Spoken Term Research Articles

Learning Speaker-Specific Characteristics With a Deep Neural Architecture

Speech signals convey various yet mixed information ranging from linguistic to speaker-specific information. However, most of acoustic representations characterize all different kinds of information as whole, which could hinder either a speech or a speaker recognition (SR) system from producing a better performance. In this paper, we propose a novel deep neural architecture (DNA) especially for learning speaker-specific characteristics from mel-frequency cepstral coefficients, an acoustic representation commonly used in both speech recognition and SR, which results in a speaker-specific overcomplete representation. In order to learn intrinsic speaker-specific characteristics, we come up with an objective function consisting of contrastive losses in terms of speaker similarity/dissimilarity and data reconstruction losses used as regularization to normalize the interference of non-speaker-related information. Moreover, we employ a hybrid learning strategy for learning parameters of the deep neural networks: i.e., local yet greedy layerwise unsupervised pretraining for initialization and global supervised learning for the ultimate discriminative goal. With four Linguistic Data Consortium (LDC) benchmarks and two non-English corpora, we demonstrate that our overcomplete representation is robust in characterizing various speakers, no matter whether their utterances have been used in training our DNA, and highly insensitive to text and languages spoken. Extensive comparative studies suggest that our approach yields favorite results in speaker verification and segmentation. Finally, we discuss several issues concerning our proposed approach.

A contrastive study of verbs of remembering and forgetting in English and Spanish

This article presents a contrastive study of the verbs REMEMBER, FORGET and REMIND and their Spanish equivalents RECORDAR, ACORDARSE and OLVIDAR(SE), from a broad systemic-functional perspective. Through a database built with occurrences obtained from authentic corpora, a quantitative analysis was carried out on these verbs in terms of spoken and written mode and a number of clausal factors: mood, syntax and semantics of the Phenomenon, voice, polarity, person and modality. These factors were considered both individually and in relation to one another. The analysis uncovered similarities in the use of the English and Spanish verbs and, more significantly, a number of differences, due to a diversity of factors such as mode, lexical distribution, grammatical features and type of discourse, as well as politeness and other pragmatic reasons.

Web Based Hindi to Punjabi Machine Translation System

Hindi and Punjabi are closely related languages with lots of similarities in syntax and vocabulary Both Punjabi and Hindi languages have originated from Sanskrit which is one of the oldest language. In terms of speakers, Hindi is third most widely spoken language and Punjabi is twelfth most widely spoken language. Punjabi language is mostly used in the Northern India and in some areas of Pakistan as well as in UK, Canada and USA. Hindi is the national language of India and is spoken and used by the people all over the country. In the present research, Basic Hindi to Punjabi machine translation system using direct translation approach has been developed. The results of this translation system are surprisingly good. The system includes lexicon based translation, transliteration and continuously improving the system through machine learning module. It also takes care of basic word sense disambiguation.

Lattice-based Indexing for Spontaneous Mandarin Speech

We examine the task of spoken term detection in Chinese spontaneous speech with a lattice-based approach.We compare lattices generated with different units and lattices converted from one unit to another.We find that the best system is with toneless-syllable lattices converted from word lattices whose figure of merit (FOM) is 73.7 % from the baseline 69.2 %.By combining lattices from multiple systems into a single lattice and fully exploiting the redundant information in the combined lattice with a time-based node/arc merging,we achieve the result of a compact lattice index with the accuracy improved up to 79.2 %.

The Clinical Observation on the Curative Effect of Integrated Traditional Chinese and Western Medicine Treating 152 Cases of Lung Tuberculosis

Objective To investigate the therapeudc effect of treating lung mbereulosis with integrated traditional Chinese and western medicine. Methods 152 patients with lung tuberculosis aging from 25 to 35 years were randomly recruited into 2 groups. Western medicine and integrated traditional Chinese and western medicine were applied to each group respectively. Observe the therapeutic effects after the treatment. Results The group treated with integrated traditional Chinese and western medicine was significantly better than the group treated with western medicine in terms of therapeutic effect speaking, with P＜0.05. Conclusion The therapy of integrated traditional Chinese and western medicine has good effect in treating lung tuberculosis. Key words: Lung tuberculosis; Integrated traditional Chinese and western medicine

‘Is English we speaking’: Trinbagonian in the twenty-first century

Some notes and comments on the English usage of Trinidad and Tobago. The paper argues that for Trinidadians to think in terms of speaking and owning ‘only’ their distinctive Creole, setting aside their long-established indigenous variety of Standard English as if not really their own, is a complex distortion of social and linguistic reality. The reality has emerged from social, cultural, and psychological factors present in the Anglophone Caribbean at large, includes both conventional English and the Creole with which it inter-operates on a daily basis, and is an issue that stands in need of a positive revision that acknowledges the islands' dual inheritance.

The myth of a homogeneous speech Community: a sociolinguistic study of the speech of Japanese women in diverse gender roles

The overall objective of Ms paper is to provide empirical evidence against a mythical, stereotyped view of Japan as a homogeneous speech Community. As the most revealing variable, I have focused on the speech of Japanese women whose gender roles have undergone drastic transformation in contemporary society. The study uses a variationist approach to analyze the speech of three groups of women leading distinctive social lives: full-time homemakers, full-time employed women in clerical positions, and those in positions of authority. The results refute as overgeneralizations the Claims of past mainstream work on Japanese gender dijferentiation, which has consistently defined women s language use based exclusively on middleclass full-time homemakers under the influence of the traditional ideology of complimentary gender roles. Variable rule analysis reveals that differential performance grammars are operating among the three groups of women, and that the intergroup differentiation can be interpreted as social stratification more meaningfully correlated with Speakers' concrete occupation-bound categories than with abstract categories such as social class membership. Potential causesfor such dijferentiation are accountedfor in terms of Speakers' everyday contacts with people and types ofcommunicative routines and experiences in their occupation-bound communication networks.

The Southern Shift in Memphis, Tennessee

This article investigates the Southern Vowel Shift—a possibly interrelated series of rotations in vowel space currently affecting the dialects of southern speakers—in terms of examining its classification as a chain-shift process and, more generally, providing a descriptive account of the phonetic character of the changes in each individual vowel class. Based on the work of Labov (1991, 1994) and Feagin (1986), it has been suggested that the Southern Shift involves changes in both the front vowels and the back vowels, with the tense and lax front vowel nuclei essentially switching places and the back vowels moving forward. The relationship of these changes in the front vowels and those in the back vowels has not been firmly established, but they appear to be driven by different social and linguistic forces. What is happening to the low front, the mid, and the low back vowel classes in the Southern Shift has only been superficially explored. A detailed instrumental analysis of the vowel systems of 25 native Memphians of selected ages, socioeconomic classes, and genders is presented, revealing the movement of vowel classes which seem to be playing an important role in the instigation or perpetuation of the Southern Shift. This analysis points out discrepancies about how previously cited vocalic changes are embedded in mid-southern speech and provides a picture of how these changes are affecting other changes in the system. The results suggest that, while many of the changes cited in the literature are indeed present in the sample, the interrelatedness of these changes and their prognosis to move to completion are not so clear.

Meaning, belief, and language acquisition

Abstract A very plausible and common view of meaning supposes that linguistic meaning is to be understood in terms of speakers’ intentions. This program proposes to analyse the meaning of a sentence in terms of what speakers mean by or in uttering it; and this speaker meaning in turn is to be analysed in terms of the speaker's intentions. This essay argues that intention‐based semantics cannot provide an adequate analysis of linguistic meaning: not because of contrived counterexamples, nor because it conflicts with scruples about intentionality which we do or should have. It fails because research in psychology shows that children do not attribute beliefs to others in the way demanded by the theory. Empirical evidence is provided for the claim that two‐ and three‐year‐old children do not satisfy the conditions for speaker meaning, and thus cannot be said to mean anything by their utterances. It seems to me that children both mean something by their utterances and that their utterances have linguistic mean...

Evaluative communications between affectively ill and well mothers and their children.

Earlier research suggests that the natural verbal discourse of mothers with their children can be important in clarifying, verifying, and evaluating the behavior in which a child is engaged, in attributing qualities to the child, and in influencing the child's self-perceptions. We investigated the potential influences of parental affective illness (bipolar affective disorder and unipolar depression in contrast to no history of psychiatric illness) on such "labeling" behavior in a sample of 61 mothers and their older (school-age) and younger (preschool-age) children. It was hypothesized that the dispositions characterizing affective illness (specifically, negativity and disengagement) would be reflected in the labeling statements of mothers with a diagnosis as they interacted with their children. Based on videotaped interactions during a visit to a home-like laboratory apartment, labeling statements were identified in terms of speaker and person being labeled ("addressee") and coded (positive, negative, mixed, or neutral) for judgmental and affective quality of the statement and reaction of the addressee. Data were analyzed (a) by family unit and (b) my mother to child statements. The general pattern of findings indicated, in relative terms, an excess of negativity on the part of family members in the bipolar group and a dearth of negative affect for mothers in the unipolar group. Negativity in the bipolar group appeared to be especially likely when the setting involved mothers and two male children. Additionally, findings are discussed in terms of sex differences in vulnerability to depression.

Methods for the analysis of two‐party question and answer dialogues

Methods for the analysis of written two‐party question and answer dialogues are described. Written dialogue is distinguished from spoken dialogue and from descriptive prose. The concept of perspective of a dialogue is analyzed in terms of speaker, addressee, reporter and person reported to, is illustrated for several common dialogue contexts, and is applied to two short texts. Tense‐shifting in indirect speech is analyzed to illustrate temporal shifts in perspective. Two models are developed to describe turn‐taking in dialogues, and one of them is applied to a sample text. The notion of global structure, as developed in traditional theories of narrative, is discussed as it applies to dialogue, and a tentative analysis of the global structure is presented.

Observations on elicited language imitation with the severely retarded.

Word, phrase, and sentence imitation responses by 42 institutionalized, severely retarded adolescents were examined with a view to validity of imitation in language assessment with the retarded, and for developmental characteristics of language processing by retarded speakers. High response consistency in imitation, and semantic and syntactic substitutions were interpreted as evidence of internalized control of grammar by these speakers. A mental-age but not chronological-age correlation with imitation ability was found. One third of the speakers did not produce good imitations for three-word, Subject-Verb-Object phrases; speakers who did produce good imitations of such phrases could also imitate longer phrases and sentences. Evidence of metalinguistic awareness was presented, and results and teaching implications were discussed in terms of speakers' ability to process basic semantic relationships.

Multidimensional classification of abnormal voices

Twenty samples of voice disordered speakers were randomly selected from a library of 200 tape recordings to determine their perceptual attributes. Twenty-three subjects judged the similarity of the ordered pairs of these voices on a seven-point scale. One of the three groups of seven judges rated the hoarseness, breathiness, and nasality. The judged similarities were analyzed by INDSCAL. The physical measures of airflow, F0, F1, F2, F2/F1, F2−F1, and the acoustic spectrum and the psychophysical measures of hoarseness, breathiness, and nasality were used to interpret a five-dimensional INDSCAL solution. Dimension 1 was explained by absence versus presence of periodicity and by judgments of “hoarseness.” Dimension 2 was explained acoustically by elevation versus nonelevation of F2, and diagnostically by presence versus absence of mass lesion. Dimension 3 was explained physiologically by airflow rate and psychologically as “breathiness” and to a lesser extent, as “nasality.” Dimension 5 was interpreted acoustically in terms of speakers' F0. The five-dimensional solution accounted for 50% of the variance in the data.