Processing of Pitch by the Auditory System

听觉系统对音调的处理

• 批准号: EP/D501571/2
• 负责人: Chris Plack
• 金额: --
• 依托单位: University of Manchester
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2008
• 资助国家: 英国
• 起止时间: 2008 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FD501571%2F2
• 关键词: Processing; Pitch; Auditory; System

Many of the sounds we hear in everyday life, such as musical notes and the vowels of speech, have waveforms that repeat overtime. These tones often give us the sensation of a distinct pitch that corresponds to the repetition rate or fundamental frequency (FO) of the tone. Pitch perception is important not only in music, but also for the perception of speech: it helps us tell the difference between a question and a statement, and, because different voices differ in pitch, it helps us to separate the voices of different people speaking at the same time. Because of this, scientists and engineers are interested in programming computers to perform pitch perception automatically. These computer models can be used to help us understand the human auditory system, and to help us develop machines that mimic aspects of human perception, such as pitch tracking devices for converting the sounds of instruments into musical notation, and automatic speech recognition devices. Current computer models are quite successful, but not nearly as good as a human. We are trying to understand more about the pitch mechanisms in human hearing, in order to improve these computer models. Complex tones, such as those in speech and music, are composed of a set of pure tone components, called harmonics. Each harmonic has a sinusoidal waveform that repeats at a frequency that is an integer multiple of FO. The multiplying factor gives the harmonic number. For example, the first six harmonics of a complex tone with an FO of 100 Hz have frequencies of 100, 200, 300, 400, 500 and 600 Hz. When we listen to a complex tone, the cochlea in the inner ear separates out the individual low-numbered harmonics (less than about harmonic number ten). The brain derives pitch mainly from the frequencies of these low-numbered harmonics. Each of the low-numbered harmonics excites a distinct place in the cochlea. In addition, the frequency of each harmonic is represented by a synchronized pattern of neural activity, in that neurons in the auditory nerve will tend to fire (produce electrical impulses) at the same time during each cycle of the sinusoidal waveform of the harmonic. In other words, the information about the harmonic frequencies is represented by a place code (place in cochlea) and by a temporal code (temporal pattern of neural activity).The most popular computer models propose that the pitch mechanism uses only the temporal code, however this is still very much open to question. Our experiments will provide tests of the assertion, using techniques that require human subjects to make comparisons between the sounds that are played to them. One set of experiments will determine if it is essential that the temporal code for a particular frequency be conveyed by the auditory neurons that are connected to the place in the cochlea that normally responds to the same frequency. In other words, does accurate pitch perception depend on a match between the place and temporal codes? Another study will focus on a peculiar phenomenon known as dichotic pitch . This occurs when a sound is presented simultaneously to the two ears, such that the sound in each ear is identical except for a narrow frequency region, in which the sounds in the two ears are independent. Listeners hear a faint pitch corresponding to this region. The independence between the two ears is processed in the brain by an array of neurons. We will test whether the outputs of these detectors are coded using a place code (i.e., which neurons in the array are active). If so, this will mean that the pitch mechanism can exploit this code when extracting pitch, and that the current temporal models will have to be revised. Finally, we will investigate whether the combination of different harmonics from the two ears is automatic, or depends on the listener rapidly switching attention from one ear to the other. This experiment will help determine how the input to the pitch mechanism is controlled.

我们在日常生活中听到的许多声音，如音符和语音的元音，都有随时间重复的波形。这些音调通常给我们一种与音调的重复率或基频（FO）相对应的独特音高的感觉。音高感知不仅在音乐中很重要，而且对言语的感知也很重要：它帮助我们区分问题和陈述，而且，由于不同的声音在音高上不同，它帮助我们区分不同人同时说话的声音。正因为如此，科学家和工程师对计算机编程以自动执行音调感知很感兴趣。这些计算机模型可以帮助我们理解人类的听觉系统，并帮助我们开发模仿人类感知的机器，例如将乐器的声音转换为乐谱的音高跟踪设备，以及自动语音识别设备。目前的计算机模型相当成功，但远不如人类。我们正试图更多地了解人类听觉中的音高机制，以改进这些计算机模型。复杂的音调，如语音和音乐中的音调，是由一组纯音成分组成的，称为谐波。每个谐波具有以FO的整数倍的频率重复的正弦波形。乘法因子给出谐波数。例如，FO为100 Hz的复音调的前六个谐波具有100、200、300、400、500和600 Hz的频率。当我们听到一个复杂的音调时，内耳中的耳蜗会分离出单个的低次谐波（大约小于10次谐波）。大脑主要从这些低频谐波的频率中获得音高。每一个低编号的谐波都会刺激耳蜗中的一个不同的地方。此外，每个谐波的频率由神经活动的同步模式表示，因为在谐波的正弦波形的每个周期期间，听觉神经中的神经元将倾向于同时激发（产生电脉冲）。换句话说，关于谐波频率的信息由位置代码（耳蜗中的位置）和时间代码（神经活动的时间模式）表示。最流行的计算机模型提出音高机制仅使用时间代码，然而这仍然是非常值得商榷的。我们的实验将提供对这一断言的测试，使用的技术要求人类受试者对播放给他们的声音进行比较。一组实验将确定是否有必要由听觉神经元传达特定频率的时间代码，这些听觉神经元连接到耳蜗中通常对相同频率做出反应的位置。换句话说，准确的音高感知是否取决于位置码和时间码之间的匹配？另一项研究将集中在一个特殊的现象称为双耳分音音高。这发生在当声音同时呈现给两只耳朵时，使得每只耳朵中的声音除了窄频率区域之外是相同的，在窄频率区域中两只耳朵中的声音是独立的。听众听到一个微弱的音高对应于这个区域。双耳之间的独立性在大脑中由一系列神经元处理。我们将测试这些检测器的输出是否使用位置代码（即，阵列中的哪些神经元是活跃的）。如果是这样，这将意味着音高机制在提取音高时可以利用此代码，并且必须修改当前的时间模型。最后，我们将研究来自双耳的不同谐波的组合是否是自动的，或者取决于听者迅速将注意力从一只耳朵切换到另一只耳朵。这个实验将有助于确定如何控制变桨机制的输入。