Automatic Speech Recognition on Mobile Devices and over Communication Networks
Part of the series Advances in Pattern Recognition pp 255275
FixedPoint Arithmetic
 Enrico BocchieriAffiliated withAT&T Labs Research
There are two main requirements for embedded/mobile systems: one is low power consumption for long battery life and miniaturization, the other is low unit cost for components produced in very large numbers (cell phones, settop boxes). Both requirements are addressed by CPU’s with integeronly arithmetic units which motivate the fixedpoint arithmetic implementation of automatic speech recognition (ASR) algorithms. Large vocabulary continuous speech recognition (LVCSR) can greatly enhance the usability of devices, whose small size and typical onthego use hinder more traditional interfaces. The increasing computational power of embedded CPU’s will soon allow realtime LVCSR on portable and lowcost devices. This chapter reviews problems concerning the fixedpoint implementation of ASR algorithms and it presents fixedpoint methods yielding the same recognition accuracy of the floatingpoint algorithms. In particular, the chapter illustrates a practical approach to the implementation of the framesynchronous beamsearch Viterbi decoder, Ngrams language models, HMM likelihood computation and melcepstrum frontend. The fixedpoint recognizer is shown to be as accurate as the floatingpoint recognizer in several LVCSR experiments, on the DARPA Switchboard task, and on an AT&T proprietary task, using different types of acoustic frontends, HMM’s and language models. Experiments on the DARPA Resource Management task, using the StrongARM1100 206 MHz and the XScale PXA270 624 MHz CPU’s show that the fixedpoint implementation enables realtime performance: the floating point recognizer, with floatingpoint software emulation is several times slower for the same accuracy.
 Title
 FixedPoint Arithmetic
 Book Title
 Automatic Speech Recognition on Mobile Devices and over Communication Networks
 Book Part
 III
 Pages
 pp 255275
 Copyright
 2008
 DOI
 10.1007/9781848001435_12
 Print ISBN
 9781848001428
 Online ISBN
 9781848001435
 Series Title
 Advances in Pattern Recognition
 Series ISSN
 16177916
 Publisher
 Springer London
 Copyright Holder
 SpringerVerlag London Limited
 Additional Links
 Topics
 Industry Sectors
 eBook Packages
 Authors

 Enrico Bocchieri ^{(3)}
 Author Affiliations

 3. AT&T Labs Research, Florham Park, New Jersey, USA
Continue reading...
To view the rest of this content please follow the download PDF link above.