Viterbi Score Research Articles

Improved dynamic match phone lattice search for Persian spoken term detection system in online and offline applications

Spoken term detection (STD) refers to discovering all occurrences of a given term in a set of speech utterances. One of the well-known approaches for the STD system is the phone lattice search (PLS) that produces a phone-based lattice of speech utterances. Since the accuracy of a phone recognizer affects the accuracy of the STD system, the PLS approach utilizes the minimum edit distance (MED) measure to compensate the phone recognizer errors. While this measure increases the detection rate, it also raises the false alarm rate. In this paper, we consider the PLS approach as the baseline. Then, we use Viterbi scoring and Jaro-Winkler similarity measure in order to decrease the false alarm rate. Since the proposed approach uses more techniques than the baseline approach, the search speed may decrease. To overcome this problem, we use lattice pruning and indexing techniques such as depth first search algorithm to increase the search speed in online and offline applications, respectively. We report the experimental results for monophone-based and triphone-based STD system. The results indicate that using triphone-based STD system improved the performance about 2% in comparison with monophone-based STD system. Moreover, when we used triphone-based models, the proposed approach including MED measure, Viterbi scores and Jaro-Winkler similarity measure improved the accuracy of the method with only MED measure, about 17%.

Modified Viterbi Scoring for HMM-based Speech Recognition

Modified Viterbi Scoring for HMM-based Speech Recognition

Compact Wake-Up Word Speech Recognition on Embedded Platforms

The wake-up word speech recognition system is a new paradigm in the field of automatic speech recognition (ASR). This new paradigm is not yet widely recognized but useful in many applications such as mobile phones and smart home systems. In this paper we describe the development of a compact wake-up word recognizer for embedded platforms. To keep resource cost low, a variety of simplification techniques are used. Speech feature observations are compressed to lower dimension and the simple distance-based template matching method is used in place of complex Viterbi scoring. We apply double scoring method to achieve a better performance. To cooperate with double scoring method, the support vector machine classifier is used as well. We were able to accomplish a performance improvement with false rejection rate reduced from 6.88% to 5.50% and false acceptance rate reduced from 8.40% to 3.01%.

Fast Learning of Grammar Production Probabilities in Radar Electronic Support

Although stochastic context-free grammars (SCFG) appear promising for the recognition and threat assessment of complex radar emitters in radar electronic support (ES) systems, the computational requirements for learning their production rule probabilities can be onerous. The two most popular methods, the inside-outside (IO) algorithm and the Viterbi score (VS) algorithm, are both iterative. IO maximizes the likelihood of a training data set, whereas VS maximizes the likelihood of its best parse trees. Even though VS is known to have lower overall computational costs in practice, both algorithms can be impractical for complex grammatical models. Several techniques have been previously developed to accelerate learning. In this paper, two fast variants of the traditional IO algorithm, known as graphical expectation-maximization (gEM(IO)) and tree-scanning (TS(IO)), are reviewed, along with a third technique called HOLA. In addition, two novel algorithms are proposed that apply the gEM (gEM(VS)) and TS (TS(VS)) principles to the Viterbi technique. An experimental protocol is defined and implemented so that the performance of all five techniques (gEM(IO), TS(IO), gEM(VS), TS(VS), and HOLA) can be compared using simulated training sets of complex radar signals. These techniques are compared from several perspectives-perplexity (the likelihood of a test data set), error rate on estimated states, time and memory complexity per iteration, and convergence time. Estimation of the average case and worst case execution time and storage requirements allow for the assessment of complexity, while computer simulations, performed using radar data sets, allow for the assessment of the other performance measures. The impact on performance of the number of sequences in the training set is observed. Results indicate that gEM(IO) and TS(IO) provide the same level of accuracy, yet the resources requirements depend on the ambiguity of the grammars. As expected, the gEM(VS) and TS(VS) techniques provide significantly lower convergence times and time complexities in practice than gEM(IO) and TS(IO), for a comparable level of accuracy. All of these algorithms may provide a greater level of accuracy than HOLA, yet their computational complexities may be orders of magnitude higher.

Recognition unit model training based on competing word and word string models

A system pattern-based speech recognition, e.g., a hidden Markov model (HMM) based speech recognizer using Viterbi scoring. The principle of minimum recognition error rate is applied by the present invention using discriminative training. Various issues related to the special structure of HMMs are presented. Parameter update expressions for HMMs are provided.

Automatic speech recognition in a noisy automotive environment

Recently, a great deal of interest has been shown in performing reliable speech recognition in an automotive environment. In order to determine the level of recognition accuracy to be expected in such a hostile environment, a series of experiments was performed. The data base for these experiments was created by having four talkers (two male, two female) speak ten replications of an 11‐digit vocabulary, zero‐nine and oh (in an isolated format), while driving in an automobile. Five different environmental conditions were studied: (1) car off, fan off; (2) car idling, fan on high; (3) car traveling at 30 mph, fan on high; (4) car traveling at 60 mph, fan on high; and (5) car traveling at 60 mph, fan off. Two different recognition algorithms were tested, namely; (1) standard LPS‐based template recognition, using dynamic time warping; and (2) continuous density hidden Markov model recognition, using Viterbi scoring and cepstral analysis. Several different distance measures were also evaluated for each of the recognition sytems. Results from the template‐based algorithms showed speaker‐dependent recognition accuracies from 96%–99% depending on which condition was used for training. A five‐state HMM system yielded recognition accuracies from 93%–98% depending on the training condition.