Unverified50% confidenceFactExact time
协同过滤通过分析数百万用户的行为相似性来预测个体偏好，内容过滤则分析音频信号的音色、节奏、调性等特征
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedThe hybrid retrieval strategy comprises a three-layer filtering mechanism—keyword retrieval + semantic retrieval + reranking (Reranker)—which can significantly reduce noise interference.65% similarUnverified扩散/流匹配模型（如Voicebox、CosyVoice）通过迭代去噪生成连续语音特征，在质量和一致性间取得较好平衡64% similarUnverified科林·切里通过二分听觉（Dichotic Listening）实验发现人类大脑能依靠音调、音色、空间方位和语义线索选择性关注特定声源63% similarUnverifiedIn multimodal on-device inference, audio signals need to be converted into feature vectors via methods such as mel spectrograms, while motion data comes from the phone's accelerometer and gyroscope.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/571848API
curl https://kongchang.com/api/v1/knowledge/claims/571848MCP
get_claim(id=571848)