25-08-2017, 09:32 PM
Abstract
Due to increasing demand for speech communications, efficient techniques in low-rate speech coding are of interest.Here a new compression technique using root cepstral analysis has been proposed. Implementing the proposed method causes the coder to deal with root cepstrum coefficients instead of speech samples. The main idea in using root cepstrum analysis is that some of the trivial coefficients can be ignored to send toward the decoder. Thus considerable reduction in the total bit-rate is achieved. Moreover it has the advantage of adjustability which can be used to optimize the coding procedure. The experimental results confirm the ability of the proposed method in speech coding problems
Despite the fact, cepstrum has been widely used in speech recognition, this paper intend to apply cepstrum in speech coding problems; the proposed algorithm aims to put cepstrum in a CELP coder to obtain a low-rate coder. Here the main purpose is to demonstrate and also compare the strength of Root rather than log cepstrum in low-rate coders.
As description on the proposed method, a pre-processing step i.e. framing, windowing, etc, is applied to the input speech at first. Then the cepstral coefficients for each frame are computed. In the next step, in order to reduce the bit rate for transmitting the samples (cepstral coefficients), the frame length has to be decreased. We can do this by zeroing out some samples (cepstrum coeffs.) of each frame. The main idea of applying root cepstrum analysis and then zero out trivial coefficients is: cepstral analysis separates the vocal tract and pitch information. Data related to the vocal tract is located in low qufrequencies, and the pitch information in upper part of qufrequencies axis. Hence the samples located in the central zone of qu-frequency domain carry insignificant speech information. This can be found as a suitable feature for reducing the bit-rates in speech coders.
If you want more info on this please leave a post here.
Due to increasing demand for speech communications, efficient techniques in low-rate speech coding are of interest.Here a new compression technique using root cepstral analysis has been proposed. Implementing the proposed method causes the coder to deal with root cepstrum coefficients instead of speech samples. The main idea in using root cepstrum analysis is that some of the trivial coefficients can be ignored to send toward the decoder. Thus considerable reduction in the total bit-rate is achieved. Moreover it has the advantage of adjustability which can be used to optimize the coding procedure. The experimental results confirm the ability of the proposed method in speech coding problems
Despite the fact, cepstrum has been widely used in speech recognition, this paper intend to apply cepstrum in speech coding problems; the proposed algorithm aims to put cepstrum in a CELP coder to obtain a low-rate coder. Here the main purpose is to demonstrate and also compare the strength of Root rather than log cepstrum in low-rate coders.
As description on the proposed method, a pre-processing step i.e. framing, windowing, etc, is applied to the input speech at first. Then the cepstral coefficients for each frame are computed. In the next step, in order to reduce the bit rate for transmitting the samples (cepstral coefficients), the frame length has to be decreased. We can do this by zeroing out some samples (cepstrum coeffs.) of each frame. The main idea of applying root cepstrum analysis and then zero out trivial coefficients is: cepstral analysis separates the vocal tract and pitch information. Data related to the vocal tract is located in low qufrequencies, and the pitch information in upper part of qufrequencies axis. Hence the samples located in the central zone of qu-frequency domain carry insignificant speech information. This can be found as a suitable feature for reducing the bit-rates in speech coders.
If you want more info on this please leave a post here.