The Dragon Continuous Speech Recognition System: A Real-Time Implementation

The Dragon Continuous Speech Recognition System: A Real-Time Implementation Paul Bamberg, Yen-lu Chow, Laurence Gillick, Robert Roth and Dean Sturtevant Dragon Systems, Inc. 90 Bridge Street Newton, MA 02158 Abstract We present a 1000-word continuous speech recognition (CSR) system that operates in real time on a personal computer (PC). The system, designed for large vocabulary natural language tasks, makes use of phonetic Hidden Markov models (HMM) and incorporates acoustic, phonetic, and linguistic sources of knowledge to achieve high recognition performance. We describe the various components of this system. We also present our strategy for achieving real time recognition on the PC. Using a 486- based PC with a 29K-based add-on board, the recognizer has been timed at 1.1 times real time. 1. Introduction This paper describes the Dragon continuous speech recognition system that runs in real time on the PC with a 1000-word vocabulary. To achieve the goal of real-time recognition on a personal computer is a process that requires analysis of the computational requirements of the recognition algorithm along several dimensions, and the improving the recognizer's performance along those dimensions. We first present an overview of the Dragon CSR system architecture, and describe its various components, including signal processing, recognition, rapid match, phonetic modeling, and the application task. We discuss our strategy for achieving real-time continuous speech recognition, and demonstrate how it is actually achieved by developing multiple solutions and applying them in combination. 2. System Description The architecture of the continuous speech recognition system is shown in Figure 1. The various components of this system are described below. SYSTEM ARCHITECTURE Speech L I s,g.,/ MatDchPer tcaWn;:~atee[ MRa~Pclhder Recognized Sentence Figure 1: Speech recognition system architecture 2.1 Signal Processing An TMS32010-based board that plugs into a slot on the AT-bus performs analog-to-digital conversion and digital signal processing of the input speech waveform, and extracts spectral features used in recognition. Input speech is sampled at 12 KHz and lowpass filtered at 6 KHz. Eight spectral parameters are computed every 20 milliseconds, and are used as input to the HMM-based recognizer. 78 2.2 Recognition The recognition search to find the most likely sentence hypothesis is based on the time-synchronous decoding algorithm that is used in almost all current CSR systems for this vocabulary size. In this algorithm, partial paths (representing incomplete sentential hypotheses) are extended synchronously using dynamic programming (DP), and all span the same length of the input signal, so that their path cost functions are directly comparable. To reduce recognition search, a beam pruning technique is applied to eliminate all paths that score poorly relative to the best path, and therefore would have very low probability of being the global best hypotheses that spans the entire utterance. We also explored another family of speech decoding algorithms, the stack decoder[l], in our recognizer. It is our conclusion at this time that at least for a task of this complexity, time- synchronous algorithms are considerably more efficient for finding a single most likely answer. 2.3 Rapid Matcher An important component of …

Content truncated to 500 words