Unverified50% confidenceFactExact time
Filler words, non-lexical expressions, and false starts lack stable pronunciation patterns and easily cause mismatches between the acoustic model and the text sequence.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTraditional pipeline architectures for speech translation have fundamental flaws including error propagation, cumulative latency, and loss of paralinguistic information.73% similarUnverified热词偏置的偏置权重设置过高会导致误召回，设置过低则无法有效纠正音译行为71% similarUnverified传统三段式语音管线中每个模块独立训练，误差会逐级累积，且文字化过程会不可逆地丢失说话者的语调、停顿、情绪等副语言信息71% similarVerifiedLarge language models suffer from 'attention decay' when generating long text, causing adherence to earlier instructions to decrease as context length increases69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/914087API
curl https://kongchang.com/api/v1/knowledge/claims/914087MCP
get_claim(id=914087)