23 March 2018 in IEEE Transactions on Emerging Topics in Computational Intelligence
Speech enhancement (SE) aims to reduce noise in speech signals. Most SE techniques focus only on addressing audio information. In this paper, inspired by multimodal learning, which utilizes data from different modalities, and the recent success of convolutional neural networks (CNNs) in SE, we propose an audio-visual deep CNNs (AVDCNN) SE model, which incorporates audio and visual streams into a unified network model. We also propose a multitask learning framework for reconstructing audio and visual signals at the output layer. Precisely speaking, the proposed AVDCNN model is structured as an audio-visual encoder-decoder network, in which audio and visual data are first processed using individual CNNs, and then fused into a joint network to generate enhanced speech (the primary task) and reconstructed images (the secondary task) at the output layer. The model is trained in an endto-end manner, and parameters are jointly learned through back propagation. We evaluate enhanced speech using five instrumental criteria. Results show that the AVDCNN model yields a notably superior performance compared with an audio-only CNN-based SE model and two conventional SE approaches, confirming the effectiveness of integrating visual information into the SE process. In addition, the AVDCNN model also outperforms an existing audio- visual SE model, confirming its capability of effectively combining audio and visual information in SE.
1 September 2020 in Neurophysiologie Clinique
24 November 2023 in Mayo Clinic Proceedings: Digital Health
To investigate the feasibility and accuracy of artificial intelligence (AI) methods of facial deidentification in hospital-recorded epileptic seizure videos, for improved patient privacy protection while preserving clinically important features of seizure semiology. Videos of epileptic seizures displaying seizure-related involuntary facial changes were selected from recordings at Taipei Veterans General Hospital Epilepsy Unit (between August 1, 2020 and February 28, 2023), and a single representative video frame was prepared per seizure. We tested 3 AI transformation models: (1) morphing the original facial image with a different male face; (2) substitution with a female face; and (3) cartoonization. Facial deidentification and preservation of clinically relevant facial detail were calculated based on: (1) scoring by 5 independent expert clinicians and (2) objective computation. According to the clinician scoring of 26 facial frames in 16 patients, the best compromise between deidentification and preservation of facial semiology was the cartoonization model. A male facial morphing model was superior to the cartoonization model for deidentification, but clinical detail was sacrificed. Objective similarity testing of video data reported deidentification scores in agreement with the clinicians’ scores; however, preservation of semiology gave mixed results likely due to inadequate existing comparative databases. Artificial intelligence-based face transformation of medical seizure videos is feasible and may be useful for patient privacy protection. In our study, the cartoonization approach provided the best compromise between deidentification and preservation of seizure semiology.
1 April 2026 in IEEE Transactions on Biomedical Engineering
Objective: Individuals with hearing impairments face challenges in their ability to comprehend speech, particularly in noisy environments. This study explores the effectiveness of audio-visual speech enhancement (AVSE) in improving the intelligibility of vocoded speech in cochlear implant (CI) simulations. Methods: We propose a speech enhancement framework called Self-Supervised Learning-based AVSE (SSL-AVSE), which uses visual cues such as lip and mouth movements along with corresponding speech. Features are extracted using the AV-HuBERT model and refined through a bidirectional LSTM. Experiments were conducted using the Taiwan Mandarin speech with video (TMSV) dataset. Results: Objective evaluations showed improvements in PESQ from 1.43 to 1.67 and in STOI from 0.70 to 0.74. NCM scores increased by up to 87.2% over the noisy baseline. Subjective listening tests further demonstrated maximum gains of 45.2% in speech quality and 51.9% in word intelligibility. Conclusion: SSL-AVSE consistently outperforms audio-only speech enhancement (AOSE) and conventional AVSE baselines. Listening tests with statistically significant results confirm its effectiveness. In addition to its strong performance, SSL-AVSE demonstrates cross-lingual generalization: although it was pretrained on English data, it performs effectively on Mandarin speech. This finding highlights the robustness of the features extracted by a pretrained foundation model and their applicability across languages. Significance: To the best of our knowledge, no prior work has explored the application of AVSE to CI simulations. This study provides the first evidence that incorporating visual information can significantly improve the intelligibility of vocoded speech in CI scenarios.