作業幫直播課“語音彈幕”功能研發實踐

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"點擊屏幕上的小話筒,大聲喊出答案,就會看到自己的語音變成文字飄在屏幕上方,不僅如此,還能看到同一小組其他夥伴的發言。在作業幫直播課“小組直播間”,“語音彈幕”這一互動形式讓小學生感到新奇。"}]},{"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":"喜歡這一功能的不僅是小學生。在“小組直播間”產品用戶調研中,“語音彈幕”功能在語文學科的主講老師中的使用率高達92%。看圖說話、讀古詩是老師們最愛的使用場景。"}]},{"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":"從沉默到活躍,“語音彈幕”這一“短頻快”的互動方式打開了作業幫直播課全新的教學互動形式。在AI技術的賦能下,在線課堂熱鬧了起來。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"大班小組化:用互動打破孤獨感"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"孤獨感,是曹越對傳統大班直播課的評價,“一個學生對着一塊屏幕,孤獨地看視頻。”"}]},{"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":"曹越是作業幫直播課小學產品的負責人。在曹越看來,由課件、老師、聊天區構成的“三分屏”模式並未充分尊重小學生的天性。“學習並非單向投餵的過程,氛圍感的建立非常重要。”"}]},{"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":"在大班課內建立以6人爲單位的小組,通過師生間互相可以看見的“雙向視頻”模式,以及小組PK贏學分等激勵機制,“小組直播間”重塑了大班直播課的形態。"}]},{"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":"2020年1月,作業幫在小學各年段的大班直播課中上線了“小組直播間”功能。“我們希望打造一個學習的‘場’,讓學員感受到有同伴,就跟線下一樣,和自己的夥伴一起學習。老師和學生給你點贊加油,就跟真實的課堂一樣。”"}]},{"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":"互動是打破孤獨感的利器,小組則能營造沉浸的學習氛圍。在小組化學習理念的驅使下,作業幫產品和技術用了近一年的時間,藉助AI語音交互、視覺互動等手段,不斷充盈着小組直播間,而“語音彈幕”作爲首個語音識別類的功能,拉近了在線直播課上師生、同學之間的距離。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/69\/af\/69b4bab10768b0485f6123ea649b46af.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","text":"圖注:“語音彈幕”功能使用場景示例"}]},{"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":"在三分屏模式下,“是否卡”、“選項卡”、“投票卡”是學生和老師交流的唯一途徑。在一節60~90分鐘的課上機械而重複的點擊回答卡片,學生難免會感覺無聊。"}]},{"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":"在和小學生和家長的大量接觸中,作業幫小學部產品經理聶靖騏感到,對於小學階段的孩子來說,激發學習興趣、營造學習氛圍纔是重點。"}]},{"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":"在一次刷短視頻彈幕的時候,聶靖騏腦子裏突然閃過一個念頭,“如果上課可以發彈幕,孩子們會發什麼?”因爲互動性強,彈幕已經成爲年輕人中頗受歡迎的交流方式。“這麼有趣的方式,爲什麼不能接入‘小組直播間’的課堂?”"}]},{"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":"但對於小學生來說,困難也是顯而易見的。"}]},{"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":"“有時候自己發彈幕,都會擔心打字的過程錯過了精彩內容,對於小孩子來說,像成年人那樣快速打字,更是不太現實的。”"}]},{"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":"再退一步來說,即使打字速度能跟上,打字過程中,鍵盤遮擋了一半的屏幕,非常容易錯過關鍵信息。"}]},{"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":"如果換成語音,會不會好一些?"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"語音彈幕功能上線"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2020年7月,在團隊內部討論後,研發“語音彈幕”的需求正式提上了日程。"}]},{"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":"剛接到這一需求時,作業幫智能語音技術負責人王強強覺得這是個“科技含量”並不高的功能。王強強曾是清華大學語音實驗室的技術負責人,在他看來,“語音彈幕”技術並無值得同行稱道的創新之處,但一次和阿里研發人員的聊天卻轉變了他的看法。"}]},{"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":"\"我女兒上二年級,特別喜歡你們的語音彈幕功能,她說電腦竟然能聽懂她說的話,非常‘amazing’。\"專業技術人員覺得稀鬆平常的設計,但卻給低齡兒童帶來十足驚喜。"}]},{"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":"本着“試水”的心態研發上線後,“語音彈幕”功能良好的反饋也讓團隊感到驚喜。"}]},{"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":"從技術層面來說,語音識別在行業裏並無祕密,但從在線教育場景來看,語音彈幕的瞬時高併發成爲了最大難點。由於之前並無準備,暑假期間高頻次的使用,讓後端機器差點兒沒扛住。"}]},{"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":"“你就想象一堂課上萬學生同時開麥說話,要同時扛住幾萬路併發,咱們機器壓力得有多大?”"}]},{"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":"原本這功能都沒有申請新的預算,但爲了扛住這份“甜蜜的負擔”,王強強團隊緊急增加機器,投入幾十萬把集羣擴了一倍。“我們技術總負責人拍板,學員反饋好,加機器也要上。”"}]},{"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":"回顧這一次研發經歷,王強強更加深刻地體會到用戶反饋的重要性。“我們做技術的不能閉門造車,研發的最終目的還是爲了用戶體驗更好。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"行業“最強算法”:不只差了一點點"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不打無準備之戰。在扛過了2020年暑期課程的難關後,王強強化被動爲主動,通過更加靈活的架構、行業最低的動態擴收容成本,同時還有阿里、騰訊等多家技術備份,迎接寒假課程。"}]},{"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":"在硬件準備到位後,王強強開始對語音識別技術的準確率精細打磨。新聞聯播主持人口播,系統的語音識別準確率可以達到99%,但在低齡小學生語音的場景下,準確率會因各種原因大打折扣。"}]},{"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":"“尚處語言學習階段的小學生語音表達不清晰、答非所問,又或者家長在一旁干擾,加之方言和口音、環境噪音等因素,都會影響語音識別的準確率。”"}]},{"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":"在王強強看來,方言識別在全行業尚無好的解決辦法,只能靠手動標記;而對於環境噪音,採取傳統噪聲抑制加深度學習的方法,在語音彈幕場景下取得了良好的效果。"}]},{"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":"在教育領域內容識別上,作業幫有足夠大的資源庫。在這方面,即使和互聯網大廠相比,作業幫都有足夠的底氣做到最好。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/ea\/4f\/eab9f332c05cdf74bd3335790736ee4f.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":"center","origin":null},"content":[{"type":"text","text":"圖注:作業幫產品研發團隊討論場景"}]},{"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":"“我們的用戶量足夠大。你讓幾千個孩子去說,一定有十幾二十個說得標準的,我們依託這樣的場景,研發一套算法將好的話語快速‘挑’出來。”"}]},{"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":"擯棄手工標數據,王強強帶領團隊研發一套自動篩選算法,能在短時間內自動挑選領域內數據,並迅速學習增強自身模型能力。“目前我們這套算法應該是行業裏最快,不是說差了一點點,應該是差了很多。”"}]},{"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":"除了在技術上不斷精進外,王強強也積極推動着產品、研發閉環的打造。在他看來,技術埋頭研發,然後推銷給別的部門的方式,非常被動。“我相信所有的大廠,阿里、騰訊其實他們都這麼幹,但現實情況是,如果不能預判用戶的需求,總是被動響應,做完一個東西再去推銷,肯定行不通。”"}]},{"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":"過去一年,王強強花了大量的精力在公司內部做培訓,打入業務方,充分尊重用戶反饋,而他對自己的定位,也不再是單純的技術人員,而是多面手。"}]},{"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":"“我們是做技術的,但我們也可以是產品,可以是研發,也可以是銷售,甚至可以是HR。”"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"迴歸教育本質:科技助力拓展更多可能性"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2016年,阿爾法狗的爆火,讓AI技術突然進入了普通大衆的視野,但對王強強來說,AI技術的發展應用前景遠不止於此。“我出去都不說自己是做AI的,就說是做語音識別的。”在王強強看來,目前AI功能在應用層面還是比較簡單的,遠未到人工智能的程度。"}]},{"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":"在他看來,AI行業現在有點類似於20世紀初汽車行業的發展狀況。1903年內燃機剛上路時,倫敦街頭仍會出現“馬拉汽車”的景象,“當時好多人都不看好它,因爲它老出問題,老拋錨,但100年過去,馬車消失了,馬路上跑的都是汽車。”王強強相信,雖然目前AI技術仍處於感知智能階段,遠未達到認知智能的預期,但是它的前景依然遠大。"}]},{"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":"而在線教育,則是AI技術落地很廣的一個應用場景,每一個在線課堂上的學生,對它的應用都有感知。而未來,AI交互將更加智能、自然,更加顯著地提升在線課堂的交互體驗。"}]},{"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":"基於對未來AI發展趨勢的預判,新一代家長的教育理念也隨之升級,越早讓孩子接觸,便能觸發越多的可能性。而對於那些暫無機會直接接觸優質信息資源的孩子來說,在線教育則以跨越時間、空間的普惠性,縮短着由資源差異帶來的信息差距。"}]},{"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":"正如作業幫創始人侯建彬所說,“在線教育的最大價值,還是在於普惠。在此之前,整個教培體系從不缺乏優秀的老師和優質的內容,缺乏的是把這些資源放大、遷移、匹配的便捷管道。”"}]}]}
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章