TTSlow: Slow Down Text-to-Speech with Efficiency Robustness Evaluations

Text-to-speech (TTS) has been extensively studied for generating high-quality speech with textual inputs, playing a crucial role in various real-time applications. For real-world deployment, ensuring stable and timely generation in TTS models against minor input perturbations is of paramount importance. Therefore, evaluating the robustness of TTS models against such perturbations, commonly known as adversarial attacks, is highly desirable. In this paper, we propose TTSlow, a novel adversarial approach specifically tailored to slow down the speech generation process in TTS systems. To induce long TTS waiting time, we design novel efficiency-oriented adversarial loss to encourage endless generation process. TTSlow encompasses two attack strategies targeting both text inputs and speaker embedding. Specifically, we propose TTSlow-text, which utilizes a combination of homoglyphs-based and swap-based perturbations, along with TTSlow-spk, which employs a gradient optimization attack approach for speaker embedding. TTSlow serves as the first attack approach targeting a wide range of TTS models, including autoregressive and non-autoregressive TTS ones, thereby advancing exploration in audio security. Extensive experiments are conducted to evaluate the inference efficiency of TTS models, and in-depth analysis of generated speech intelligibility is performed using Gemini. The results demonstrate that TTSlow can effectively slow down two TTS models across three publicly available datasets. We are committed to releasing the source code upon acceptance, facilitating further research and benchmarking in this domain.

Unsupervised Face-Mask Speech Enhancement Using Generative Adversarial Networks with Human-in-the-Loop Assessment Metrics

The utilization of face masks is an essential healthcare measure, particularly during times of pandemics, yet it can present challenges in communication in our daily lives. To address this problem, we propose a novel approach known as the human-in-the-loop StarGAN (HL-StarGAN) face-masked speech enhancement method. HL-StarGAN comprises discriminator, classifier, metric assessment predictor, and generator that leverages an attention mechanism. The metric assessment predictor, referred to as MaskQSS, incorporates human participants in its development and serves as a "human-in-the-loop" module during the learning process of HL-StarGAN. The overall HL-StarGAN model was trained using an unsupervised learning strategy that simultaneously focuses on the reconstruction of the original clean speech and the optimization of human perception. To implement HL-StarGAN, we curated a face-masked speech database named "FMVD," which comprises recordings from 34 speakers in three distinct face-masked scenarios and a clean condition. We conducted subjective and objective tests on the proposed HL-StarGAN using this database. The outcomes of the test results are as follows: (1) MaskQSS successfully predicted the quality scores of face mask voices, outperforming several existing speech assessment methods. (2) The integration of the MaskQSS predictor enhanced the ability of HL-StarGAN to transform face mask voices into high-quality speech; this enhancement is evident in both objective and subjective tests, outperforming conventional StarGAN and CycleGAN-based systems.

Bridging the Gap: Integrating Pre-trained Speech Enhancement and Recognition Models for Robust Speech Recognition

Noise robustness is critical when applying automatic speech recognition (ASR) in real-world scenarios. One solution involves the used of speech enhancement (SE) models as the front end of ASR. However, neural network-based (NN-based) SE often introduces artifacts into the enhanced signals and harms ASR performance, particularly when SE and ASR are independently trained. Therefore, this study introduces a simple yet effective SE post-processing technique to address the gap between various pre-trained SE and ASR models. A bridge module, which is a lightweight NN, is proposed to evaluate the signal-level information of the speech signal. Subsequently, using the signal-level information, the observation addition technique is applied to effectively reduce the shortcomings of SE. The experimental results demonstrate the success of our method in integrating diverse pre-trained SE and ASR models, considerably boosting the ASR robustness. Crucially, no prior knowledge of the ASR or speech contents is required during the training or inference stages. Moreover, the effectiveness of this approach extends to different datasets without necessitating the fine-tuning of the bridge module, ensuring efficiency and improved generalization.

SVSNet+: Enhancing Speaker Voice Similarity Assessment Models with Representations from Speech Foundation Models

Representations from pre-trained speech foundation models (SFMs) have shown impressive performance in many downstream tasks. However, the potential benefits of incorporating pre-trained SFM representations into speaker voice similarity assessment have not been thoroughly investigated. In this paper, we propose SVSNet+, a model that integrates pre-trained SFM representations to improve performance in assessing speaker voice similarity. Experimental results on the Voice Conversion Challenge 2018 and 2020 datasets show that SVSNet+ incorporating WavLM representations shows significant improvements compared to baseline models. In addition, while fine-tuning WavLM with a small dataset of the downstream task does not improve performance, using the same dataset to learn a weighted-sum representation of WavLM can substantially improve performance. Furthermore, when WavLM is replaced by other SFMs, SVSNet+ still outperforms the baseline models and exhibits strong generalization ability.

Abnormal Respiratory Sound Identification Using Audio-Spectrogram Vision Transformer

Respiratory disease, the third leading cause of deaths globally, is considered a high-priority ailment requiring significant research on identification and treatment. Stethoscope-recorded lung sounds and artificial intelligence-powered devices have been used to identify lung disorders and aid specialists in making accurate diagnoses. In this study, audio-spectrogram vision transformer (AS-ViT), a new approach for identifying abnormal respiration sounds, was developed. The sounds of the lungs are converted into visual representations called spectrograms using a technique called short-time Fourier transform (STFT). These images are then analyzed using a model called vision transformer to identify different types of respiratory sounds. The classification was carried out using the ICBHI 2017 database, which includes various types of lung sounds with different frequencies, noise levels, and backgrounds. The proposed AS-ViT method was evaluated using three metrics and achieved 79.1% and 59.8% for 60:40 split ratio and 86.4% and 69.3% for 80:20 split ratio in terms of unweighted average recall and overall scores respectively for respiratory sound detection, surpassing previous state-of-the-art results.

An Investigation of Incorporating Mamba for Speech Enhancement

This work aims to study a scalable state-space model (SSM), Mamba, for the speech enhancement (SE) task. We exploit a Mamba-based regression model to characterize speech signals and build an SE system upon Mamba, termed SEMamba. We explore the properties of Mamba by integrating it as the core model in both basic and advanced SE systems, along with utilizing signal-level distances as well as metric-oriented loss functions. SEMamba demonstrates promising results and attains a PESQ score of 3.55 on the VoiceBank-DEMAND dataset. When combined with the perceptual contrast stretching technique, the proposed SEMamba yields a new state-of-the-art PESQ score of 3.69.

Unmasking Illusions: Understanding Human Perception of Audiovisual Deepfakes

The emergence of contemporary deepfakes has attracted significant attention in machine learning research, as artificial intelligence (AI) generated synthetic media increases the incidence of misinterpretation and is difficult to distinguish from genuine content. Currently, machine learning techniques have been extensively studied for automatically detecting deepfakes. However, human perception has been less explored. Malicious deepfakes could ultimately cause public and social problems. Can we humans correctly perceive the authenticity of the content of the videos we watch? The answer is obviously uncertain; therefore, this paper aims to evaluate the human ability to discern deepfake videos through a subjective study. We present our findings by comparing human observers to five state-ofthe-art audiovisual deepfake detection models. To this end, we used gamification concepts to provide 110 participants (55 native English speakers and 55 non-native English speakers) with a webbased platform where they could access a series of 40 videos (20 real and 20 fake) to determine their authenticity. Each participant performed the experiment twice with the same 40 videos in different random orders. The videos are manually selected from the FakeAVCeleb dataset. We found that all AI models performed better than humans when evaluated on the same 40 videos. The study also reveals that while deception is not impossible, humans tend to overestimate their detection capabilities. Our experimental results may help benchmark human versus machine performance, advance forensics analysis, and enable adaptive countermeasures.

Towards Environmental Preference Based Speech Enhancement For Individualised Multi-Modal Hearing Aids

Since the advent of Deep Learning (DL), Speech Enhancement (SE) models have performed well under a variety of noise conditions. However, such systems may still introduce sonic artefacts, sound unnatural, and restrict the ability for a user to hear ambient sound which may be of importance. Hearing Aid (HA) users may wish to customise their SE systems to suit their personal preferences and day-to-day lifestyle. In this paper, we introduce a preference learning based SE (PLSE) model for future multi-modal HAs that can contextually exploit audio information to improve listening comfort, based upon the preferences of the user. The proposed system estimates the Signal-to-noise ratio (SNR) as a basic objective speech quality measure which quantifies the relative amount of background noise present in speech, and directly correlates to the intelligibility of the signal. Additionally, to provide contextual information we predict the acoustic scene in which the user is situated. These tasks are achieved via a multi-task DL model, which surpasses the performance of inferring the acoustic scene or SNR separately, by jointly leveraging a shared encoded feature space. These environmental inferences are exploited in a preference elicitation framework, which linearly learns a set of predictive functions to determine the target SNR of an AV (Audio-Visual) SE system. By greatly reducing noise in challenging listening conditions, and by novelly scaling the output of the SE model, we are able to provide HA users with contextually individualised SE. Preliminary results suggest an improvement over the non-individualised baseline model in some participants.

Self-Supervised Speech Quality Estimation and Enhancement Using Only Clean Speech

Speech quality estimation has recently undergone a paradigm shift from human-hearing expert designs to machine-learning models. However, current models rely mainly on supervised learning, which is time-consuming and expensive for label collection. To solve this problem, we propose VQScore, a self-supervised metric for evaluating speech based on the quantization error of a vector-quantized-variational autoencoder (VQ-VAE). The training of VQ-VAE relies on clean speech; hence, large quantization errors can be expected when the speech is distorted. To further improve correlation with real quality scores, domain knowledge of speech processing is incorporated into the model design. We found that the vector quantization mechanism could also be used for self-supervised speech enhancement (SE) model training. To improve the robustness of the encoder for SE, a novel self-distillation mechanism combined with adversarial training is introduced. In summary, the proposed speech quality estimation method and enhancement models require only clean speech for training without any label requirements. Experimental results show that the proposed VQScore and enhancement model are competitive with supervised baselines. The code will be released after publication.

Audio-Visual Speech Enhancement in Noisy Environments via Emotion-Based Contextual Cues

In real-world environments, background noise significantly degrades the intelligibility and clarity of human speech. Audio-visual speech enhancement (AVSE) attempts to restore speech quality, but existing methods often fall short, particularly in dynamic noise conditions. This study investigates the inclusion of emotion as a novel contextual cue within AVSE, hypothesizing that incorporating emotional understanding can improve speech enhancement performance. We propose a novel emotion-aware AVSE system that leverages both auditory and visual information. It extracts emotional features from the facial landmarks of the speaker and fuses them with corresponding audio and visual modalities. This enriched data serves as input to a deep UNet-based encoder-decoder network, specifically designed to orchestrate the fusion of multimodal information enhanced with emotion. The network iteratively refines the enhanced speech representation through an encoder-decoder architecture, guided by perceptually-inspired loss functions for joint learning and optimization. We train and evaluate the model on the CMU Multimodal Opinion Sentiment and Emotion Intensity (CMU-MOSEI) dataset, a rich repository of audio-visual recordings with annotated emotions. Our comprehensive evaluation demonstrates the effectiveness of emotion as a contextual cue for AVSE. By integrating emotional features, the proposed system achieves significant improvements in both objective and subjective assessments of speech quality and intelligibility, especially in challenging noise environments. Compared to baseline AVSE and audio-only speech enhancement systems, our approach exhibits a noticeable increase in PESQ and STOI, indicating higher perceptual quality and intelligibility. Large-scale listening tests corroborate these findings, suggesting improved human understanding of enhanced speech.