Residual coding method of linear prediction coding coefficient based on collaborative quantization, and computing device for performing the method
Disclosed are a method for coding a residual signal of LPC coefficients based on collaborative quantization and a computing device for performing the method. The residual signal coding method includes: generating encoded LPC coefficients and LPC residual signals by performing LPC analysis and quantization on an input speech; Determining a predicted LPC residual signal by applying the LPC residual signal to cross module residual learning; Performing LPC synthesis using the coded LPC coefficients and the predicted LPC residual signal; It may include the step of determining an output speech that is a synthesized output according to a result of performing the LPC synthesis.
1. A residual signal coding method of LPC (Linear Prediction Coding) coefficients performed by a computing device, the residual signal coding method comprising:
generating coded LPC coefficients and LPC residual signals by performing, by a computing device, LPC analysis and quantization on an input speech;
determining a predicted LPC residual signal by applying the LPC residual signal to cross module residual learning;
performing LPC synthesis using the coded LPC coefficients and the predicted LPC residual signal;
determining an output speech that is a synthesized output according to the result of performing the LPC synthesis.
2. The residual signal coding method of claim 1 , wherein the cross module residual learning including:
applying a high-pass filter to the input speech;
applying a pre-emphasis filter to a result applied by the high pass filter;
determining the LPC coefficient from the result of applying the pre-emphasis filter;
generating a soft assignment matrix of the coded LPC coefficients and softmax by quantizing the LPC coefficients; and
determining an LPC residual signal based on a result of applying the pre-emphasis filter and a result of quantizing the LPC coefficients.
3. The residual signal coding method of claim 2 , wherein the determining the LPC coefficient comprises:
performing cross-frame windowing by applying a window to all frames of the input speech to which the pre-emphasis filter has been applied;
performing sub-frame windowing by applying a window to a plurality of sub-frames corresponding to an intermediate region among all frames of the input speech from the cross-frame windowing result;
performing synthesis windowing by overlapping the sub-frame windowing result.
4. The residual signal coding method of claim 1 , wherein the LPC coefficient may be quantized by applying a trainable softmax to the LPC coefficient of a LSP (line spectrum pair) domain.
5. The residual signal coding method of claim 1 , wherein the LPC residual signal is encoded by autoencoders of 1D-CNN (convolution neural network).
6. The residual signal coding method of claim 5 , wherein the autoencoders of the 1D-CNN are trained sequentially by using a residual signal that is the output of a previous autoencoder, as the input of a following autoencoder.
7. The residual signal coding method of claim 5 ,
wherein a differential coding is applied to the output of the autoencoder with respect to the autoencoders of the 1D-CNN,
wherein the output of the autoencoder is applied with differential coding based on a length of the code for each frame of the autoencoder.
8. A computing device for performing a residual signal coding method of LPC coefficients, the computing device comprising: one or more processors configured to:
generate encoded LPC coefficients and LPC residual signals by performing, by a computing device, LPC (Linear Prediction Coding) analysis and quantization on the input speech;
determine a predicted LPC residual signal by applying the LPC residual signal to cross module residual learning;
perform LPC synthesis using the coded LPC coefficients and the predicted LPC residual signal;
determine an output speech that is a synthesized output according to the result of performing the LPC synthesis.
9. The computing device of claim 8 ,
wherein the one or more processors perform:
applying a high-pass filter to the input speech;
applying a pre-emphasis filter to a result applied by the high pass filter,
determining the LPC coefficient from the result of applying the pre-emphasis filter;
generating a soft assignment matrix of the coded LPC coefficients and softmax by quantizing the LPC coefficients; and
determining an LPC residual signal based on a result of applying the pre-emphasis filter and a result of quantizing the LPC coefficients.
10. The computing device of claim 9 ,
wherein the one or more processors for determining the LPC coefficient performs:
cross-frame windowing by applying a window to all frames of the input speech to which the pre-emphasis filter has been applied;
sub-frame windowing by applying a window to a plurality of sub-frames corresponding to an intermediate region among all frames of the input speech from the cross-frame windowing result;
synthesis windowing by overlapping the sub-frame windowing result.
11. The computing device of claim 8 , wherein the LPC coefficient may be quantized by applying a trainable softmax to the LPC coefficient of a LSP domain.
12. The computing device of claim 8 , wherein the LPC residual signal is encoded by 1D-CNN autoencoders.
13. The computing device of claim 12 , wherein the autoencoders of the 1D-CNN are trained sequentially by using a residual signal that is the output of a previous autoencoder, as the input of a following autoencoder.
14. The computing device of claim 12 ,
wherein a differential coding is applied to the output of the autoencoder with respect to the autoencoders of the 1D-CNN,
wherein the output of the autoencoder is applied with differential coding based on a length of the code for each frame of the autoencoder.