推荐telotortium · 2022-10
“Impressive results - on CD-quality audio too, not just speech as with many recent codecs.”
推荐它,结果令人印象深刻,连 CD 音质的音频也行,不只是语音
流式编码器–解码器结构的实时神经音频编解码器 EnCodec
提出实时、高保真的神经音频编解码器:流式编码器–解码器结构,端到端训练、潜空间量化;用单个多尺度频谱对抗器简化并加速训练,并引入损失均衡器稳定训练。还研究用轻量 Transformer 把码流再压缩至多 40%;在语音、带噪混响语音和音乐上的主观评测(MUSHRA)中,各带宽设置下都优于基线。
评论者对其在 CD 级音频(而不只是语音)上的结果印象深刻,另有人指出官方 GitHub 仓库只有推理、没有训练代码。
“Impressive results - on CD-quality audio too, not just speech as with many recent codecs.”
“High Fidelity Neural Audio Compression (the github repo doesn't include training, just inference)”