Unverified50% confidenceFactExact time
After zero-training PCA injection, audio-to-audio tasks improve: gender clustering jumps from 0.05% to 63.1%, artist retrieval rises from 55.2% to 93.8%, emotion recognition from 19.7% to 40.2%, and environmental sounds from 38.8% to 56.3%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/28/2026
First Seen
Valid until: 1/26/2027
Sources
Related Entities
Related Claims
Unverified有研究表明在自发语音场景下WER从5%提升到8%，用户感知的任务完成率可能下降超过30%63% similarUnverifiedThe addition of fixed acoustic features slightly reduces the proportion of the 512d semantic component after normalization, harming audio-to-audio tasks that need to ignore speaker timbre: spoken topic classification (SIBFLEURS) drops from 57.2% to 21.3%, and spoken intent (MInDS14) drops from 97.6% to 88.3%.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1108288API
curl https://kongchang.com/api/v1/knowledge/claims/1108288MCP
get_claim(id=1108288)