论文解读
浅正字法不等于无重音：用弱监督让 ByT5 在句子级学会菲律宾语重音
针对菲律宾语重音需靠句子上下文消歧但句子级音素标注稀缺的问题，该研究用 Wiktionary 词典加 LLM 辅助管线合成句子级弱监督数据微调 CharsiuG2P 初始化的 ByT5，在 1173 句人工校对集上把音素错误率从约 19.74% 降到约 0.54%、字符错误率降到约 2.50%，代价是 malumi 类与非标准词仍易错且大合成集增益主要体现在 …
弱监督学习
论文解读
EAR: Enhancing Uni-Modal Representations for Weakly Supervised Audio-Visual Video Parsing
音频事件检测 | 5.8/10
论文解读
DBFT-SD: Weakly Supervised Multimodal Detection of Sensitive Audio-Visual Content
音频事件检测 | 8.0/10
论文解读
Deepaq: A Perceptual Audio Quality Metric Based on Foundational Models and Weakly Supervised Learning
音频质量评估 | 7.5/10
论文解读
Is Phase Really Needed for Weakly-Supervised Dereverberation?
语音增强 | 6.0/10
论文解读
Look, Listen and Segment: Towards Weakly Supervised Audio-Visual Semantic Segmentation
音视频 | 7.0/10
论文解读
Subsequence SDTW: Differentiable Alignment with Flexible Boundary Conditions
音乐信息检索 | 8.0/10
论文解读
Teacher-Guided Pseudo Supervision and Cross-Modal Alignment for Audio-Visual Video Parsing
音视频 | 7.0/10