AECodec: High Fidelity Neural Audio Codec Based On Speech And Electroglottograph

With the rapid development of communication devices and voice interactions, speech is playing an increasingly important role. Given the growing demand for voice transmission, efficiently transmitting high-quality speech signals within limited bandwidth becomes a challenge. This paper introduces a high-fidelity speech codec that utilizes both speech signals and electroglottograph signals for speech compression. The codec is capable of compressing and decompressing speech, significantly reducing the bitrate while maintaining high fidelity. The proposed codec employs a streaming encoder-decoder architecture and is trained in an end-to-end manner. In addition to using speech signals as input, the electroglottograph signal is used as an auxiliary input, leveraging its ability to capture vocal fold movement characteristics such as closure degree, closure speed, and cycle duration, thus enhancing the model’s feature extraction capability. Moreover, the encoder-decoder structure integrates a Transformer Encoder module with residual connections, further improving the model's ability to process time series data. To validate the effectiveness of this approach, we conducted extensive objective evaluations and experimental studies across various bandwidths, proving our approach is superior to the baselines methods.