful information for further use are applied in most systems. In general, this is the preliminary processing of speech signals.