Please do not reprint this article in any form, thank you.
5.2 Silk decoding process
The linear prediction layer of the decoder mainly uses the long-short time prediction synthesis filter to filter the excitation signal. The internal working bandwidth of the linear prediction layer is NB, MB and WB. For the mixed coding mode of SWB and FB, the linear prediction layer still works in WB Under the bandwidth, after interval decoding, the symbols of interval encoding are decoded, and several parameters of SILK/CELT encoding are obtained by analyzing these symbols. For speech frames, the content of SILK encoding includes parameters and excitation, and the index refers to the index of the codebook after parameter vectorization , the index of the codebook will eventually be mapped to encoding parameters, such as LTP, LSF parameters, etc. The process of SILK decoding is as follows, after decoding by the interval decoder, the source encoding information is obtained, and then generated by LTP and LPC filtering according to the decoded channel parameters The decoded voice signal and the generated voice signal are resampled according to the set number of channels and the API Fs sampling value to output the final audio signal. This process is shown in Figure 5-4.
Figure 5-4 SILK decoding frame
The next section analyzes the decoding process of Opus ordinary frame SILK in combination with the meaning of encoding symbols and decoding functions. The function call relationship is shown in Figure 5-5.
Figure 5-5 Decoding function call flow of Opus common frame SILK
5.2.1 LP layer
The smallest unit of Opus encoding is the encoding frame. The duration of an encoding frame is 10ms or 20ms. An Opus encoding packet can include multiple encoding frames, but an encoding packet can contain up to 120ms of speech encoding information. Opus may also contain Low Bit-