Word detection in recorded speech using textual queries

Abstract

Abstract-The paper presents unsupervised method for word detection in recorded spoken language signal. The method is based on examining signal similarity of two analyzed media description: registered voice and a word (textual query) synthesized by using Text-to-Speech tools. The descriptions of media were given by a sequence of Mel-Frequency Cepstral Coefficients or Human-Factor Cepstral Coefficients. Dynamic Time Warping algorithm has been applied to provide time alignment of the given media description. The detection involved classification method based on cost function, calculated upon signal similarity and alignment path. Potential false matches were eliminated in the algorithm by comparing costs of the path subsequences to a threshold value. The results of the work could provide incentives to build affordable commercial or non-commercial solutions for specific and multilingual applications.Index Terms-Speech processing, speech analysis, pattern matching, keyword search, audio information retrieval

Using formant frequencies to word detection in recorded speech

“…This paper considers the same use case as in [1]. In this approach the KWS method supports human operator in searching for specific words in a given speech medium.…”

“…Using textual query and TTS make it easy to extend the approach to reflect language variations assumed in the scenario, to search for the same word translated to several languages [1].…”

“…Optimal means here the lowest cost path P for passing from one point of matrix to another, within given constraints. For details of applying DTW to exemplary speech features vectors, see [1]. …”

“…Regarding this trend [1] considers an approach that could be used to detect words in recorded speech of unknown language without training, by using publicly available, free of charge online translation services with Text-To-Speech support e.g. : Google Translate, Bing Translator, Yandex Translate …”

“…The approach presented in [1] employs cepstrum-based features: MelFrequency Cepstral Coefficients (MFCC) and Human-Factor Cepstral Coefficients (HFCC). As for classification strategy that approach points Dynamic Time Warping (DTW) algorithm.…”