Applied Speech and Audio Processing: With matlab examples
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
4.2. Psychoacoustics
71 distinct echoes. For this reason, echoes are only noticeable in large rooms – they are present in small rooms but we are unable to hear them. Matlab can be used to construct a demo of the Haas effect. First record a word or two of clear speech into Matlab, then use the following code to listen to the speech replayed with progressively longer unattenuated echoes. Fs is set to the sample rate used to record the speech, and the echo starts at 10 ms, then increases in steps of 20 ms up to 100 ms: audio=reshape(audio,1,length(audio)); for echo=0.01:0.020:0.1 pad=zeros(1,fix(Fs*echo)); input(’Press any key to hear next echo’); soundsc([audio,pad]+[pad,audio],Fs); end Note that the reshape() function is used to ensure that the audio vector is (1 × N) rather than (N × 1) in dimension, so it can be combined with the variable sized padding. One useful rule of thumb when working with speech is that echoes of 100 ms or longer in two-way conversational speech, such as during a telephone call, will become annoying and distracting to callers. This is a major quality factor for telephony systems, and presents a hard-to-meet target for IP-based Internet telephony systems such as Skype which use digital processing and relatively slow or congested communications networks. 4.2.16 Speech perception Evidence exists to indicate that the human aural system processes speech in a completely different way from other non-similar sounds [4]. One experiment which uses sinewave speech to demonstrate this, was discussed in the introduction to Chapter 3. Whatever the exact cause, certain brain mechanisms cause a listener to perceive two sentences of speech as being similar when the physical differences between them may be very large (for example, same speaker, same words, different acoustic situation). Two examples of one speaker saying the same words in two different environments would be judged by listeners as similar, irrespective of possibly major physical differences in amplitude, timing, pitch, and so on. Speech can be detected and understood when the noise-to-signal power is such that single tones would be inaudible. Despite the evidence suggesting that speech processing in the brain is separate from sound processing, speech perception still suffers from simultaneous and non- simultaneous masking, binaural masking and auditory adaptation (indeed this may be more pronounced: monotonic speech rapidly leads to drowsiness or lack of interest). Speech perception interlinks speech production with the psychological aspects of the communication channel, the subject matter and the state of mind of the recipient [3]. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling