Viterbi Search Algorithm Research Articles

A Low-Power, High-Performance Speech Recognition Accelerator

Automatic Speech Recognition (ASR) is becoming increasingly ubiquitous, especially in the mobile segment. Fast and accurate ASR comes at high energy cost, not being affordable for the tiny power-budgeted mobile devices. Hardware acceleration reduces energy-consumption of ASR systems, while delivering high-performance. In this paper, we present an accelerator for large-vocabulary, speaker-independent, continuous speech-recognition. It focuses on the Viterbi search algorithm representing the main bottleneck in an ASR system. The proposed design consists of innovative techniques to improve the memory subsystem, since memory is the main bottleneck for performance and power in these accelerators’ design. It includes a prefetching scheme tailored to the needs of ASR systems that hides main memory latency for a large fraction of the memory accesses, negligibly impacting area. Additionally, we introduce a novel bandwidth-saving technique that removes off-chip memory accesses by 20 percent. Finally, we present a power saving technique that significantly reduces the leakage power of the accelerators scratchpad memories, providing between 8.5 and 29.2 percent reduction in entire power dissipation. Overall, the proposed design outperforms implementations running on the CPU by orders of magnitude, and achieves speedups between 1.7x and 5.9x for different speech decoders over a highly optimized CUDA implementation running on Geforce-GTX-980 GPU, while reducing the energy by 123-454x.

최대 빈도모델 탐색을 이용한 동물소리 인식용 소리모델생성

본 논문에서는 동물소리 인식시스템을 위하여 최대 빈도모델 탐색 알고리즘을 고안하고 이를 이용한 소리모델을 생성하는 방법을 제안하였다. 소리모델 생성 방법은 동물종의 소리 데이터로부터 학습과정, 비터비 탐색과정 및 최대 빈도모델 탐색과정을 반복하면서 HMM(Hidden Makcov Model)모델의 구조(상태의 수와 GMM의 수)를 탐색하여 최적의 인식률을 갖는 모델집합이 생성하는 방법이다. 최대 빈도모델 탐색 알고리즘은 입력 소리 데이터를 비터비(Viterbi) 알고리즘으로 탐색하여 모델리스트를 생성하고 이 리스트 중에서 최대 빈도수의 모델을 탐색하여 최종 인식결과로 결정하는 방법이다. 알고리즘에서 소리특징으로 MFCC(Mel Frequency Cepstral Coefficient), 모델형식으로 HMM을 이용하고 C# 프로그래밍언어로 구현 하였다. 알고리즘의 성능을 평가하기 위하여 27종의 동물소리를 선정하고 실험을 하였으며 27개의 HMM 모델집합이 97.29 퍼센트의 인식률로 생성됨을 확인하였다. In this paper, I proposed a sound model generation and a most frequent model search algorithm for recognizing animal vocalization. The sound model generation algorithm generates a optimal set of models through repeating processes such as the training process, the Viterbi Search process, and the most frequent model search process while adjusting HMM(Hidden Markov Model) structure to improve global recognition rate. The most frequent model search algorithm searches the list of models produced by Viterbi Search Algorithm for the most frequent model and makes it be the final decision of recognition process. It is implemented using MFCC(Mel Frequency Cepstral Coefficient) for the sound feature, HMM for the model, and C# programming language. To evaluate the algorithm, a set of animal sounds for 27 species were prepared and the experiment showed that the sound model generation algorithm generates 27 HMM models with 97.29 percent of recognition rate.

A fast hierarchical search algorithm for discriminative keyword spotting

A keyword spotter can be considered as a binary classifier which classifies a set of uttered sentences into two groups on the basis of whether they contain target keywords or not. For this classification task, the keyword spotter needs to identify the target keywords locations based on a fast and accurate search algorithm. In our previous works, we exploited a modified Viterbi (M-Viterbi) search algorithm which has two known drawbacks. First, to locate the target keywords, it runs an exhaustive search through all possible segments of input speech. Second, while computing the start and end time-frames of each new phone, it makes the keyword spotter to trace-back and re-evaluate the timing alignments of all previous one(s), despite the fact that very limited amount of data -if any- would get updated as a result. These two pitfalls cause a dramatically enlarged search space as well as a significant increase in computational complexity. In this paper, we propose a Hierarchical Search (H-Search) algorithm which allows the system to ignore some segments of input speech at each level of hierarchy, according to their lower likelihood of containing the target keywords. In addition, unlike the M-Viterbi algorithm, the H-Search algorithm does not demand repeated evaluations when computing the current phone alignment which, in turn, results in a narrowed-down search space (O(TP) versus O(TPLmax) – where T is number of frames, P is number of keyword phones and Lmax is the maximum phone duration) as well as a decreased computational complexity (O(TPLmax) versus O(TPLmax3)) compared to those of the M-Viterbi algorithm. We applied the H-Search algorithm to the classification part of an Evolutionary Discriminative Keyword Spotting (EDKWS) system introduced in our previous works. The experimental results indicate that the H-Search algorithm is executed 100 times faster than the M-Viterbi algorithm while the performance of the EDKWS system degrades no more than two percent compared to that of the M-Viterbi algorithm.

Improving Machine Translation Using Hybrid Dictionary-Graph Based Word Sense Disambiguation with Semantic and Statistical Methods

This paper describes a machine translation prototype in which noun phrase translation is defined as a subtask of machine translation. A dedicated noun phrase translation subsystem is built and improved to translate Arabic noun phrase into English using only minimal resources for both the source and the target language. This work proposes a dictionary-graph based WSD approach to improve machine translation using hybrid semantic-statistical method based on computing words relatedness and a statistical measure of association to get the relation between ambiguous words. This relation was used with viterbi search algorithm to find the appropriate translation of the Arabic noun phrase. A shallow source language analysis, combined with a translation dictionary and a mapping system of source language phenomena into the target language and a target language corpus for generation are all the resources needed in the described system.

Adaptive Optimization of an Iterative Multiuser Detector for Turbo-Coded CDMA

Extrinsic information transfer (EXIT) charts are utilized to optimize the iterative multiuser detector receiver in a multiuser turbo-coded CDMA system. The (receive) power levels are optimized for the system load using a constrained nonlinear optimization approach. The optimal decoding schedule is derived dynamically using the power optimized EXIT chart and a Viterbi search algorithm. Dynamic scheduling is shown to be a more flexible approach which results in a more stable QoS for a typical system configuration than one-shot scheduling, and large complexity savings over a receiver without scheduling. We verify through simulations that complexity savings of over 50% and power savings of over 8dB can be achieved. We show that the optimized power levels combined with adaptive scheduling allows for efficient utilization of receiver resources for heavily loaded systems.

Large-Vocabulary Continuous Sign Language Recognition Based on Transition-Movement Models

The major challenges that sign language recognition (SLR) now faces are developing methods that solve large-vocabulary continuous sign problems. In this paper, transition-movement models (TMMs) are proposed to handle transition parts between two adjacent signs in large-vocabulary continuous SLR. For tackling mass transition movements arisen from a large vocabulary size, a temporal clustering algorithm improved from k-means by using dynamic time warping as its distance measure is proposed to dynamically cluster them; then, an iterative segmentation algorithm for automatically segmenting transition parts from continuous sentences and training these TMMs through a bootstrap process is presented. The clustered TMMs due to their excellent generalization are very suitable for large-vocabulary continuous SLR. Lastly, TMMs together with sign models are viewed as candidates of the Viterbi search algorithm for recognizing continuous sign language. Experiments demonstrate that continuous SLR based on TMMs has good performance over a large vocabulary of 5113 Chinese signs and obtains an average accuracy of 91.9%

Continuous Mandarin speech recognition for Chinese language with large vocabulary based on segmental probability model

The author presents a study of large-vocabulary continuous Mandarin speech recognition based on a segmental probability model (SPM) approach. SPM was found to be very suitable for recognition of isolated Mandarin syllables especially considering the monosyllabic structure of Chinese language. To extend the application of the model to continuous Mandarin speech recognition, a concatenated syllable matching (CSM) algorithm in place of the conventional Viterbi search algorithm is first introduced. Also, to utilise the available training material efficiently, a training procedure is proposed to re-estimate the SPM parameters using the maximum a posteriori (MAP) algorithm. A few special techniques integrating acoustic and linguistic knowledge are developed further to improve the performance step by step. Preliminary experimental results show that the final achievable rate is as high as 91.62%, which indicates a 18.48% error rate reduction and more than three times faster than the well studied subsyllable-based CHMM.