Optimization of computationally intensive part of speech recognizer

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F09%3A00164813" target="_blank" >RIV/68407700:21230/09:00164813 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Optimization of computationally intensive part of speech recognizer

Popis výsledku v původním jazyce

A part of computer speech recognizer is a computationally very intensive probability function, which compares parametrized input data with the vast amount of trained parameters and evaluate their similarity. This contribution deals with optimization of the probability function to be better feasible on a DSP hardware architectures. The calculation of the original function is composed of several elementary operations (difference, multiplication, square, accumulation), but it is advantageous to convert such operations to an operation of single type, MAC (multiply and accumulate), which is effectively feasible, especially on Digital Signal Processors. We have found such a transformed function, but at the cost of small reducing the accuracy of the calculations. The result of our experiments is an algorithm which has less elementary operations in its core compared to the original function.

Název v anglickém jazyce

Optimization of computationally intensive part of speech recognizer

Popis výsledku anglicky

A part of computer speech recognizer is a computationally very intensive probability function, which compares parametrized input data with the vast amount of trained parameters and evaluate their similarity. This contribution deals with optimization of the probability function to be better feasible on a DSP hardware architectures. The calculation of the original function is composed of several elementary operations (difference, multiplication, square, accumulation), but it is advantageous to convert such operations to an operation of single type, MAC (multiply and accumulate), which is effectively feasible, especially on Digital Signal Processors. We have found such a transformed function, but at the cost of small reducing the accuracy of the calculations. The result of our experiments is an algorithm which has less elementary operations in its core compared to the original function.

Druh

D - Stať ve sborníku

CEP obor

JA - Elektronika a optoelektronika, elektrotechnika

OECD FORD obor

—

Projekt

Výsledek vznikl pri realizaci vícero projektů. Více informací v záložce Projekty.

Návaznosti

Z - Vyzkumny zamer (s odkazem do CEZ)

Rok uplatnění

2009

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

19th Czech-German Workshop on Speech Processing

ISBN

978-80-86269-18-4

ISSN

—

e-ISSN

—

Počet stran výsledku

5

Strana od-do

—

Název nakladatele

Institute of Photonics and Electronics AS CR

Místo vydání

Prague

Místo konání akce

Prague

Datum konání akce

29. 9. 2009

Typ akce podle státní příslušnosti

EUR - Evropská akce

Kód UT WoS článku

—