A new architecture is presented to support the general class of real-time large-vocabulary speaker-independent continuous speech recognizers incorporating language models. Many such recognizers require multiple high-performance central processing units (CPU's) as well as high interprocessor communication bandwidth. This array processor provides a peak CPU performance of 2.56 giga-floating point operations per second (GFLOPS) as well as a high-speed communication network. In order to efficiently utilize these resources, algorithms were devised for partitioning speech models for mapping into the array processor. Also, a novel scheme is presented for a functional partitioning of the speech recognizer computations. The recognizer is functionally partitioned into six stages, namely, the linear predictive coding (LPC) based feature extractor, mixture probability computer, (phone) state probability computer, word probability computer, phrase probability computer, and traceback computer. Each of these stages is further subdivided as many times as necessary to fit the individual processing elements (PE's). The functional stages are pipelined and synchronized with the frame rate of the incoming speech signal. This partitioning also allows a multistage stack decoder to be implemented for reduction of computation.>

A problem with many speech understanding systems is that grammars that are more suitable for representing the relation between sentences and their meanings, such as context free grammars (CFGs) and augmented phrase structure grammars (APSGs), are computationally very demanding. On the other hand, finite state grammars are efficient, but cannot represent directly the sentence-meaning relation. The authors describe how speech recognition and language analysis can be tightly coupled by developing an APSG for the analysis component and deriving automatically from it a finite-state approximation that is used as the recognition language model. Using this technique, the authors have built an efficient translation system that is fast compared to others with comparably sized language models.>

The authors investigate the parallel computation of large-vocabulary speech recognition on a tree-structured parallel processor. Having seriously established results on parallel level-building with continuously variable hidden Markov models they extend this work to the case in which the spoken sentences are constrained by a finite state grammar. The two key ideas are: (1) a pipelined sorting function on the processor array that efficiently transmits a sorted list of the best scores from all processors to the host; and (2) a level-based pruning technique in which paths through the dynamic programming network are pruned only at the ends of words. These ideas are evaluated on the BT-100 processor, a binary-tree parallel processor. A performance model is presented that estimates execution time as a function of algorithm parameters. Real-time speech recognition has been achieved for a data-entry task with a 70-word vocabulary and average branching factor of 23, and for an airline reservation task with a vocabulary of 132 words. The performance model predicts real-time execution of the 991-word DARPA Resource Management Task on a 127-processor machine.>

The authors describe a parallel frame-synchronous level-building algorithm, utilizing HMM word-models, for connected-speech recognition on a tree-structured parallel computer. The algorithm is scalable in the sense that the source code and execution time remain essentially the same as vocabulary size increases, so long as the hardware is scaled proportionally. An illustrative sizing and timing analysis of a speaker-independent connected-digit recognizer on the ASPEN tree-machine is described. This algorithm executes in real-time and achieves 98.3% string accuracy on the Texas Instruments digit data base.>

Speech recognizers trained in quiet conditions but operating in noise usually have poor accuracy. This paper reports two methods for improving the accuracy of an LPC vector-quantization speech recognizer by adapting the vector codebook to noisy conditions. First, each codebook vector is changed to reflect the way people speak in noise. Second, the estimated spectrum of the background noise is added to the codebook vectors. These ideas have been tested on a total of 2400 utterances of digits recorded in a car by 4 speakers. A baseline word spotter similar to NTT's SPLIT system was modified by adapting its vector codebook to noise. This adapted codebook, when used with a new word decision criterion, yields error rates at least 4 times lower for noisy conditions. The accuracy is significantly better than without codebook adaptation techniques.