论文解读
15 秒参考合成整库：固定音色泰语 TTS 的质量与覆盖取舍
该工作用 15 秒参考驱动 OmniVoice 教师批量合成泰语语料并训练 82M 固定音色 Kokoro 学生，以 68.2% 挑战集关键词准确率与 91.4% 停顿精确率保留音色并减少词内停顿，代价是难词实现仍落后于教师与更大系统。
知识蒸馏
论文解读
只抄最后一层表示：预量化潜变量蒸馏压缩流式音频前端
该文只训练窄而深的学生编码器逐帧回归教师预量化潜变量，在 2.8 倍参数压缩下六组配对中五组相对词错误率退化不超过 1.9%，同容量独立训练对照优 3.9% 相对，而联合训练的 J3 退化 7.7% 揭示教师与阶段边界。
论文解读
单塔为何还能赢双塔：BiMTokenizer 用双向状态与固定格点重做 1.1 kbps 的语义-声学平衡
语音编码 | 8.3/10
论文解读
7B 做原生音画同步：用门控与路由把对齐做轻，而不是把参数做大
扩散模型 | 5.9/10
论文解读
把教师的回声残差教给小模型：AEC 蒸馏到底补回了什么
回声消除 | 8.5/10
论文解读
Long-Horizon Audio-Visual Generation for Persistent Stories and Interactive Worlds
音视频生成 | 8.9/10
论文解读
sanoTTS: The Smallest Real-Time Neural TTS on a General-Purpose Microcontroller
语音合成 | 10.0/10
论文解读
DAVSS: Distilled Audio-Visual State Space Models
音视频理解 | 6.4/10
论文解读
AnyTalk: Speech Animation for Arbitrary Characters Leveraging a Video Generation Model
音视频语音合成 | 6.7/10
论文解读
Speaker-Normalized Semantic Speech Tokens via Iterative S2U-T2U Refinement
语音编码 | 6.0/10
论文解读
Avatar-Forever: Decoupled Parallel Training for High-Quality Real-Time Infinite Avatars
音视频生成 | 6.6/10
论文解读
Omni-LiveAvatar: Minute-Level Real-Time Streaming Joint Audio-Visual Avatar Generation
音视频生成 | 7.7/10
论文解读
On-Policy Self-Distillation for Multi-Dialect ASR: Mastering Dialects, Retaining Mandarin
语音识别 | 7.0/10
论文解读
RT-SEMamba: Real-Time Speech Enhancement Mamba via Progressive Knowledge Distillation
语音增强 | 6.6/10
论文解读
UniSwap: Streaming Audio-Visual Identity Swapping for Talking Videos
音视频生成 | 7.8/10
论文解读
DINO-A: Adapting Self-Distillation Vision Transformers to General Audio Representation Learning
音频分类 | 6.2/10
论文解读
CuteTTS: Efficient and High-Quality Speech Synthesis via Autoregressive Modeling of Continuous Latents
语音合成 | 7.0/10
论文解读
ReLMCodec: Designing Predictable Speech Tokens from Pre-Quantization Phoneme Structure
语音编码 | 7.7/10
论文解读
Cloud-Boosted Low-Compute Multi-Channel Speech Enhancement
语音增强 | 6.0/10
论文解读
Equivariant Music Transformer
音乐生成 | 8.6/10