BigCodec：推进低比特率神经语音编解码器的极限
BigCodec: Pushing the Limits of Low-Bitrate Neural Speech Codec
AI总结：
本文提出 159M 参数的低比特率神经语音编解码器 BigCodec，通过融合序列与卷积架构及低维向量量化，在 1.04 kbps 下取得优于现有方案的主客观表现。
摘要：
我们提出 BigCodec，一种低比特率神经语音编解码器。尽管近期神经语音编解码器已展现出显著进展，但在低比特率（约 1 kbps）下，其性能会明显下降。虽然低比特率本身会限制性能，但模型容量等其他因素也阻碍了进一步提升。为解决这一问题，我们将模型规模扩大到 159M 参数，比参数量约 10M 的流行编解码器大 10 倍以上。此外，我们将序列模型集成到传统卷积架构中，以更好地捕捉时间依赖关系，并采用低维向量量化来确保较高的码本利用率。全面的客观和主观评估表明，比特率为 1.04 kbps 的 BigCodec 显著优于若干现有低比特率编解码器。此外，BigCodec 的客观性能可与运行在高 4 至 6 倍比特率下的流行编解码器相媲美，甚至在主观感知质量上优于真实参考音频。
英文摘要：
We present BigCodec, a low-bitrate neural speech codec. While recent neural speech codecs have shown impressive progress, their performance significantly deteriorates at low bitrates (around 1 kbps). Although a low bitrate inherently restricts performance, other factors, such as model capacity, also hinder further improvements. To address this problem, we scale up the model size to 159M parameters that is more than 10 times larger than popular codecs with about 10M parameters. Besides, we integrate sequential models into traditional convolutional architectures to better capture temporal dependency and adopt low-dimensional vector quantization to ensure a high code utilization. Comprehensive objective and subjective evaluations show that BigCodec, with a bitrate of 1.04 kbps, significantly outperforms several existing low-bitrate codecs. Furthermore, BigCodec achieves objective performance comparable to popular codecs operating at 4-6 times higher bitrates, and even delivers better subjective perceptual quality than the ground truth.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
