Unverified50% confidenceFactExact time
传统语音助手依赖固定唤醒词的关键词检测模型（KWS），参数量通常在100KB到1MB量级，在毫瓦级功耗下持续运行
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedWake word detection typically runs lightweight models of only a few hundred KB to a few MB to continuously monitor audio streams; OpenWakeWord and Porcupine are mature open-source implementation solutions.72% similarUnverified传统语音助手依赖固定唤醒词如Hey Siri或OK Google，底层为轻量级关键词检测模型71% similarUnverified本地语音识别处理通常需要4-8GB内存用于模型加载，较大模型可能需要更多68% 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token67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561346API
curl https://kongchang.com/api/v1/knowledge/claims/561346MCP
get_claim(id=561346)