Unverified50% confidenceFactExact time
语音被转换为文字中间表示时，原始音频中约40%的信息随之消失，包括情绪、强调、语速等副语言信息
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
VerifiedIn traditional voice interaction systems, paralinguistic information carried by speech—such as emotion, tone, and speech rate—is completely lost when converted to plain text during the ASR stage.83% similarUnverified说话人嵌入向量捕捉声音中与内容无关的身份信息，包括声带共振频率、基频范围和发音方式74% similarUnverified语音数据通常被编码为离散的音频token，每秒语音大约对应25-50个token73% similarUnverifiedResearch indicates that as many as 60%–70% of syllables in native speakers' everyday conversation undergo varying degrees of reduction.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/517801API
curl https://kongchang.com/api/v1/knowledge/claims/517801MCP
get_claim(id=517801)