Reverberation can severely degrade the quality of speech signals recorded using microphones in an enclosure. In acoustic sensor networks with spatially distributed microphones, a similar dereverberation performance may be achieved using only a subset of all available microphones. Using the popular convex relaxation method, in this paper we propose to perform microphone subset selection for the weighted prediction error (WPE) multi-channel dereverberation algorithm by introducing a group sparsity penalty on the prediction filter coefficients. The resulting problem is shown to be solved efficiently using the accelerated proximal gradient algorithm. Experimental evaluation using measured impulse responses shows that the performance of the proposed method is close to the optimal performance obtained by exhaustive search, both for frequency-dependent as well as frequency-independent microphone subset selection. Furthermore, the performance using only a few microphones for frequency-independent microphone subset selection is only marginally worse than using all available microphones.
In this work, we propose a novel method for modeling numerous speakers, which enables expressing the overall characteristics of speakers in detail like a trained multi-speaker model without additional training on the target speaker's dataset. Although various works with similar purposes have been actively studied, their performance has not yet reached that of trained multi-speaker models due to their fundamental limitations. To overcome previous limitations, we propose effective methods for feature learning and representing target speakers' speech characteristics by discretizing the features and conditioning them to a speech synthesis model. Our method obtained a significantly higher similarity mean opinion score (SMOS) in subjective similarity evaluation than seen speakers of a best-performing multi-speaker model, even with unseen speakers. The proposed method also outperforms a zero-shot method by significant margins. Furthermore, our method shows remarkable performance in generating new artificial speakers. In addition, we demonstrate that the encoded latent features are sufficiently informative to reconstruct an original speaker's speech completely. It implies that our method can be used as a general methodology to encode and reconstruct speakers' characteristics in various tasks.
Emotion recognition is a topic of significant interest in assistive robotics due to the need to equip robots with the ability to comprehend human behavior, facilitating their effective interaction in our society. Consequently, efficient and dependable emotion recognition systems supporting optimal human-machine communication are required. Multi-modality (including speech, audio, text, images, and videos) is typically exploited in emotion recognition tasks. Much relevant research is based on merging multiple data modalities and training deep learning models utilizing low-level data representations. However, most existing emotion databases are not large (or complex) enough to allow machine learning approaches to learn detailed representations. This paper explores modalityspecific pre-trained transformer frameworks for self-supervised learning of speech and text representations for data-efficient emotion recognition while achieving state-of-the-art performance in recognizing emotions. This model applies feature-level fusion using nonverbal cue data points from motion capture to provide multimodal speech emotion recognition. The model was trained using the publicly available IEMOCAP dataset, achieving an overall accuracy of 77.58% for four emotions, outperforming state-of-the-art approaches
We consider the task of animating 3D facial geometry from speech signal. Existing works are primarily deterministic, focusing on learning a one-to-one mapping from speech signal to 3D face meshes on small datasets with limited speakers. While these models can achieve high-quality lip articulation for speakers in the training set, they are unable to capture the full and diverse distribution of 3D facial motions that accompany speech in the real world. Importantly, the relationship between speech and facial motion is one-to-many, containing both inter-speaker and intra-speaker variations and necessitating a probabilistic approach. In this paper, we identify and address key challenges that have so far limited the development of probabilistic models: lack of datasets and metrics that are suitable for training and evaluating them, as well as the difficulty of designing a model that generates diverse results while remaining faithful to a strong conditioning signal as speech. We first propose large-scale benchmark datasets and metrics suitable for probabilistic modeling. Then, we demonstrate a probabilistic model that achieves both diversity and fidelity to speech, outperforming other methods across the proposed benchmarks. Finally, we showcase useful applications of probabilistic models trained on these large-scale datasets: we can generate diverse speech-driven 3D facial motion that matches unseen speaker styles extracted from reference clips; and our synthetic meshes can be used to improve the performance of downstream audio-visual models.
Due to the successful application of deep learning, audio spoofing detection has made significant progress. Spoofed audio with speech synthesis or voice conversion can be well detected by many countermeasures. However, an automatic speaker verification system is still vulnerable to spoofing attacks such as replay or Deep-Fake audio. Deep-Fake audio means that the spoofed utterances are generated using text-to-speech (TTS) and voice conversion (VC) algorithms. Here, we propose a novel framework based on hybrid features with the self-attention mechanism. It is expected that hybrid features can be used to get more discrimination capacity. Firstly, instead of only one type of conventional feature, deep learning features and Mel-spectrogram features will be extracted by two parallel paths: convolution neural networks and a short-time Fourier transform (STFT) followed by Mel-frequency. Secondly, features will be concatenated by a max-pooling layer. Thirdly, there is a Self-attention mechanism for focusing on essential elements. Finally, ResNet and a linear layer are built to get the results. Experimental results reveal that the hybrid features, compared with conventional features, can cover more details of an utterance. We achieve the best Equal Error Rate (EER) of 9.67\% in the physical access (PA) scenario and 8.94\% in the Deep fake task on the ASVspoof 2021 dataset. Compared with the best baseline system, the proposed approach improves by 74.60\% and 60.05\%, respectively.
This study focuses on emotion-sensitive spoken dialogue in human-machine speech interaction. With the advancement of Large Language Models (LLMs), dialogue systems can handle multimodal data, including audio. Recent models have enhanced the understanding of complex audio signals through the integration of various audio events. However, they are unable to generate appropriate responses based on emotional speech. To address this, we introduce the Emotional chat Model (E-chat), a novel spoken dialogue system capable of comprehending and responding to emotions conveyed from speech. This model leverages an emotion embedding extracted by a speech encoder, combined with LLMs, enabling it to respond according to different emotional contexts. Additionally, we introduce the E-chat200 dataset, designed explicitly for emotion-sensitive spoken dialogue. In various evaluation metrics, E-chat consistently outperforms baseline LLMs, demonstrating its potential in emotional comprehension and human-machine interaction.
Emotional voice conversion (EVC) seeks to modify the emotional tone of a speaker's voice while preserving the original linguistic content and the speaker's unique vocal characteristics. Recent advancements in EVC have involved the simultaneous modeling of pitch and duration, utilizing the potential of sequence-to-sequence (seq2seq) models. To enhance reliability and efficiency in conversion, this study shifts focus towards parallel speech generation. We introduce Duration-Flexible EVC (DurFlex-EVC), which integrates a style autoencoder and unit aligner. Traditional models, while incorporating self-supervised learning (SSL) representations that contain both linguistic and paralinguistic information, have neglected this dual nature, leading to reduced controllability. Addressing this issue, we implement cross-attention to synchronize these representations with various emotions. Additionally, a style autoencoder is developed for the disentanglement and manipulation of style elements. The efficacy of our approach is validated through both subjective and objective evaluations, establishing its superiority over existing models in the field.
In this paper, we introduce DiarizationLM, a framework to leverage large language models (LLM) to post-process the outputs from a speaker diarization system. Various goals can be achieved with the proposed framework, such as improving the readability of the diarized transcript, or reducing the word diarization error rate (WDER). In this framework, the outputs of the automatic speech recognition (ASR) and speaker diarization systems are represented as a compact textual format, which is included in the prompt to an optionally finetuned LLM. The outputs of the LLM can be used as the refined diarization results with the desired enhancement. As a post-processing step, this framework can be easily applied to any off-the-shelf ASR and speaker diarization systems without retraining existing components. Our experiments show that a finetuned PaLM 2-S model can reduce the WDER by rel. 55.5% on the Fisher telephone conversation dataset, and rel. 44.9% on the Callhome English dataset.
Speech dereverberation aims to alleviate the detrimental effects of late-reverberant components. While the weighted prediction error (WPE) method has shown superior performance in dereverberation, there is still room for further improvement in terms of performance and robustness in complex and noisy environments. Recent research has highlighted the effectiveness of integrating physics-based and data-driven methods, enhancing the performance of various signal processing tasks while maintaining interpretability. Motivated by these advancements, this paper presents a novel dereverberation frame-work, which incorporates data-driven methods for capturing speech priors within the WPE framework. The plug-and-play strategy (PnP), specifically the regularization by denoising (RED) strategy, is utilized to incorporate speech prior information learnt from data during the optimization problem solving iterations. Experimental results validate the effectiveness of the proposed approach.
As a fundamental tool for natural language processing (NLP), the part-of-speech (POS) tagger assigns the POS label to each word in a sentence. A novel lightweight POS tagger based on word embeddings is proposed and named GWPT (green word-embedding-based POS tagger) in this work. Following the green learning (GL) methodology, GWPT contains three modules in cascade: 1) representation learning, 2) feature learning, and 3) decision learning modules. The main novelty of GWPT lies in representation learning. It uses non-contextual or contextual word embeddings, partitions embedding dimension indices into low-, medium-, and high-frequency sets, and represents them with different N-grams. It is shown by experimental results that GWPT offers state-of-the-art accuracies with fewer model parameters and significantly lower computational complexity in both training and inference as compared with deep-learning-based methods.