开通博客赚积分
发布资源赚积分
分类

源码开发语言/平台
当前位置：首页 > 源码/资料 > 嵌入式/单片机编程 > DSP编程 > 查看源码
DHMM_RECOG_MFC.cpp
资源名称：VQ-DHMM.rar　[点击查看]
上传用户：avbj512
上传日期：2013-09-18
资源大小：6239k
文件大小：15k
源码类别：

DSP编程

开发平台：

Visual C++

DHMM_RECOG_MFC.cpp：源码内容
  1. // DHMM_RECOG_MFC.cpp:
  2. // Implementation of the DHMM_RECOG_MFC Module.
  3. // That is to compute Max_likelihood or Max_Output_Probably.
  4. //
  5. // Created 2001/08, By DongMing, MDSR.
  6. //
  7. //////////////////////////////////////////////////////////////////////
  9. #include "stdafx.h"
  10. #include "DHMM_Recog_MFC.h"
  11. #include "kwspot.h"
  12. #include "DHMM_Model_MFC.h"
  13. #include "DAT_File_Access.h"
  14. #include "DHMM_VQ_MFC.h"
  16. #include "DHMM_GL.h"
  17. #include "DHMM_LHS.h"
  18. #include "DHMM_HQ.h"
  20. extern PRO_CONFIG u_Pro_Config;
  22. //////////////////////////////////////////////////////////////////////
  23. // API functions
  25. int DHMM_Recog_Kitty(void)
  26. {
  27. u_Pro_Config.l_DHMM_Recog_Config_Set = RECOG_CONFIG_DHMM_RECOG_TRAINSET;
  28. PRO_LOG("tRecognition using Train Set...n");
  29. //DHMM_Recog_One_Set();
  30. RESULT_LOG("n--------Train Set. Accuracy Ratet%dttt%8.4fn", u_Pro_Config.n_DHMM_Model_Person_Start_Index, u_Pro_Config.d_Recognition_Accurate_Rate);
  32. u_Pro_Config.l_DHMM_Recog_Config_Set = RECOG_CONFIG_DHMM_RECOG_TESTSET;
  33. PRO_LOG("tRecognition using Test Set...n");
  34. DHMM_Recog_One_Set();
  35. RESULT_LOG("n--------Test Set. Accuracy Ratet%dttt%8.4fn", u_Pro_Config.n_DHMM_Model_Person_Start_Index, u_Pro_Config.d_Recognition_Accurate_Rate);
  36. return 0;
  37. }
  39. int DHMM_Recog_One_Set(void)
  40. {
  41. int n_Recog_Result, n_True_Result;
  42. char sz_Feature_File_Name[256];
  43. int n_Recog_Result_Good_Count = 0, n_Recog_Result_Bad_Count = 0;
  44. int n_Person_Index, n_Word_Sample_Index;
  46. int Total_Model_Num;
  48. Total_Model_Num =u_Pro_Config.n_DHMM_Model_Num;
  50. double* pd_DHMM_Model_Probably;
  51. pd_DHMM_Model_Probably = new double[Total_Model_Num];
  52. ASSERT(pd_DHMM_Model_Probably);
  53. DYNA_2DIM_INT_ARRAY d2dna_Recog_Table = d2dna_New(u_Pro_Config.n_DHMM_Model_Num, Total_Model_Num);
  54. for (int n_Sentence_Index = 0; n_Sentence_Index < u_Pro_Config.n_DHMM_Model_Num; n_Sentence_Index++)
  55. for (int nTmp = 0; nTmp < Total_Model_Num; nTmp++)
  56. d2dna_Recog_Table[n_Sentence_Index][nTmp] = 0;
  58. // 对所有测试集的人进行循环
  59. for (n_Person_Index = 0; n_Person_Index < u_Pro_Config.n_Recog_Person_Num;  n_Person_Index++)//u_Pro_Config.n_Recog_Person_Num; n_Person_Index++)
  60. {
  61. //  判断是否测试集中的人，不是则跳出此轮循环
  62. if(u_Pro_Config.l_DHMM_Recog_Config_Set == RECOG_CONFIG_DHMM_RECOG_TRAINSET)
  63. {
  64. if (u_Pro_Config.n_Recog_Person_Start_Index <= u_Pro_Config.n_Recog_Person_End_Index)
  65. {
  66. if (!((n_Person_Index < u_Pro_Config.n_Recog_Person_Start_Index) || (n_Person_Index > u_Pro_Config.n_Recog_Person_End_Index))) continue;
  67. }
  68. else
  69. {
  70. if (!((n_Person_Index > u_Pro_Config.n_Recog_Person_End_Index) && (n_Person_Index < u_Pro_Config.n_Recog_Person_Start_Index))) continue;
  71. }
  72. }
  73. else if(u_Pro_Config.l_DHMM_Recog_Config_Set == RECOG_CONFIG_DHMM_RECOG_TESTSET)
  74. {
  75. if (u_Pro_Config.n_Recog_Person_Start_Index <= u_Pro_Config.n_Recog_Person_End_Index)
  76. {
  77. if ((n_Person_Index < u_Pro_Config.n_Recog_Person_Start_Index) || (n_Person_Index > u_Pro_Config.n_Recog_Person_End_Index)) continue;
  78. }
  79. else
  80. {
  81. if ((n_Person_Index > u_Pro_Config.n_Recog_Person_End_Index) && (n_Person_Index < u_Pro_Config.n_Recog_Person_Start_Index)) continue;
  82. }
  83. }
  85. // 对每个人的每个词，进行一次识别测试
  86. sprintf(sz_Feature_File_Name, u_Pro_Config.sz_Recog_Origin_File_Name_Format, n_Person_Index);
  87. for (n_Word_Sample_Index = 0; n_Word_Sample_Index < u_Pro_Config.n_Recog_Sentence_Num; n_Word_Sample_Index++)
  88. {
  89. int Sen_Index = n_Word_Sample_Index + u_Pro_Config.n_Sentence_Start_Index;
  90. n_Recog_Result = DHMM_Recog_Basic(sz_Feature_File_Name, Sen_Index, pd_DHMM_Model_Probably);
  92. n_True_Result = n_Word_Sample_Index % WORD_ITEM_NUM;
  94. if (u_Pro_Config.WATCH_RECOG_RESULT == 1)
  95. {
  96. DHMM_LLS_Ret(pd_DHMM_Model_Probably, Total_Model_Num, n_True_Result, n_Recog_Result);
  97. }
  99. // 统计识别正确性
  100. d2dna_Recog_Table[n_True_Result][n_Recog_Result]++;
  102. if (n_Recog_Result == n_True_Result)
  103. n_Recog_Result_Good_Count++;
  104. else
  105. {
  106. n_Recog_Result_Bad_Count++;
  107. }
  108. }
  109. PRO_LOG("tNumber:%4d person, recognition %4d right!r", n_Person_Index, n_Recog_Result_Good_Count);
  110.   }
  112. // 计算识别率
  113. u_Pro_Config.d_Recognition_Accurate_Rate = n_Recog_Result_Good_Count / double(n_Recog_Result_Good_Count + n_Recog_Result_Bad_Count) * 100;
  114. PRO_LOG("nAccurate Rate = %d / %d = %10.4f%%.n", n_Recog_Result_Good_Count, (n_Recog_Result_Good_Count + n_Recog_Result_Bad_Count), u_Pro_Config.d_Recognition_Accurate_Rate);
  116. /**/
  117. PRO_LOG("  Recognition Table:n");
  118. PRO_LOG("t");
  119. for (int nTmp = 0; nTmp < Total_Model_Num; nTmp++)
  120. PRO_LOG("%2d  ", nTmp);
  121. PRO_LOG("n");
  122. PRO_LOG("t-------------------------------------------n");
  123. for (n_Sentence_Index = 0; n_Sentence_Index < u_Pro_Config.n_DHMM_Model_Num; n_Sentence_Index++)
  124. {
  125. PRO_LOG("t");
  126. for (nTmp = 0; nTmp < Total_Model_Num; nTmp++)
  127. PRO_LOG("%2d  ", d2dna_Recog_Table[n_Sentence_Index][nTmp]);
  128. PRO_LOG("n");
  129. }
  130. PRO_LOG("n");
  132. d2dna_Free(d2dna_Recog_Table, u_Pro_Config.n_DHMM_Model_Num, Total_Model_Num);
  133. delete[] pd_DHMM_Model_Probably;
  135. return 0;
  136. }
  139. int DHMM_LLS_Ret(double* pd_DHMM_Model_Probably, const int Total_Model_Num, const int n_True_Result, const int n_Recog_Result)
  140. {
  141. char RetFile[100];
  142. strcpy(RetFile,"..\..\data\Recog_Ret\Kitty_Recog_LLS_Ret.txt");
  143. FILE* fp;
  144. fp = fopen(RetFile,"a+");
  145. ASSERT(fp!=NULL);
  147. for (int n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  148. {
  149. fprintf(fp, "%7.0f, ", pd_DHMM_Model_Probably[n_Model_Index]);
  150. }
  152. fprintf(fp, "  %2d ", n_Recog_Result);
  153. fprintf(fp, "%2d ", n_True_Result);
  154. fclose(fp);
  155. return 0;
  156. }
  158. //////////////////////////////////////////////////////////////////////
  159. // 函数名称：DHMM_Recog_Basic
  160. // 函数功能：基本识别（一次识别）
  161. // 函数性质：API
  162. // 输入参数：
  163. // sz_Feature_File_Name，待识别说话人的特征文件名
  164. // n_Word_Sample_Index，待识别词条标号
  165. // 输出参数：
  166. // 无
  167. // 返回值：
  168. // 识别结果
  169. // 备注：仅仅使用Viterbi或前后向，比较似然分数
  170. int DHMM_Recog_Basic(char * sz_Feature_File_Name, int n_Word_Sample_Index, double* pd_DHMM_Model_Probably)
  171. {
  172. int nRetCode;
  173. DYNA_2DIM_DOUBLE_ARRAY d2dda_Code_Book;
  174. DHMM_MODEL * pu_DHMM_Model;
  175. WORD_SAMPLE u_Word_Sample;
  176. //double * pd_DHMM_Model_Probably;
  177. DYNA_2DIM_INT_ARRAY d2dna_DHMM_Model_Sequence;
  178. int n_Model_Index;
  179. int n_Recog_Result;
  180. double d_Recog_Result_Probably;
  181. static int n_file_count = 0;
  182. int Total_Model_Num;
  184. Total_Model_Num =u_Pro_Config.n_DHMM_Model_Num;
  186. // 读入码书
  187. d2dda_Code_Book = d2dda_New(u_Pro_Config.n_VQ_Code_Book_Size, u_Pro_Config.n_Feature_Dim);
  188. ASSERT(d2dda_Code_Book != NULL);
  189. nRetCode = DHMM_VQ_Load_Code_Book_File(u_Pro_Config.sz_Toload_Code_Book_File_Name,
  190. d2dda_Code_Book, u_Pro_Config.n_VQ_Code_Book_Size, u_Pro_Config.n_Feature_Dim);
  191. ASSERT(nRetCode == 0);
  193. if ((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) == (MODEL_CONFIG_TRAIN_WITH_SILENCE_MODEL))
  194. u_Pro_Config.n_DHMM_Model_State_Num += 2;
  195. else if((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) ==  MODEL_CONFIG_LOAD_WITH_SILENCE_MODEL)
  196. u_Pro_Config.n_DHMM_Model_State_Num += 2;
  198. // 为各个词条的模型准备内存，并读入模型
  199. pu_DHMM_Model = new DHMM_MODEL[Total_Model_Num];
  200. ASSERT(pu_DHMM_Model);
  201. for (n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  202. {
  203. pu_DHMM_Model[n_Model_Index].n_State_Num = u_Pro_Config.n_DHMM_Model_State_Num;
  204. pu_DHMM_Model[n_Model_Index].n_Code_Book_Size = u_Pro_Config.n_VQ_Code_Book_Size;
  205. pu_DHMM_Model[n_Model_Index].pdPi = new double[u_Pro_Config.n_DHMM_Model_State_Num];
  206. pu_DHMM_Model[n_Model_Index].d2dda_A = d2dda_New(u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_DHMM_Model_State_Num);
  207. pu_DHMM_Model[n_Model_Index].d2dda_B = d2dda_New(u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_VQ_Code_Book_Size);
  208. ASSERT((pu_DHMM_Model[n_Model_Index].pdPi != NULL)
  209. && (pu_DHMM_Model[n_Model_Index].d2dda_A != NULL)
  210. && (pu_DHMM_Model[n_Model_Index].d2dda_B != NULL));
  211. }
  212. if ((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) == MODEL_CONFIG_LOAD_WITH_SILENCE_MODEL ||
  213. (u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) == MODEL_CONFIG_TRAIN_WITH_SILENCE_MODEL)
  214. nRetCode = DHMM_Model_Load_DHMM_Model_File_With_Silence(u_Pro_Config.sz_Toload_DHMM_Model_File_Name, pu_DHMM_Model, Total_Model_Num);
  215. ASSERT(nRetCode == 0);
  217. // 准备该词条的内存空间
  218. u_Word_Sample.n_Feature_Sequence_Len = dfa_Feature_Get_Sentence_Frame_Num(sz_Feature_File_Name, n_Word_Sample_Index);
  219. u_Word_Sample.n_Feature_Dim = u_Pro_Config.n_Feature_Dim;
  221. int word_len = u_Word_Sample.n_Feature_Sequence_Len;
  222. //ASSERT(word_len >= 0);
  223. if(word_len <= 0)
  224. {
  225. // 释放模型内存，释放码书内存
  226. for (n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  227. {
  228. delete[] pu_DHMM_Model[n_Model_Index].pdPi;
  229. d2dda_Free(pu_DHMM_Model[n_Model_Index].d2dda_A, u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_DHMM_Model_State_Num);
  230. d2dda_Free(pu_DHMM_Model[n_Model_Index].d2dda_B, u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_VQ_Code_Book_Size);
  231. }
  232. if ((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) == (MODEL_CONFIG_TRAIN_WITH_SILENCE_MODEL))
  233. u_Pro_Config.n_DHMM_Model_State_Num -= 2;
  234. else if((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) ==  MODEL_CONFIG_LOAD_WITH_SILENCE_MODEL)
  235. u_Pro_Config.n_DHMM_Model_State_Num -= 2;
  237. delete[] pu_DHMM_Model;
  238. d2dda_Free(d2dda_Code_Book, u_Pro_Config.n_VQ_Code_Book_Size, u_Pro_Config.n_Feature_Dim);
  239. return 0;
  240. }
  242. u_Word_Sample.d2dda_Feature_Sequence =
  243. d2dda_New(u_Word_Sample.n_Feature_Sequence_Len, u_Word_Sample.n_Feature_Dim);
  245. u_Word_Sample.pn_VQed_Feature_Sequence = 
  246. new int[u_Word_Sample.n_Feature_Sequence_Len];
  247. ASSERT((u_Word_Sample.d2dda_Feature_Sequence  != NULL)
  248. && (u_Word_Sample.pn_VQed_Feature_Sequence != NULL));
  250. // 读入词条特征
  251. nRetCode = dfa_Feature_Read_A_Sentence(sz_Feature_File_Name, n_Word_Sample_Index, u_Pro_Config.n_Feature_Dim,
  252. u_Word_Sample.d2dda_Feature_Sequence);
  253. ASSERT(nRetCode == u_Word_Sample.n_Feature_Sequence_Len);
  255. // VQ特征
  256. nRetCode = DHMM_VQ_Encode_A_Word_Sample(d2dda_Code_Book, u_Pro_Config.n_VQ_Code_Book_Size, u_Pro_Config.n_Feature_Dim,
  257. &u_Word_Sample);
  258. ASSERT(nRetCode == 0);
  260. int fea_num = u_Word_Sample.n_Feature_Dim;
  262. // 释放原始特征参数空间
  263. d2dda_Free(u_Word_Sample.d2dda_Feature_Sequence, u_Word_Sample.n_Feature_Sequence_Len, u_Word_Sample.n_Feature_Dim);
  265. // 为匹配分数准备内存
  266. //pd_DHMM_Model_Probably = new double[u_Pro_Config.n_DHMM_Model_Num];
  267. d2dna_DHMM_Model_Sequence = d2dna_New(Total_Model_Num, u_Word_Sample.n_Feature_Sequence_Len);
  268. ASSERT((pd_DHMM_Model_Probably != NULL) && (d2dna_DHMM_Model_Sequence != NULL));
  270. // 计算该词条与每一个模型的匹配分数
  271. for (n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  272. {
  273. if ((u_Pro_Config.l_DHMM_Recog_Config & RECOG_CONFIG_METHOD_MASK) == RECOG_CONFIG_METHOD_VITERBI_ONLY)
  274. {
  275. nRetCode = DHMM_Recog_Viterbi(&pu_DHMM_Model[n_Model_Index], &u_Word_Sample, &pd_DHMM_Model_Probably[n_Model_Index], d2dna_DHMM_Model_Sequence[n_Model_Index]);
  276. ASSERT(nRetCode == 0);
  277. }
  278. }
  280. // 选出最佳匹配
  281. n_Recog_Result = -1;
  282. d_Recog_Result_Probably = -MAX_DOUBLE_VALUE;
  283. for (n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  284. {
  285. pd_DHMM_Model_Probably[n_Model_Index] /= word_len;
  286. if (pd_DHMM_Model_Probably[n_Model_Index] > d_Recog_Result_Probably)
  287. {
  288. n_Recog_Result = n_Model_Index;
  289. d_Recog_Result_Probably = pd_DHMM_Model_Probably[n_Model_Index];
  290. }
  291. }
  293. // 释放词条VQ后特征参数占用空间
  294. delete[] u_Word_Sample.pn_VQed_Feature_Sequence;
  296. // 释放匹配分数占用空间
  297. //delete[] pd_DHMM_Model_Probably;
  298. d2dna_Free(d2dna_DHMM_Model_Sequence, Total_Model_Num, u_Word_Sample.n_Feature_Sequence_Len);
  300. // 释放模型内存，释放码书内存
  301. for (n_Model_Index = 0; n_Model_Index < Total_Model_Num; n_Model_Index++)
  302. {
  303. delete[] pu_DHMM_Model[n_Model_Index].pdPi;
  304. d2dda_Free(pu_DHMM_Model[n_Model_Index].d2dda_A, u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_DHMM_Model_State_Num);
  305. d2dda_Free(pu_DHMM_Model[n_Model_Index].d2dda_B, u_Pro_Config.n_DHMM_Model_State_Num, u_Pro_Config.n_VQ_Code_Book_Size);
  306. }
  307. if ((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) == (MODEL_CONFIG_TRAIN_WITH_SILENCE_MODEL))
  308. u_Pro_Config.n_DHMM_Model_State_Num -= 2;
  309. else if((u_Pro_Config.l_DHMM_Model_Config & MODEL_CONFIG_GENERATE_DHMM_MODEL_MASK) ==  MODEL_CONFIG_LOAD_WITH_SILENCE_MODEL)
  310. u_Pro_Config.n_DHMM_Model_State_Num -= 2;
  312. delete[] pu_DHMM_Model;
  313. d2dda_Free(d2dda_Code_Book, u_Pro_Config.n_VQ_Code_Book_Size, u_Pro_Config.n_Feature_Dim);
  315. return n_Recog_Result;
  316. }
  320. //////////////////////////////////////////////////////////////////////
  321. // 函数名称：DHMM_Recog_Viterbi
  322. // 函数功能：计算Viterbi方法的最大似然值
  323. // 函数性质：API
  324. // 输入参数：
  325. // pu_DHMM_Model，DHMM模型
  326. // pu_Word_Sample，要计算的词
  327. // 输出参数：
  328. // pd_Max_Likelihood，存放计算的似然值
  329. // pn_Status_Sequence，存放各帧所经历的状态
  330. // 返回值：
  331. // 0 表示成功
  332. // 备注：该函数是一个分发函数，按照u_Pro_Config.l_DHMM_Recog_Config配置的信息
  333. // 将具体调用不同的函数体
  334. int DHMM_Recog_Viterbi(DHMM_MODEL * pu_DHMM_Model,
  335.    WORD_SAMPLE * pu_Word_Sample,
  336.    double * pd_Max_Likelihood, int * pn_Status_Sequence)
  337. {
  338. switch (u_Pro_Config.l_DHMM_Recog_Config & RECOG_CONFIG_RECOG_PROCEDURE_MASK)
  339. {
  340. case RECOG_CONFIG_RECOG_PROCEDURE_STD:
  341. ASSERT(0);
  342. break;
  343. case RECOG_CONFIG_RECOG_PROCEDURE_GL:
  344. ASSERT(0);
  345. break;
  346. case RECOG_CONFIG_RECOG_PROCEDURE_LHS:
  347. return DHMM_Recog_Viterbi_LHS(pu_DHMM_Model, pu_Word_Sample, pd_Max_Likelihood, pn_Status_Sequence);
  348. break;
  349. case RECOG_CONFIG_RECOG_PROCEDURE_HQ:
  350. return DHMM_Recog_Viterbi_HQ(pu_DHMM_Model, pu_Word_Sample, pd_Max_Likelihood, pn_Status_Sequence);
  351. break;
  352. case RECOG_CONFIG_RECOG_PROCEDURE_WP:
  353. ASSERT(0);
  354. break;
  355. default:
  356. ASSERT(0);
  357. break;
  358. }
  360. return 0;
  361. }
  363. //////////////////////////////////////////////////////////////////////
  364. // 函数名称：DHMM_Recog_Forward_Backward
  365. // 函数功能：计算前后向算法的最大似然值
  366. // 函数性质：API
  367. // 输入参数：
  368. // pu_DHMM_Model，DHMM模型
  369. // pu_Word_Sample，要计算的词
  370. // 输出参数：
  371. // pd_Max_Likelihood，存放计算的似然值
  372. // 返回值：
  373. // 0 表示成功
  374. // 备注：该函数是一个分发函数，按照u_Pro_Config.l_DHMM_Recog_Config配置的信息
  375. // 将具体调用不同的函数体
  376. int DHMM_Recog_Forward_Backward(DHMM_MODEL * pu_DHMM_Model,
  377. WORD_SAMPLE * pu_Word_Sample,
  378. double * pd_Max_Likelihood)
  379. {
  380. switch (u_Pro_Config.l_DHMM_Recog_Config & RECOG_CONFIG_RECOG_PROCEDURE_MASK)
  381. {
  382. case RECOG_CONFIG_RECOG_PROCEDURE_STD:
  383. ASSERT(0);
  384. break;
  385. case RECOG_CONFIG_RECOG_PROCEDURE_GL:
  386. return DHMM_Recog_Forward_Backward_GL(pu_DHMM_Model, pu_Word_Sample, pd_Max_Likelihood);
  387. break;
  388. case RECOG_CONFIG_RECOG_PROCEDURE_LHS:
  389. ASSERT(0);
  390. break;
  391. case RECOG_CONFIG_RECOG_PROCEDURE_HQ:
  392. return DHMM_Recog_Forward_Backward_HQ(pu_DHMM_Model, pu_Word_Sample, pd_Max_Likelihood);
  393. break;
  394. case RECOG_CONFIG_RECOG_PROCEDURE_WP:
  395. ASSERT(0);
  396. break;
  397. default:
  398. ASSERT(0);
  399. break;
  400. }
  402. return 0;
  403. }