CNSV VID × DSpotter KWS 整合架構

CNSV VID × DSpotter KWS 整合架構 An architecture diagram generated by Archify. Audio capture · 16 kHz · mono · PCM16 · Architecture component Audio capture 16 kHz · mono · PCM16 Audio fan-out · application-owned · Architecture component · same samples · same order Audio fan-out application-owned same samples · same order DSpotter KWS · trigger phrase detection · Architecture component · independent handle DSpotter KWS trigger phrase detection independent handle CNSV VID · ring buffer + speaker ID · Architecture component · independent handle CNSV VID ring buffer + speaker ID independent handle Hit coordinator · application policy · Architecture component · timing → sample offsets Hit coordinator application policy timing → sample offsets Wake + identity · application callback · Architecture component Wake + identity application callback DSpotter assets · command pack + license · Architecture component DSpotter assets command pack + license CNSV assets · pretrained + speakers · Architecture component · model + license CNSV assets pretrained + speakers model + license PCM samples DSpotterAddSample CNSVApi_AddSample InitMultiWithPackBin Init + AddSpeaker hit · word duration · silence · latency nStart/nEnd · TrainSpeaker/GetResult trigger phrase speaker + score Legend Frontend Backend Database Security External

整合邊界

  • • CNSV 與 DSpotter 各自初始化、重置、釋放
  • • 兩個 engine 不直接互呼；應用程式負責 dual-feed 與結果仲裁

Audio contract

  • • 16,000 Hz · mono · signed PCM16
  • • nNumSample 是 sample 數；API header 沒有規定固定 frame size

命中區間

  • • nStart = word duration + end silence + network latency
  • • nEnd = end silence + network latency
  • • DSpotter header 未定義 timing 單位；跨版本前需向 vendor 確認