SRCodec: Split-residual vector quantization for neural speech codec

We present samples for SRCodec, our proposed framwork. Read our paper for more details.

We provide samples including four females and four males, randomly selected from our test set. We compare our proposed method with both traditional codecs and neural codecs, including Opus at 6 kbps and 9 kbps, Encodec at 3 kbps, Lyra-v1 at 3 kbps, and Lyra-v2 at 3.2 kbps, 6 kbps . The samples of Encodec are reported without Entropy Coding. All utterance signals are resampled to 16 kHz.

Thanks for your listening!