Signal processing for multichannel audio

多通道音频信号处理

• 批准号: 2466018
• 金额: --
• 依托单位: Imperial College London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2466018
• 关键词: Signal; processing; multichannel; audio

The PhD research will focus on algorithms to improve signals in audio systems where there are more than one channel, otherwise known as multichannel audio systems. Multichannel systems are in common use as they are vital to many enhancement algorithms for example removing background noise from conferencing calls. One problem that is being addressed is finding ways of reducing the amount of information that is needed to represent signals in multichannel systems. This could be useful in situations where there is a limited bandwidth to transmit many signals, where without this technology multichannel audio systems could not be used. Another problem is removing the effect of reverberation from microphone signals. De-reverberation reduces the performance of voice recognition software such as is commonly found in home assistants, therefore effective algorithms that remove the reverberation before the signal is sent to a voice recognition system are desirable. Another problem that may be addressed is the interpretation of machine learning methods. Algorithms that rely on machine learning are highly effective at tasks that are otherwise very difficult. However, there is a lack of interpretability, that is to say, an output is generated for a given input but it is very difficult for experts to say what is actually happening in machine learning algorithms. Some work may be conducted in finding ways of interpreting these systems when they are applied to signal processing. This would be beneficial in many cases for example it would allow researchers to understand and design better algorithms, and also means machine learning could be applied to safety critical applications with a greater understanding of the risks. The main approach to conducting the research will be to use programming languages to simulate situations such as rooms etc. and implement the algorithms in simulations. This is the most widespread technique used in the field as it is often the simplest and most effective way of designing and assessing these algorithms.The research could be categorised under with the following research areas: Digital Signal Processing Music and Acoustics TechnologyArtificial intelligence Technologies Speech TechnologyNatural Language Processing

博士研究将专注于算法，以改善音频系统中的信号，其中有一个以上的通道，也称为多通道音频系统。多通道系统是常用的，因为它们对于许多增强算法是至关重要的，例如从会议呼叫中去除背景噪声。正在解决的一个问题是找到减少多通道系统中表示信号所需的信息量的方法。这在传输许多信号的带宽有限的情况下可能很有用，如果没有这种技术，就无法使用多声道音频系统。另一个问题是从麦克风信号中去除混响的影响。去混响降低了诸如在家庭助理中常见的语音识别软件的性能，因此期望在信号被发送到语音识别系统之前去除混响的有效算法。另一个可以解决的问题是机器学习方法的解释。依赖于机器学习的算法在其他非常困难的任务中非常有效。然而，缺乏可解释性，也就是说，对于给定的输入生成输出，但专家很难说出机器学习算法中实际发生的事情。当这些系统应用于信号处理时，可以进行一些工作来寻找解释这些系统的方法。在许多情况下，这将是有益的，例如，它将使研究人员能够理解和设计更好的算法，也意味着机器学习可以应用于安全关键应用，更好地了解风险。进行研究的主要方法将是使用编程语言来模拟房间等情况，并在模拟中实现算法。这是该领域中使用最广泛的技术，因为它通常是设计和评估这些算法的最简单和最有效的方法。研究可以分为以下研究领域：数字信号处理音乐和声学技术人工智能技术语音技术自然语言处理