[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fB87dCyifgYsX46JFRxiyT8sFWP4Gahid2v_gIfwV5pU":3},{"answer":4,"createTime":5,"id":6,"options":7,"origin":12,"question":19,"related":20,"source":31,"type":32},[],"2024-06-19 10:41:56",153136253,[8,9,10,11],"ngram-count -read sample.count -order 3 -lm sample.lm","ngram-count -read sample.count -order 3 -lm sample_wb.lm -wbdiscount","ngram-count -read sample.count -order 3 -lm sample_kn.lm -interpolate -kndiscount","ngram-count -text sample.txt -order 3 -write sample.count",{"count":13,"courseId":14,"courseImg":15,"courseName":16,"workId":17,"workName":18},18,"e435984f691240c6fc7eed81e23d857c","https:\u002F\u002Ftihai-oss-cloud.itihey.com\u002Fimg\u002F9c1e48361b00f3ee2086f4e259ed792b.jpg","语音识别综合实践案例","work_35262542","第五次作业","使用SPILM训练语言模型时,如果语料太少,可以使用以下哪种方式训练()",[21,33,42,51,60,69,77,86,95,104],{"answer":22,"createTime":23,"id":24,"options":25,"question":30,"source":31,"type":32},[],"2024-06-19 10:41:55",153136244,[26,27,28,29],"语音识别","语音合成","语音增强","说话人识别","在语音领域,均方误差损失函数主要用于()","v1",0,{"answer":34,"createTime":23,"id":35,"options":36,"question":41,"source":31,"type":32},[],153136245,[37,38,39,40],"梯度裁剪","权重衰减","参数归一化","随机丢弃","L1和L2正则化方法也叫做()",{"answer":43,"createTime":23,"id":44,"options":45,"question":50,"source":31,"type":32},[],153136246,[46,47,48,49],"1:1","n:m","1:n","n:1","DNN-HMM的输出节点与HMM的状态节点是()关系",{"answer":52,"createTime":23,"id":53,"options":54,"question":59,"source":31,"type":32},[],153136247,[55,56,57,58],"351","390","350","389","DNN的优势在于能够对包含多帧数据的特征进行处理,表征更丰富的语音上下文特征,DNN的输入层往往先进行多帧拼接,加入每帧语音特征39维度,对前10帧进行拼接,拼接后的特征有几维()",{"answer":61,"createTime":23,"id":62,"options":63,"question":68,"source":31,"type":32},[],153136248,[64,65,66,67],"MFCC","PLP","FBANK","以上都不对","以下哪种声学特征适合基于深度学习的语音模型()",{"answer":70,"createTime":23,"id":71,"options":72,"question":76,"source":31,"type":32},[],153136249,[73,74,75,67],"人工打标","基于其它模型输出的帧与状态的对齐文件","基于人工输出的帧与状态的对齐文件","DNN-HMM语音识别模型是有监督学习,真实标签是从哪里获取到的()",{"answer":78,"createTime":23,"id":79,"options":80,"question":85,"source":31,"type":32},[],153136250,[81,82,83,84],"1,250","1,150","50,250","50,150","假如有50个音素,每个音素有使用5个状态建模,每个音素有效发射状态是3个,那么要训练()个GMM-HMM模型,DNN-HMM的输出节点有多少个()",{"answer":87,"createTime":23,"id":88,"options":89,"question":94,"source":31,"type":32},[],153136251,[90,91,92,93],"(1,0,1)","(-1,0,-1)","(-1,0,1)","(-1,1,0)","TDNN描述当前帧与过去1帧、未来1帧拼接为3帧的节点作为输入节点的表示是()",{"answer":96,"createTime":23,"id":97,"options":98,"question":103,"source":31,"type":32},[],153136252,[99,100,101,102],"1","2","3","n","n-gram模型表示()词之间的组合概率",{"answer":105,"createTime":5,"id":6,"options":106,"question":19,"source":31,"type":32},[],[8,9,10,11]]