在工廠生產線中,產品之生產與加工多透過作業人員使用機台設備進行之;而不當之操作機台設備易導致工安事故之發生,故機台設備之安全操作成為工廠重視之課題。機台設備之運作狀況多透過聲音訊息傳達予現場作業人員,以讓現場工作人員可藉由辨識聲音訊息進而執行聲音訊息所對應之處置動作。而工業指導書或工廠作業手冊中多以文字形式描述機台設備運作所發出之聲音訊息以及其對應之處置方式,相關人員往往需花費時間揣摩或瞭解以文字形式表達之聲音訊息,故以文字型態表達之聲音知識往往讓知識吸收者陷入需花費時間閱讀與理解內容之困境,且往往無法讓員工具體且清楚地掌握聲音訊息與其對應之處置措施。 為能讓知識吸收者可透過具體之聽覺感受瞭解文字形式所描述之聲音訊息,本研究乃發展一套「聲音知識語音化」方法論,以將以文字型聲音知識轉以語音化之方式呈現予知識吸收者。本研究乃先針對具聲音知識之文件內容(如機械設備使用說明書或設備操作手冊)進行收集、分析與整理,以瞭解表達聲音知識內容之結構與元素。之後,再根據聲音知識之解析結果建立聲音知識判斷詞庫,以作為由自由形式知識文件擷取聲音知識目標文句之基礎。最後,本研究即根據聲音知識判斷詞庫與表達聲音知識之文句結構發展一套聲音知識語音化方法論,此方法論之詳細作法為先將自由形式之知識文件全文進行文句標示,再對標示文句進行篩選,以取得可能含有聲音知識之待選文句,進而針對待選文句進行文句關聯解析,以取得具聲音知識之文句。之後,由自由形式表達聲音知識之目標文句中擷取呈現聲音類型、音量大小與時間久暫三種聲音特性之關鍵內容轉化為具結構化之內容,以利用此結構化之內容精鍊表達聲音知識文句之內涵,最後再配合語音化方式呈現結構化之聲音知識文句內容,達到聲音知識語音化之目的。 本研究除發展聲音知識語音化之方法論外,亦根據此方法論建構一套聲音知識語音化資訊分享系統,並以「設備操作知識文件」為案例進行系統驗證,以確認本方法論之可行性及績效。而由驗證結果得知,本系統僅需一定數量之訓練資料即可使系統推論績效達一定水準。整體而言,本研究所提出之聲音知識語音化模式與技術可有效將聲音知識以語音化之方式呈現與知識吸收者,以協助知識吸收者快速理解與吸收抽象之聲音知識,進而提升知識之再利用率。
In the shop floor of a factory, operators have to use machines to perform manufacturing processes. Usually the operation status of machines can be identified via sounds. In the equipment user manuals, knowledge is usually represented via texts or illustrations and knowledge receivers might spend much time to recognize the text-based sound expressions. Thus, a vocalization representation scheme for the text-based sound expressions can assist knowledge receivers to efficiently and effectively recognize this type of knowledge. This research aims at developing a knowledge vocalization methodology in order to convert the knowledge contents with text-based sound expressions into the vocalized expressions. The proposed methodology consists of three modules namely Sound Expression Identification (SEI), Target Sentence Extraction and Formatting (TSEF) and Knowledge Content Vocalization (KCV). In the SEI module, the components with sound expressions are identified from the sentences. Based on the identified sound components, the target sentences with sound expressions are extracted from the free-form documents and expressed as formatted matrices via the TSEF module. In the KCV module, all text-based, formatted sound expressions are represented via vocalized expressions. As the knowledge contents with sound expressions can be represented via knowledge vocalization, knowledge receivers can efficiently recognize the knowledge contents and knowledge reuse can be facilitated. Moreover, based on the proposed methodology, a Web-based prototype system for vocalized knowledge sharing is also developed and the equipment manuals are employed to evaluate the feasibility and performance of the proposed methodology. As a whole, this research provides a knowledge representation and vocalization model to facilitate knowledge receivers to efficiently and accurately acquire the knowledge contents with sound expressions.