We introduce Speech-based Intelligence Quotient (SIQ) as a new form of human cognition-inspired evaluation pipeline for voice understanding large language models (LLM_Voice), designed to assess their voice understanding ability. Moving beyond popular voice understanding metrics such as word error rate (WER), SIQ examines LLM_Voice across three cognitive levels motivated by Bloom’s Taxonomy: (1) Remembering (i.e., WER for verbatim accuracy); (2) Understanding (i.e., similarity of LLM’s interpretations); and (3) Application (i.e., QA accuracy for simulating downstream tasks). We demonstrate that SIQ not only quantifies voice understanding abilities but also provides unified comparisons between cascaded methods (e.g., ASR-LLM) and end-to-end models, identifies annotation errors in existing benchmarks, and detects hallucinations in LLM_Voice. Our framework represents a first-of-its-kind intelligence examination that bridges cognitive principles with voice-oriented benchmarks, while exposing overlooked challenges in multi-modal training. Our code and data will be open source to encourage future studies.

Large Language Models (LLMs) have shown impressive reasoning capabilities, yet existing prompting methods face a critical trade-off: simple approaches often struggle with complex tasks and reasoning stability, while more sophisticated methods require multiple inferences and substantial computational resources, limiting their practical deployment. To address this challenge, we propose Derailer-Rerailer, a novel framework that adaptively balances reasoning accuracy and computational efficiency. At its core, our framework employs a lightweight Derailer mechanism to assess reasoning stability and selectively triggers an advanced Rerailer verification process only when necessary, thereby optimizing computational resource usage. Extensive evaluation across both open and closed-source models on more than 20 categories of mathematical, symbolic, and commonsense reasoning tasks demonstrates our framework’s effectiveness: Derailer-Rerailer achieves significant accuracy improvements (8-11% across various reasoning tasks) while maintaining 2-3 times better efficiency than existing verification methods, with particularly strong performance in mathematical and symbolic reasoning, offering a practical solution for enhancing LLM reasoning reliability while significantly reducing computational overhead.

This paper explores emotion-aware speech-to-text translation (ST) using generative error correction (GER) by large language models (LLMs). Despite recent advancements in ST, the impact of the emotional content has been overlooked. First, we enhance the translation of emotional speech by adopting the GER paradigm: Finetuned an LLM to generate the translation based on the decoded N-best hypotheses. Moreover, we combine the emotion and sentiment labels into the LLM finetuning process to enable the model to consider the emotion content. In addition, we project the ST model’s latent representation into the LLM embedding space to further improve emotion recognition and translation. Experiments on an English-Chinese dataset show the effectiveness of the combination of GER, emotion and sentiment labels, and the projector for emotion-aware ST. Our code is available at https://github.com/N-Orien/EmoST.

Self-consistency mitigates hallucinations in Large Language Models (LLMs) by sampling multiple reasoning paths, but it lacks a systematic approach to determine the optimal number of samples or select the most faithful rationale. To address this limitation, we introduce Reasoning-Aware Self-Consistency (RASC), a novel framework that enhances sampling efficiency and reasoning faithfulness by dynamically evaluating both outputs and rationales. RASC assesses the quality of reasoning and the consistency of answers for each generated sample, using these assessments to guide early stopping decisions and rationale selection. The framework employs criteria-based stopping and weighted majority voting, enabling more informed choices on when to halt sampling and which rationale to select. Our comprehensive experiments across diverse question-answering datasets demonstrate that RASC outperforms existing methods, reducing sample usage by approximately 70% while maintaining accuracy. Moreover, RASC facilitates the selection of high-fidelity rationales, thereby improving the faithfulness of LLM outputs. Our approach effectively addresses the efficiency-accuracy trade-off in LLM reasoning tasks, offering a new perspective for more nuanced, faithful, and effective utilization of LLMs in resource-constrained environments.

Despite recent advances in the general visual instruction-following ability of Multimodal Large Language Models (MLLMs), they still struggle with critical problems when required to provide a precise and detailed response to a visual instruction: (1) failure to identify novel objects or entities, (2) mention of non-existent objects, and (3) neglect of object’s attributed details. Intuitive solutions include improving the size and quality of data or using larger foundation models. They show effectiveness in mitigating these issues, but at an expensive cost of collecting a vast amount of new data and introducing a significantly larger model. Standing at the intersection of these approaches, we examine the three object-oriented problems from the perspective of the image-to-text mapping process by the multimodal connector. In this paper, we first identify the limitations of multimodal connectors stemming from insufficient training data. Driven by this, we propose to enhance the mapping with retrieval-augmented tag tokens, which contain rich object-aware information such as object names and attributes. With our Tag-grounded visual instruction tuning with retrieval Augmentation (TUNA), we outperform baselines that share the same language model and training data on 12 benchmarks. Furthermore, we show the zero-shot capability of TUNA when provided with specific datastores.

We present a new task, speech dialogue translation mediating speakers of different languages. We construct the SpeechBSD dataset for the task and conduct baseline experiments. Furthermore, we consider context to be an important aspect that needs to be addressed in this task and propose two ways of utilizing context, namely monolingual context and bilingual context. We conduct cascaded speech translation experiments using Whisper and mBART, and show that bilingual context performs better in our settings.

As the core of task-oriented dialogue systems, dialogue state tracking (DST) is designed to track the dialogue state through the conversation between users and systems. Multi-domain DST has been an important challenge in which the dialogue states across multiple domains need to consider. In recent mainstream approaches, each domain and slot are aggregated and regarded as a single query feeding into attention with the dialogue history to obtain domain-slot specific representations. In this work, we propose disentangled domain-slot attention for multi-domain dialogue state tracking. The proposed approach disentangles the domain-slot specific information extraction in a flexible and context-dependent manner by separating the query about domains and slots in the attention component. Through a series of experiments on MultiWOZ 2.0 and MultiWOZ 2.4 datasets, we demonstrate that our proposed approach outperforms the standard multi-head attention with aggregated domain-slot query.

This paper describes the Kyoto speech-to-speech translation system for IWSLT 2023. Our system is a combination of speech-to-text translation and text-to-speech synthesis. For the speech-to-text translation model, we used the dual-decoderTransformer model. For text-to-speech synthesis model, we took a cascade approach of an acoustic model and a vocoder.

Dialogue state tracking (DST) is designed to track the dialogue state during the conversations between users and systems, which is the core of task-oriented dialogue systems. Mainstream models predict the values for each slot with fully token-wise slot attention from dialogue history. However, such operations may result in overlooking the neighboring relationship. Moreover, it may lead the model to assign probability mass to irrelevant parts, while these parts contribute little. It becomes severe with the increase in dialogue length. Therefore, we investigate sparse local slot attention for DST in this work. Slot-specific local semantic information is obtained at a sub-sampled temporal resolution capturing local dependencies for each slot. Then these local representations are attended with sparse attention weights to guide the model to pay attention to relevant parts of local information for subsequent state value prediction. The experimental results on MultiWOZ 2.0 and 2.4 datasets show that the proposed approach effectively improves the performance of ontology-based dialogue state tracking, and performs better than token-wise attention for long dialogues.

Language models (LM) have played crucial roles in automatic speech recognition (ASR) to enhance end-to-end (E2E) ASR systems’ performance. There are two categories of approaches: finding better ways to integrate LMs into ASR systems and adapting on LMs to the task domain. This article will start with a reflection of interpolation-based integration methods of E2E ASR’s scores and LM’s scores. Then we will focus on LM augmentation approaches based on the noisy channel model, which is intrigued by insights obtained from the above reflection. The experiments show that we can enhance an ASR E2E model based on encoder-decoder architecture by pre-training the decoder with text data. This implies the decoder of an E2E model can be treated as an LM and reveals the possibility of enhancing the E2E model without an external LM. Based on those ideas, we proposed the implicit language model canceling method and then did more discussion about the decoder part of an E2E ASR model. The experimental results on the TED-LIUM2 dataset show that our approach achieves a 3.4% relative WER reduction compared with the baseline system, and more analytic experiments provide concrete experimental supports for our assumption.

With the advent of the General Data Protection Regulation (GDPR) and increasing privacy concerns, the sharing of speech data is faced with significant challenges. Protecting the sensitive content of speech is the same important as the voiceprint. This paper proposes an effective speech content protection method by constructing a frame-by-frame adversarial speech generation system. We revisited the adversarial examples generating method in the recent machine learning field and selected the phonetic state sequence of sensitive speech for the adversarial examples generation. We build an adversarial speech collection. Moreover, based on the speech collection, we proposed a neural network-based frame-by-frame mapping method to recover the speech content by converting from the adversarial speech to the human speech. Experiment shows our proposed method can encode and recover any sensitive audio, and our method is easy to be conducted with publicly available resources of speech recognition technology.

As an important component of task-oriented dialogue systems, dialogue state tracking is designed to track the dialogue state through the conversations between users and systems. Multi-domain dialogue state tracking is a challenging task, in which the correlation among different domains and slots needs to consider. Recently, slot self-attention is proposed to provide a data-driven manner to handle it. However, a full-support slot self-attention may involve redundant information interchange. In this paper, we propose a top-k attention-based slot self-attention for multi-domain dialogue state tracking. In the slot self-attention layers, we force each slot to involve information from the other k prominent slots and mask the rest out. The experimental results on two mainstream multi-domain task-oriented dialogue datasets, MultiWOZ 2.0 and MultiWOZ 2.4, present that our proposed approach is effective to improve the performance of multi-domain dialogue state tracking. We also find that the best result is obtained when each slot interchanges information with only a few slots.

Knowledge graphs suffer from sparsity which degrades the quality of representations generated by various methods. While there is an abundance of textual information throughout the web and many existing knowledge bases, aligning information across these diverse data sources remains a challenge in the literature. Previous work has partially addressed this issue by enriching knowledge graph entities based on “hard” co-occurrence of words present in the entities of the knowledge graphs and external text, while we achieve “soft” augmentation by proposing a knowledge graph enrichment and embedding framework named Edge. Given an original knowledge graph, we first generate a rich but noisy augmented graph using external texts in semantic and structural level. To distill the relevant knowledge and suppress the introduced noise, we design a graph alignment term in a shared embedding space between the original graph and augmented graph. To enhance the embedding learning on the augmented graph, we further regularize the locality relationship of target entity based on negative sampling. Experimental results on four benchmark datasets demonstrate the robustness and effectiveness of Edge in link prediction and node classification.

Offensive language detection (OLD) has received increasing attention due to its societal impact. Recent work shows that bidirectional transformer based methods obtain impressive performance on OLD. However, such methods usually rely on large-scale well-labeled OLD datasets for model training. To address the issue of data/label scarcity in OLD, in this paper, we propose a simple yet effective domain adaptation approach to train bidirectional transformers. Our approach introduces domain adaptation (DA) training procedures to ALBERT, such that it can effectively exploit auxiliary data from source domains to improve the OLD performance in a target domain. Experimental results on benchmark datasets show that our approach, ALBERT (DA), obtains the state-of-the-art performance in most cases. Particularly, our approach significantly benefits underrepresented and under-performing classes, with a significant improvement over ALBERT.

Emotion recognition in dialogue systems has gained attention in the field of natural language processing recent years, because it can be applied in opinion mining from public conversational data on social media. In this paper, we propose a hierarchical model to recognize emotions in the dialogue. In the first layer, in order to extract textual features of utterances, we propose a convolutional self-attention network(CAN). Convolution is used to capture n-gram information and attention mechanism is used to obtain the relevant semantic information among words in the utterance. In the second layer, a GRU-based network helps to capture contextual information in the conversation. Furthermore, we discuss the effects of unidirectional and bidirectional networks. We conduct experiments on Friends dataset and EmotionPush dataset. The results show that our proposed model(CAN-GRU) and its variants achieve better performance than baselines.

Given a question and a set of candidate answers, answer selection is the task of identifying which of the candidates answers the question correctly. It is an important problem in natural language processing, with applications in many areas. Recently, many deep learning based methods have been proposed for the task. They produce impressive performance without relying on any feature engineering or expensive external resources. In this paper, we aim to provide a comprehensive review on deep learning methods applied to answer selection.

Treebank conversion is a straightforward and effective way to exploit various heterogeneous treebanks for boosting parsing performance. However, previous work mainly focuses on unsupervised treebank conversion and has made little progress due to the lack of manually labeled data where each sentence has two syntactic trees complying with two different guidelines at the same time, referred as bi-tree aligned data. In this work, we for the first time propose the task of supervised treebank conversion. First, we manually construct a bi-tree aligned dataset containing over ten thousand sentences. Then, we propose two simple yet effective conversion approaches (pattern embedding and treeLSTM) based on the state-of-the-art deep biaffine parser. Experimental results show that 1) the two conversion approaches achieve comparable conversion accuracy, and 2) treebank conversion is superior to the widely used multi-task learning framework in multi-treebank exploitation and leads to significantly higher parsing accuracy.

When evaluating a potential product purchase, customers may have many questions in mind. They want to get adequate information to determine whether the product of interest is worth their money. In this paper we present a simple deep learning model for answering questions regarding product facts and specifications. Given a question and a product specification, the model outputs a score indicating their relevance. To train and evaluate our proposed model, we collected a dataset of 7,119 questions that are related to 153 different products. Experimental results demonstrate that –despite its simplicity– the performance of our model is shown to be comparable to a more complex state-of-the-art baseline.