Current Speaker Research Articles

Harnessing the power of Wav2Vec2 and CNNs for Robust Speaker Identification on the VoxCeleb and LibriSpeech Datasets

Speaker identification, a cornerstone of speech processing, involves associating individuals with spoken segments within a known speaker pool. This paper presents a significant AI contribution: an innovative framework tailored for closed-set speaker identification. It concurrently emphasizes its practical engineering application in the realm of speech analysis. This paper introduces a pioneering AI framework with substantial neural network architecture enhancements, particularly focusing on optimizing the Log-Softmax function—a linchpin for speaker attribution. Additionally, we seamlessly incorporate cutting-edge data augmentation techniques into the Wav2Vec2 framework. These innovations push the boundaries of current Speaker Identification methodologies. Empirical validation demonstrates our framework’s efficacy, yielding a remarkable relative improvement of up to 3.16% in top-1% accuracy compared to the state-of-the-art. This research sets a new benchmark, surpassing existing standards and unlocking the full potential of closed-set Speaker Identification functions. In addition, the methodology presented in this paper serves as a catalyst for advancing Speaker Identification methodologies in engineering applications, underlining the transformative potential of AI-driven innovations in this domain.

Insertional code-switching as interactional resource in Mandarin–English bilingual conversation

Aims and Objectives: This study investigates the interactional relevance of insertional code-switching, that is, insertion of other-language words/phrases, in Mandarin–English bilingual conversation. It focuses on a next speaker’s (dis)alignment with an insertion in an adjacency pair and its import to the ongoing interaction. Methodology: This study adopts the framework of conversation analysis, which seeks to describe the underlying organization of talk-in-interaction. It features analyses warranted by techniques and signals in conversation that are demonstrably relevant to conversation participants. Data and Analysis: The data come from CallFriend Mandarin Corpus. Speakers are native speakers of Mandarin Chinese who resided in the United States at the time of recording. First, all instances of insertional code-switching were extracted with the extended context, filtering out those not occurring in an adjacency pair. Next, sequential analysis was conducted to examine the interactional import of a next speaker’s alignment and disalignment, respectively, when insertional code-switching occurs in an adjacency pair. Findings: This study discovers that a next speaker’s alignment with an insertion is sequentially preferred over disalignment. When different language choice over the same word occurs, speakers orient to it as problematic and then repair it, such that, alignment is achieved. Disalignment is used to claim the current speaker’s more authoritative status in the matter concerned. Originality: A large bulk of literature on insertional code-switching has focused on its syntactic constraints and semantic conditions; its interactional functions in bilingual conversation have not received as much attention. This study reveals that insertional code-switching works as an extra resource for bilingual speakers to achieve interactional alignment. Significance: This study challenges the view that insertional code-switching is not sequentially implicative by showing that it does place sequential constraints on the next speaker. It further argues that such sequential implicativeness can be exploited to establish finely calibrated alignment and index epistemic status.

Competitive points in Mandarin-speaking multiparty interaction: Speakership and epistemics

This study explores the use of hand points in competitive epistemic and speakership environments in Mandarin-speaking conversation. It examines incoming speakers’ points at a current speaker. This study employs multimodal conversation-analytic methods to analyse 334 min of triadic interaction. 40 points directed at a current speaker by an incoming speaker are the focus of analysis. This study finds that these points indicate an upcoming bid for the floor, thereby carrying out self-selection for speakership. In addition, the turn introduced by the incoming speaker's point suggests that they know at least as much as the current speaker about the issue at hand, and can implicate epistemic and speakership competition. As such, these points foreshadow delivery of a weakly aligning turn or a disaligning turn, and may accomplish either affiliation or disaffiliation. This study generates new knowledge about the interactional functions of points in multiparty conversation, and suggests potential phenomena for cross-linguistic comparison.

PASTS: Progress-aware spatio-temporal transformer speaker for vision-and-language navigation

Vision-and-language navigation (VLN) is a crucial but challenging cross-modal navigation task. One powerful technique to enhance the generalization performance in VLN is the use of an independent speaker model to provide pseudo instructions for data augmentation. However, current speaker models based on Long-Short Term Memory (LSTM) lack the ability to attend to features relevant at different locations and time steps. To address this, we propose a novel progress-aware spatio-temporal transformer speaker (PASTS) model that uses the transformer as the core of the network. PASTS uses a spatio-temporal encoder to fuse panoramic representations and encode intermediate connections through steps. Besides, to avoid the misalignment problem that could result in incorrect supervision, a speaker progress monitor (SPM) is proposed to enable the model to estimate the progress of instruction generation and facilitate more fine-grained caption results. Additionally, a multifeature dropout (MFD) strategy is introduced to alleviate overfitting. The proposed PASTS is flexible to be combined with existing VLN models. The experimental results demonstrate that PASTS outperforms previous speaker models and successfully improves the performance of previous VLN models, achieving state-of-the-art performance on the standard Room-to-Room (R2R) dataset.

Claiming insufficient knowledge in pairwork and groupwork classroom activities

This conversation analytic paper explores how students in pair and groupwork tasks produce and respond to claims of insufficient knowledge (CIKs). Based on 7 h of video recordings of peer interaction from 18 English as a Foreign Language (EFL) classes in Czech secondary schools, we analyze how students express and negotiate their epistemic status using CIKs: when producing a CIK, the current speaker assumes a not-knowing status, thus making the imminent speaker change more relevant, as the next speaker then typically reveals his or her epistemic status. We also show that when a CIK is produced dyadic interactions in second position, the first speaker then produces a knowledge display response, or another CIK, resulting in abandoning the question, which differs from sequences that can be found in frontal teaching. The findings also show that CIKs can be used to resume task-related talk and initiate repair sequences focusing on language issues that the task comprises. Thus, CIKs can be viewed as central interactional resources for students to manage the task, i.e., to invite others to contribute, to resume their talk, or to abandon the current question, and to initiate repair sequences focusing on problematic items from the task.

L2 grammar‐for‐interaction: Functions of “and”‐prefaced turns in L2 students’ collaborative talk

AbstractThis article examines how second language (L2) interactional competence is manifested in students’ use of “and”‐prefaced turns when doing meaning‐focused oral tasks in pairs and small groups. Drawing on video recordings from English‐as‐a‐foreign‐language upper‐secondary classes recorded in Czechia and Finland, 86 sequences involving “and”‐prefaced turns were scrutinized using multimodal conversation analysis, focusing on language, gaze, and material resources. The findings suggest that by producing “and”‐prefaced turns, students orient to task progression. These turns have two functions: task managerial and contribution to the emerging task answer. By using task‐managerial “and”‐prefaced turns, the current speaker invites another student to participate, while in “and”‐prefaced contributions to the task answer, a participant adds to, generalizes, or modifies the previous task answer. The analysis shows that students mobilized their L2 interactional competence in producing “and”‐prefaced turns in close coordination with embodied resources and with respect to the spatio‐material surroundings and the nature of the task. These findings contribute to the multimodal reconceptualization of the grammar–body interface and research on turn‐initial particles within L2 interactional competence.

Waḷḷāhi (‘by God’) as a marker of commitment and involvement in Egyptian Arabic conversation

This paper explores the interactional uses of the oath-expression waḷḷāhi ‘by God’ in Egyptian Arabic. Based on everyday conversational data (Arabic CALLHOME), and drawing on conversation-analytic and interactional linguistics methods, the analysis shows that waḷḷāhi rarely introduces a solemn act of swearing; rather, it is conventionally used as a marker expressing or inviting the expression of commitment and certainty. waḷḷāhi can occur in turn-initial and utterance-final position, where it frames or modifies the current speaker’s contribution, or as a freestanding response token. In epistemically divergent contexts, waḷḷāhi serves to resist the other party’s assumptions or claims about a certain state of affairs by asserting commitment to a different state of affairs, or by questioning the certainty of the prior speaker. Commitment indexed by turn-initial tokens is relativized to the speaker’s potentially restricted access and rights to knowledge, and is therefore more reserved; utterance-final tokens do not invoke this subjective reference and serve to express absolute commitment. In epistemically congruent and affiliative contexts, waḷḷāhi undergoes indexical reanalysis and is interpreted as an index of involvement, earnestness, and interest. The analysis aligns with a widely observed path of semantic and pragmatic development of discourse markers whereby markers come to assume more specific relational functions as they extend to new contexts of use.

The influence of task complexity and task modality on learners’ topic and turn management

Abstract Peer interaction is examined as an important indicator of performance in task-based language learning. However, little research has explored learners’ interactional competence under the influences of task factors. This study investigates the effects of task complexity and task modality on learners’ interactional competence in terms of topic management and turn management. Four tasks were designed – written-simple, written-complex, oral-simple, and oral-complex – and 40 Chinese EFL learners completed them in counterbalanced sequences. The results showed the significant influences of task complexity and task modality on learners’ topic and turn management. Task modality and complexity interactively influenced learners when they were expanding on topics in the role of the current speaker. Considering the cognitive demands of increasing complexity and differences between the two modalities, this paper discusses the implications of the findings for developing L2 interactional competence in speech- and text-based interactions.

When English complement clauses meet evidential adverbs

There is much literature about the licensing of complement clauses by complement-taking predicates. However, less has been written about the licensing of adverbs in a complement clause. This article addresses the licensing of English evidential adverbs in complement clauses extracted from the NOW corpus. The article discusses three factors that determine the distribution of evidential adverbs in complement clauses. These are the nature of the evidential adverbs, the constraints of the complement clause and the anchor of the evidential adverb. To explore the role of these three factors, I adopt the hierarchical scopal theory of Functional Discourse Grammar (FDG). If a complement clause licenses the inclusion of the evidential adverb, then there is a match. Should there be no alignment between the complement clause and the adverb, there will be a mismatch. The results of the analysis of the data show that there are mostly matches, which occur with either a current speaker anchor or an actor anchor. Secondly, it appears that in cases of mismatches, there is always a current speaker. It thus appears that a current speaker anchor can override the constraints of the complement clause.

A Study of the Sequential Environments of Turn-Initial “Laughter”—A Conversation Analytic Approach

“Laughter” is a common language in human society. The “laughter” in the turn-initial position is a special one, which is of great significance for analyzing the social action of the whole turn. Using the method of conversation analysis, this paper explores the sequential environments of “laughter” which is in the turn-initial position of the daily Chinese communication. This analysis finds that “laughter” in the turn-initial position often appears in the following three sequential environments: after the previous speaker’s self-deprecating; the current speaker complains to the listener about the listener or third party; after the previous speaker has stated the difficulty he or she is facing in his or her turn.

Neither speaker nor recipient: The middle-distance look of unaddressed participants

This article studies the middle-distance look of ratified unaddressed participants and the way in which it is sequentially organized in relation to the discursive conducts of current speaker. An analysis of naturally occurring video recorded discourse reveals that ratified unaddressed participants constitute themselves as non-recipients by employing middle-distance look. The article formulates a list of observations (missing in the literature) regarding the middle-distance look of ratified unaddressed participants. It raises questions which may generate knowledge on (a) the connection between middle-distance look and discursive ambiguity, (b) types of middle-distance look, (c) middle distance look as related to seating arrangements and group size.

Functions of Pauses and Hesitations in Ola Rotimi’s Our Husband Has Gone Mad Again

All normal human beings have the ability to pass across different messages in different contexts without necessarily making use of the oral form of language. These other ways of passing across messages and meanings without using the oral form include gestures, facial expressions, vocalisers, interjections as well as pauses and hesitations which are the subjects of this study. The data for the study are extracted from Ola Rotimi’s play entitled Our Husband Has Gone Mad Again using insight from the concept of context for its analysis. The play is selected for the research because it contains several instances of the use of pauses and hesitations which are capable of yielding novel functions of the two phenomena. The novel functions of pauses and hesitations discovered in the data are: pauses and hesitations perform the function of emphasis when they occur together; hesitation is used to indicate that the current speaker is emotionally unstable; hesitation and pauses are used to demonstrate strategic competence during interactions; in addition, the two phenomena when deployed effectively in the appropriate context may indicate fear of the person that one is interacting with; lastly, hesitations can be used to pass across certain feelings that may be appropriately interpreted in context.

Przyczynek do teorii cytowania Jadwigi Wajszczuk

The article consists of two parts. In the first part, the citation theory outlined by J. Wajszczuk (2005) is recalled. The second part contains an analysis of three Polish verbs referring to abusive quoting: przekręcić [coś] ‘misrepresent [something]’, przeinaczyć [coś] ‘twist [something]’ i wypaczyć [coś] ‘distort [something]’. The first of the lexemes has the broadest reference. It refers to both citation and assertory speaking. The verb wypaczyć ‘to distort [something]’ refers to a situation of significant interference in someone else’s text in conjunction with a current speaker’s’ negative evaluation. The verb przeinaczyć [coś] ‘to twist [something]’ allows for the lessening of the speaker’s guilt, who infidelly updates someone else’s statement. The verbs of the tainted message describe the logical completion of faithful quotation. They emphasize the fact that a violation of the unwritten imperative to treat someone else’s speech may be more or less harmful.

The distinctive uses of right in British and American English interaction

This paper explores distinct usages of the response particle right in American versus British English conversation. The analysis shows that, in American English, right conveys the speaker's knowing stance and, in certain environments, the speaker's claim of primary knowledge. In contrast, in British English, right registers provided information as previously unknown, informative, and relevant to the current speaker's ongoing project. The analysis draws on large corpora of audio- and video-recorded ordinary and institutional interactions in British and American English. We use the methodology of Conversation Analysis to examine sequential environments in which right is used, its interactional import, and prosodic realizations.

The role of reported speech in the evaluation of arguments from expert opinion

This paper discusses how grammatical aspects of direct and indirect speech affect the evaluation of the argument from expert opinion. Direct speech is often characterized by having two different discourse acts and, consequently, two deictic centres; and indirect speech is regarded as one discourse act with one deictic centre. These grammatical structures have different pragmatic effects, mainly because in direct speech, the reported content is assumed to be faithful in form and content. In contrast, in indirect speech, the current speaker embeds the reported content in their speech. Considering these features, this work provides a comparative analysis of the types of reported speech and their relation to Wagemans’ (2011) argument scheme and critical questions and offers modifications in the existing argument schemes to allow a more precise evaluation of arguments.

Interactivity/Reciprocity (Online Discussions/Discussion Quality)

Interactivity/Reciprocity (Online Discussions/Discussion Quality)

Turn-Taking Mechanism in Ola Rotimi’s The Gods are not to Blame

Conversation analysis is an approach which deals with verbal and non-verbal aspects of language in a situation of everyday life. Dialogues in Ola Rotimi’s The gods are not to blame are classified as naturally occurring conversations because they involve two or more interlocutors (characters) who employ turns to deliberate on issues that are related to the roles the gods play in the lives of human beings. So, as pertinent as conversation analysis is in dealing with matters bothering human beings, no enough research has been carried out on it. Therefore, this paper seeks to undertake study on turn-taking mechanism in the above named text so as to see how it is being used by the characters in it. The paper adopts sacks et al, (1974) conversational model as its theoretical framework. The text is analysed and the results show that aspects of turn-taking mechanism such as starting up, taking over, interruption, filled pauses or verbal fillers, silent fillers ellipses, questions, declarative sentences and silence run through the text. Lexical items such as “oh”, “well”, “no” are also used in the text to take turns. Turns are allocated on the bases of current speaker select next speaker, current speaker self-select and current speaker continuous. The study also found that the use of turn-taking mechanism in the text makes the conversations run smoothly. The study concludes that conversation analysis is an excellent tool for studying naturally occurring conversations. Glob Acad J Linguist Lit; 2022, 4(3): - DOI :

Loosely Portrayed Speech in Interaction: Constructing Multiple Complainable Utterances

ABSTRACT Conversation analysis is used in investigating the interactional uses of loosely portrayed speech in interaction. This device combines elements of direct and indirect portrayal, conveying some fidelity to an original while, at the same time, indicating that it is not verbatim enactment of specific utterances. The instances in the current collection are in English, deriving from informal interaction, mainly telephone calls recorded in the UK and USA. They occur in complaints about a third party, recurrently by portraying the reported speaker’s criticisms of the current speaker. The reported speaker is depicted as making multiple criticisms, which adds to the reprehensible nature of their actions. By constructing the reported speaker’s actions, and, at the same time, indicating the stance of the current speaker toward them, the complained-about speaker’s behavior is portrayed as infringing the moral order and therefore the complaint as legitimate. Data are in British and American English.

Emotional conversation generation with heterogeneous graph neural network

The successful emotional conversation system depends on sufficient perception and appropriate expression of emotions. In a real-life conversation, humans firstly instinctively perceive emotions from multi-source information, including the emotion flow hidden in dialogue history, facial expressions, audio, and personalities of speakers. Then, they convey suitable emotions according to their personalities, but these multiple types of information are insufficiently exploited in emotional conversation fields. To address this issue, in this paper, we propose a heterogeneous graph-based model for emotional conversation generation. Firstly, we design a Heterogeneous Graph-Based Encoder to represent the conversation content (i.e., the dialogue history, its emotion flow, facial expressions, audio, and speakers' personalities) with a heterogeneous graph neural network, and then predict suitable emotions for feedback. Secondly, we employ an Emotion-Personality-Aware Decoder to generate a response relevant to the conversation context as well as with appropriate emotions, through taking the encoded graph representations, the predicted emotions by the encoder and the personality of the current speaker as inputs. Experiments on both automatic and human evaluation show that our method can effectively perceive emotions from multi-source knowledge and generate a satisfactory response. Furthermore, based on the up-to-date text generator BART, our model still can achieve consistent improvement, which significantly outperforms some existing state-of-the-art models.

The effect of speech degradation on the ability to track and predict turn structure in conversation

Conversation represents a considerable amount of the daily language usage and plays an important role in language acquisition. In conversation, listeners simultaneously process their interlocutor's turn and prepare their own next turn. As such the turn-taking dynamics heavily relies on prediction. In other words, listeners avail prior knowledge to constrain both speech perception and production. Here we explored the relation between prediction and comprehension while watching two interlocutors having a conversation. We capitalize on gaze switch as a proxy of predictive behaviour to class dialogue turns as more or less well predicted and explore how this affects dialogue comprehension. Moreover, we study the extent to which speech degradation, by increasing the global uncertainty of the context, affects the relation between predictions, brain correlates of prediction errors (N400) and global comprehension. Results show that 1) listeners direct gaze to the current speaker, in particular in challenging conditions, 2) gaze behaviour possibly affects the semantic processing of the upcoming turn (N400), 3) participants with a more efficient gaze predictive behaviour better solve semantic uncertainties at the turn onset, in particular in the most challenging listening condition. Our findings contribute to a better understanding of the relation between predictions, neural predictions errors and speech comprehension under challenging conditions.