Speech and Music Quality: Perception and Prediction

语音质量评价是听力学和言语语言病理学的重要研究课题。在听力学背景下，语音质量是用户对助听器(HAS)和辅助听力设备接受和满意的一个重要因素。因此，测量听力设备中的信号处理算法对语音质量的影响是至关重要的。在SLP领域，语音质量测量用于评估嗓音障碍的严重程度，并监测嗓音治疗过程中的进展。我们的跨学科研究将工程专业知识应用于这两个领域，总体目标是开发和评估语音质量评估的新临床工具。语音质量可以通过主观或客观的工具手段进行评估。在主观语音质量评估中，招募了正常或听力受损的听者，并记录他们对语音质量的感知。虽然主观评价语音质量具有很高的表面效度，但它们也是耗时和资源密集型的。因此，使用听力计算模型从语音信号中获得的工具度量是有吸引力和可取的。我们建议的研究计划建立在我们在客观语音质量建模方面的专业知识基础上，并涵盖了主观和客观语音质量评估的询问路线。在过去的几年中，我们开发了用于紧凑表示语音质量的新特征，并开发了几个机器学习模型来将这些特征向量映射到语音质量分数。我们计划在未来五年的工作基础上，研究更有效的特征提取和特征映射方案，并纳入新的听力受损模型。特别是，在听力学主题中，我们的目标是开发和评估计算听力和机器学习模型，以预测听力受损的听众的双耳音质和音乐质量感知。以类似的方式，我们计划开发和评估总体上语音质量感知障碍的机器学习模型，但更具体地说，是帕金森氏症患者或喉部切除患者对语音的感知。我们在听力学领域的拟议工作将导致开发一种用于语音和音乐质量评估的临床工具，听觉学家可以使用该工具独立验证制造商关于助听器好处的声明。特别是，我们提出的新的双耳音质指标将允许听觉学家对新兴的双耳无线助听器进行基准测试。同样，我们提出的预测紊乱语音质量评级的工作将导致临床工具的出现，这些工具将提高对异常语音质量的评估、处方治疗方案和监测治疗过程进展的效率。

Speech quality assessment is an important research topic in Audiology and Speech Language Pathology (SLP). In the Audiology context, speech quality is an important factor in users' acceptance of and satisfaction with hearing aids (HAs) and assistive listening devices. As such, it is crucial to measure the impact of signal processing algorithms in hearing devices on speech quality. In the realm of SLP, speech quality measures are used to assess the severity of voice disorder and to monitor the progress over the course of voice therapy. Our interdisciplinary research applies engineering expertise to these two fields with the overarching goal of developing and evaluating novel clinical tools for speech quality assessment. Speech quality can be assessed through subjective or objective, instrumental means. In subjective speech quality assessment, normal or hearing impaired listeners are recruited and their perception of speech quality is recorded. While subjective opinions of speech quality have high degree of face validity, they are also time--consuming and resource intensive. Therefore, instrumental metrics derived from speech signals using computational models of audition are attractive and desirable. Our proposed research program builds on our expertise in objective speech quality modelling and encompasses lines of inquiry in both subjective and objective speech quality assessment. During the past few years, we have developed novel features for compact representation of speech quality, and several machine learning models to map these feature vectors into speech quality scores. We plan to build upon our work during the next five years by investigating more effective feature extraction and feature mapping schemes, and by incorporating new models of impaired hearing. In particular, within the Audiology theme, our goal is to develop and evaluate computational hearing and machine learning models for predicting the binaural sound quality and music quality perception by hearing impaired listeners. In a similar manner, we plan to develop and evaluate machine learning models for disordered speech quality perception in general, but more specifically the perception of speech produced by subjects with Parkinson's disease or subjects who have undergone laryngectomy. Our proposed work in the Audiology area will lead to the development of a clinical tool for speech and music quality assessment that Audiologists can use to independently verify manufacturers' claims about the benefits of their hearing aids. In particular, our proposed novel binaural metrics of sound quality will allow Audiologists to benchmark emerging binaural wireless hearing aids. Similarly, our proposed work on prediction of disordered speech quality ratings will lead to clinical tools that will increase efficiency in the assessment of abnormal voice quality, in prescribing a therapy protocol, and in monitoring the progress over the course of therapy.