基於神經網絡的多音區語音喚醒 | 論文解讀

{"type":"doc","content":[{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"1.概述"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"人工智能技術迅猛發展，人機語音交互更加自然，搭載語音喚醒、識別技術的智能設備也越來越多。語音喚醒在學術上稱爲keyword spotting(簡稱KWS)，即在連續語流中實時檢測出說話人特定片段（比如：叮噹叮噹、Hi Siri等），是一種小資源的關鍵詞檢索任務，也可以看作是一類特殊的語音識別，應用在智能設備上起到了保護用戶隱私、降低設備功耗的作用，經常扮演一個激活設備、開啓系統的入口角色，在手機助手、車載、可穿戴設備、智能家居、機器人等運用得尤其普遍。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"喚醒效果好壞的判定指標主要有召回率"},{"type":"text","text":"(recall"},{"type":"text","text":"，俗稱喚醒率"},{"type":"text","text":")"},{"type":"text","text":"、虛警率"},{"type":"text","text":"(false alarm"},{"type":"text","text":"，俗稱誤喚醒"},{"type":"text","text":")"},{"type":"text","text":"、響應時間和功耗四個指標。召回率表示正確被喚醒的次數佔總的應該被喚醒次數的比例。虛警率表示不該被喚醒卻被喚醒的概率，工業界常以"},{"type":"text","text":"12"},{"type":"text","text":"或者"},{"type":"text","text":"24"},{"type":"text","text":"小時的誤喚醒次數作爲系統虛警率的評價指標。響應時間是指用戶說出喚醒詞後，設備的反應時間，過大的響應時間會造成較差的用戶體驗。功耗是指喚醒系統的耗電情況，多數智能設備都是電池供電，且需要保證長時續航，要求喚醒系統必須是低耗能的。一個好的喚醒系統應該保證較高的召回率、較低的虛警率、響應延時短、功耗低。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}}]}