杜 淼 黃天淏 邊 彤 顏逸為 余 勤 雒瑞森





摘 要:由于無線電技術的日益成熟,盜用正常廣播頻段進行其他活動的非法廣播對國民經濟和安全造成相當大的危害,因此對非法廣播的監測非常重要。本文集成隱馬爾科夫模型對廣播關鍵詞進行識別,進而監測非法廣播。在試驗中,首先對采集的非法廣播進行人工切割與標定用于訓練,然后研究基于集成學習的方法組合多個模型,使用投票規則得到最終結果。將集成學習的PocketSphinx系統與單一模型進行比較,試驗結果顯示,與單一模型84.8%的識別率相比,集成的PocketSphinx系統識別率達到92%,并且具有更好的穩定性。
關鍵詞:關鍵詞識別;PocketSphinx;隱馬爾科夫模型;集成學習
中圖分類號:TN912.34 文獻標識碼:A 文章編號:1003-5168(2019)35-0008-04
Application of Hidden Markov Model with Ensemble Learning
in Broadcast Keyword Detection
DU Miao1 HUANG Tianhao BIAN Tong1 YAN Yiwei1 YU Qin1 LUO Ruisen1
(1.College of Electrical Engineering, Sichuan University,Chengdu Sichuan 610000;2. College of Electronic Engineering and Automation, Guilin University of Electronic Technology,Guilin Guangxi 541004)
Abstract: Due to the increasing maturity of radio technology, illegal broadcasts that use the normal broadcast band for other activities have caused considerable harm to the national economy and security, so the monitoring of illegal broadcasts is very important. This paper integrated hidden Markov models to identify broadcast keywords, and then monitored illegal broadcasts. In the experiment, the illegal broadcasts collected were first manually cut and calibrated for training, then the method based on ensemble learning was used to combine multiple models, and the voting rules were used to obtain the final results. Comparing the integrated learning PocketSphinx system with a single model, the experimental results show that compared with the single model's 84.8% recognition rate, the integrated PocketSphinx system has a recognition rate of 92% and has better stability.
Keywords: keyword recognition;PocketSphinx;Hidden Markov Model;ensemble learning
隨著無線電技術應用的成熟,部分不法分子采用盜用調頻廣播的手段進行非法活動,擾亂正常廣播秩序,對國民經濟和安全造成了嚴重威脅。在實際的應用過程中,非法廣播關鍵詞識別往往存在諸多復雜問題,例如,存在設備和環境影響產生的噪聲,沒有標準數據集進行訓練,人聲混合等。因此,本文的研究環境是一個非常復雜的實際應用場景。隱馬爾科夫模型(Hidden Markov Model,簡稱HMM)可分為:離散隱馬爾科夫模型(DHMM)、連續隱馬爾科夫模型(CHMM)和半連續隱馬爾科夫模型(SCHMM)[1]。SCHMM作為DHMM和CHMM的折衷方法,盡量避免DHMM因矢量化信息造成信息損失和CHMM由于待估計參數過多需要大量訓練集[2,3]。
由于實際的非法廣播信號中存在噪聲、不同人聲和方言等問題,所以本研究采用集成學習的方法提高模型的魯棒性。……