Abstract :
This paper proposes new waveform reconstruction techniques derived from wavelet transform and multilayer perceptron neural networks to reconstruct the speech segments that were lost in a transmission through best effort networks, such as Internet based on IP protocol. Also, a voiced-unvoiced-silence (V-U-S) recognition technique based on mutilayer perceptron neural networks is presented. The use of this recognition technique enhances the performance of the proposed reconstruction since proper configurations of them are applied to each type of speech segments. The reported results reveal that the proposed reconstruction techniques outperform the previous one when the G.722.1 standard from ITU-T is considered as a speech coding techs paper proposes new waveform reconstruction techniques derived from wavelet transform and multilayer perceptron neural networks to reconstruct the speech segments that were lost in a transmission through best effort networks, such as Internet based on IP protocol. Also, a voiced-unvoiced-silence (V-U-S) recognition technique based on mutilayer perceptron neural networks is presented. The use of this recognition technique enhances the performance of the proposed reconstruction since proper configurations of them are applied to each type of speech segments. The reported results reveal that the proposed reconstruction techniques outperform the previous one when the G.722.1 standard from ITU-T is considered as a speech coding technique of a VoIP solution. This paper proposes new waveform reconstruction techniques derived from wavelet transform and multilayer perceptron neural networks to reconstruct the speech segments that were lost in a transmission through best effort networks, such as Internet based on IP protocol. Also, a voiced-unvoiced-silence (V-U-S) recognition technique based on mutilayer perceptron neural networks is presented. The use of this recognition technique enhances the performance of the proposed reconstruction since proper- - configurations of them are applied to each type of speech segments. The reported results reveal that the proposed reconstruction techniques outperform the previous one when the G.722.1 standard from ITU-T is considered as a speech coding technique of a VoIP solution.nique of a VoIP solution.
