Implementation of Multichannel Speech Coding Based on the Opus Codec and Spatial Parameters

Multichannel speech processing has been widely studied since the spatial information contained in the multichannel signals can be exploited. To facilitate the efficient transmission and preserve the spatial information, the multichannel speech coding technique is needed. Recently, a multichannel speech coding method based on the Opus codec and spatial parameters was proposed. In the encoding stage, the speech signal of the reference channel is encoded with the Opus codec. The multichannel speech signals are decomposed through the Gammatone filter bank and the spatial parameters are extracted and quantized for each sub-band signals. In the decoding stage, the encoded signal of the reference channel is decoded with the Opus codec. Subsequently, this decoded signal is combined with the quantized spatial parameters to recover the speech signals of the remaining channels. In this paper, an improved implementation of this coding method is detailed. Specifically, a framing pattern more suitable for multichannel speech coding with the Gammatone filter bank is proposed. Besides, a newly designed window is then employed on each sub-band signal for the precise extraction of spatial parameters in the frequency domain. Additionally, the extracted spatial parameters are quantized non-uniformly. The experimental results show the effectiveness of the proposed implementation. This implementation can achieve high speech quality with a reduced bitrate. Furthermore, the spatial information contained in the multichannel speech signals can be better preserved.