Цифровая фильтрация речевых сигналов на многоядерных процессорах
- № 3 (43) 2017
Страницы:
27
–
34
Язык: русский
Аннотация
В данной статье исследуются проблемы выбора оптимального размера кадров (сегмента речевого сигнала) для цифровой фильтрации при реализации параллельной (потоковой) обработки на многоядерных процессорах. Для параллельной обработки выбран фрагмент речевого сигнала и выполнена процедура цифровой фильтрации.
Ушбу мақолада тўрт ядроли процессорда параллел (оқимли) қайта ишлашда, нутқ сигналларини фреймларга бўлиб рақамли фильтирлашда фрейм (нутқ сигнали қисми)нинг оптимал ўлчамини топиш муаммолари ечими изланилди. Параллел қайта ишлаш жараёнларида микрофондан ёзиб олинган товуш сигналининг рақамли фильтирлаш жараёнлари келтирилади. Рақамли фильтирлаш жараёнларини паралеллаштиришда қўлланилган янги усулнинг тузулмаси батафсил ёритилган.
In this article researches the problems of selecting optimal size of frames (of the speech signal segment) for digital filtering in the implementation of the parallel mode (streaming) processing on multi-core processors. For parallel processing has been performed digital filtering for the speech signals. A new structure is illustrat-ed that is parallel to the process of digital filtering of a speech signal. Used FIR fil-ter for parallel processing in speech signals. According to this algorithm, the signal is divided into frames (fragments of the speech signal) in accordance with the number of processor cores and in this order are written to the main RAM. Filtering is performed on the entire length of the speech signal. Variable parameter in this case is the length of the fragment of the speech signal. Variable parameter in the implemented parallel algorithm is the number of fragments, while the filtering the coefficients stay as the same. The algorithm is implemented on a quad-core proces-sor and the same fragment of the speech signal is selected for all implementations
Ушбу мақолада тўрт ядроли процессорда параллел (оқимли) қайта ишлашда, нутқ сигналларини фреймларга бўлиб рақамли фильтирлашда фрейм (нутқ сигнали қисми)нинг оптимал ўлчамини топиш муаммолари ечими изланилди. Параллел қайта ишлаш жараёнларида микрофондан ёзиб олинган товуш сигналининг рақамли фильтирлаш жараёнлари келтирилади. Рақамли фильтирлаш жараёнларини паралеллаштиришда қўлланилган янги усулнинг тузулмаси батафсил ёритилган.
In this article researches the problems of selecting optimal size of frames (of the speech signal segment) for digital filtering in the implementation of the parallel mode (streaming) processing on multi-core processors. For parallel processing has been performed digital filtering for the speech signals. A new structure is illustrat-ed that is parallel to the process of digital filtering of a speech signal. Used FIR fil-ter for parallel processing in speech signals. According to this algorithm, the signal is divided into frames (fragments of the speech signal) in accordance with the number of processor cores and in this order are written to the main RAM. Filtering is performed on the entire length of the speech signal. Variable parameter in this case is the length of the fragment of the speech signal. Variable parameter in the implemented parallel algorithm is the number of fragments, while the filtering the coefficients stay as the same. The algorithm is implemented on a quad-core proces-sor and the same fragment of the speech signal is selected for all implementations