A Decoder for Lvcsr Based on Fixed-Point Arithmetic

Abstract

The increasing computational power of embedded CPU's motivates the fixed-point implementation of highly accurate largevocabulary continuous-speech (LVCSR) algorithms, to achieve the same performance on the device as on the server. We report on methods for the fixed-point implementation of the frame-synchronous beam-search Viterbi decoder, N-grams language models, and HMM likelihood computation. This fixedpoint recognizer is as accurate as our best floating-point recognizer in several LVCSR experiments on the DARPA Switchboard task and on an AT&T proprietary task, with different types of acoustic front-ends and HMM's. We also present experiments on the DARPA Resource Management task using the StrongARM-1100 206 MHz CPU, where the fixed-point implementation enables real-time performance: the floatingpoint recognizer, with floating-point software emulation, is 50 times slower for the same accuracy.

GPU acceleration of automated speech recognition for mobile devices

“…This strict implementation of the mixture of Gaussian models is complex as it involves highly nonlinear operations. However, it has been shown that by taking the maximum of all of the mixtures instead of the sum, a very good approximation of the result could be obtained [10]. Hence, by taking the negative of the logarithm of equation (1) in order to convert probabilities into costs and applying the previous approximation, the acoustic cost can be evaluated by:…”

FPGA Implementation of a Pipelined Gaussian Calculation for HMM-Based Large Vocabulary Speech Recognition

“…This strict implementation of the mixture of Gaussian models is complex as it involves highly nonlinear operations. However, it has been proven that taking the maximum of all of the mixtures, instead of the sum, is a very good approximation of the result [17]. Hence, by taking the negative of the logarithm of (1) in order to convert probabilities into costs and applying the previous approximation, the acoustic cost can be evaluated by where α is a coefficient per mixture that encompasses all constants and parameters outside of the exponential function in (1).…”