Applied Speech and Audio Processing: With matlab examples
Analysis of other signals
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
6.3. Analysis of other signals
151 These methods have each been tested within a GSM structure for the determination of pitch [8]. The time-frequency distribution algorithms, with a little post-processing, were used as a replacement for the original pitch analysis used for the RPE structure. This overall system was then tested to determine what changes (if any) resulted in the intelligibility of speech conveyed. In each case the same speech and other conditions were used for the tests, and the Chinese diagnostic rhyme test (CDRT) – a Chinese language equivalent of the diagnostic rhyme test (DRT) of Section 3.3.3 – was used to assess intelligibility. Results indicated a significant improvement in intelligibility for the speech that differed in terms of sibilation, and smaller less significant improvements in several other speech classes. Overall it seems that the results are promising; TFD might well become another candidate for pitch determination in speech communications systems. However at present other methods, particularly the autocorrelation based systems, are most popular. 6.3 Analysis of other signals Arguably the predominant application of audio analysis has been related to speech com- munications applications, with a notable (and economically important) extension into music compression, such as MP3. However there is no reason why the techniques, and toolkit introduced in this chapter should not be applied elsewhere. As an example, one of the early algorithms produced by the author was capable of detecting and classifying dog barks, and in 1997 the author was contacted by a team of economists who wished to apply LSP analysis to the stock market. 1 6.3.1 Analysis of music Many researchers have investigated musical instrument recognition and coding. For example, Krishna and Sreenivas [9] evaluated three methods of recognising individual instrument sets. One of these methods was based upon LSPs and was found to be the superior method among those tested. The authors stated several advantages of LSPs including their localised spectral sensitivities, their ability to characterise both resonance locations and bandwidths (characteristics of the timbre of the instruments), and the important aspect of spectral peak location. In order to examine some of these claims, it is easy to use Matlab for analysis. As an example, Matlab was used to record a violin open A string (tuned to 440 Hz), sampled at 16 kHz. Those readers with a violin handy may wish to record a short segment using the methods of Chapter 2. It is important to ensure that the player uses an uninterrupted and smooth bowing action during the recording. Given such a recording in a floating point 1 Sadly nothing was subsequently heard from the economists, from which we can conclude that either the technique did not work, or was so successful that they immediately retired from research before writing up, and did not wish to share their key to unlimited wealth. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling