2.1 ǰ¶ËÐźŴ¦Àí
2.1.1 ÓïÒô¼ì²â£¨VAD£©
ÓïÒô¼ì²â£¨Ó¢ÎÄÒ»°ã³ÆÎª Voice Activity Detection£¬VAD£©µÄÄ¿±êÊÇ£¬×¼È·µÄ¼ì²â³öÒôƵÐźŵÄÓïÒô¶ÎÆðʼλÖ㬴Ӷø·ÖÀë³öÓïÒô¶ÎºÍ·ÇÓïÒô¶Î£¨¾²Òô»òÔëÉù£©Ðźš£ÓÉÓÚÄܹ»Â˳ý²»Ïà¸É·ÇÓïÒôÐźţ¬¸ßЧ׼ȷµÄ VAD ²»µ«ÄܼõÇáºóÐø´¦ÀíµÄ¼ÆËãÁ¿£¬Ìá¸ßÕûÌåʵʱÐÔ£¬»¹ÄÜÓÐЧÌá¸ßÏÂÓÎËã·¨µÄÐÔÄÜ¡£
VAD Ëã·¨¿ÉÒÔ´ÖÂԵķÖΪÈýÀࣺ»ùÓÚãÐÖµµÄ VAD¡¢×÷Ϊ·ÖÀàÆ÷µÄ VAD¡¢Ä£ÐÍ VAD¡£
»ùÓÚãÐÖµµÄ VAD£ºÍ¨¹ýÌáȡʱÓò£¨¶ÌʱÄÜÁ¿¡¢¶ÌÆÚ¹ýÁãÂʵȣ©»òƵÓò£¨MFCC¡¢Æ×ìØµÈ£©ÌØÕ÷£¬Í¨¹ýºÏÀíµÄÉèÖÃÃÅÏÞ£¬´ïµ½Çø·ÖÓïÒôºÍ·ÇÓïÒôµÄÄ¿µÄ¡£ÕâÊÇ´«Í³µÄ VAD ·½·¨¡£
×÷Ϊ·ÖÀàÆ÷µÄ VAD£º¿ÉÒÔ½«ÓïÒô¼ì²âÊÓ×÷ÓïÒô/·ÇÓïÒôµÄÁ½·ÖÀàÎÊÌ⣬½ø¶øÓûúÆ÷ѧϰµÄ·½·¨ÑµÁ··ÖÀàÆ÷£¬´ïµ½¼ì²âÓïÒôµÄÄ¿µÄ¡£
Ä£ÐÍ VAD£º¿ÉÒÔÀûÓÃÒ»¸öÍêÕûµÄÉùѧģÐÍ£¨½¨Ä£µ¥ÔªµÄÁ£¶È¿ÉÒԺܴ֣©£¬ÔÚ½âÂëµÄ»ù´¡£¬Í¨¹ýÈ«¾ÖÐÅÏ¢£¬ÅбðÓïÒô¶ÎºÍ·ÇÓïÒô¶Î¡£
VAD ×÷ΪÕû¸öÁ÷³ÌµÄ×îǰ¶Ë£¬ÐèÒªÔÚ±¾µØÊµÊ±µÄÍê³É¡£ÓÉÓÚ¼ÆËã×ÊÔ´·Ç³£ÓÐÏÞ£¬Òò´Ë£¬VAD Ò»°ã»á²ÉÓÃãÐÖµ·¨ÖÐijÖÖËã·¨£»¾¹ý¹¤³ÌÓÅ»¯µÄ·ÖÀà·¨Ò²¿ÉÄܱ»ÀûÓã»¶øÄ£ÐÍ VAD ĿǰÄÑÒÔÔÚ±¾µØ²¿ÊðÓ¦Óá£
2.1.2 ½µÔë
ʵ¼Ê»·¾³ÖдæÔÚ×ſյ÷¡¢·çÉÈÒÔ¼°ÆäËû¸÷ÖÖ¸÷ÑùµÄÔëÉù¡£½µµÍÔëÉù¸ÉÈÅ£¬Ìá¸ßÐÅÔë±È£¬½µµÍºó¶ËÓïÒôʶ±ðµÄÄѶȡ£
³£ÓõĽµÔëËã·¨ÓÐ ×ÔÊÊÓ¦ LMS ºÍάÄÉÂ˲¨µÈ¡£
2.1.3 Éùѧ»ØÉùÏû³ý£¨Acoustic Echo Cancellaction, AEC£©
AECÒ²ÊÇÒ»ÖÖ³£¼ûµÄ¼¼Êõ£¬ÔÚÓïÒôͨ»°ÖУ¬AECÊDZز»¿ÉÉٵĻù´¡¼¼Êõ¡£
ͼ11. AEC ʾÒ⡾À´Ô´¡¿
¾ßÌåµÄ£¬AEC µÄÄ¿µÄÊÇ£¬ÔÚÒôÏäÑïÉùÆ÷¹¤×÷£¨²¥·ÅÒôÀÖ»òÓïÒô£©Ê±£¬´ÓÂó¿Ë·çÖÐÊÕ¼¯µÄÓïÒôÖУ¬È¥³ý×ÔÉí²¥·ÅµÄÉùÒôÐźš£ÕâÊÇË«¹¤Ä£Ê½µÄǰÌá¡£·ñÔò£¬µ±ÒôÀÖ²¥·Åʱ£¬ÎÒÃǵÄÉùÒôÐźŻáÑÍûÔÚÒôÀÖÉùÖУ¬²»ÄܼÌÐø¶ÔÒôÏä½øÐÐÓÐЧµÄÓïÒô¿ØÖÆ¡£
2.1.4 È¥»ìÏì´¦Àí
ÔÚÊÒÄÚ£¬ÓïÒô»á±»Ç½±ÚµÈ¶à´Î·´É䣬Âó¿Ë·ç²É¼¯µ½£¨Í¼12£©¡£»ìÏì¶ÔÓÚÈ˶úÍêÈ«²»ÊÇÎÊÌ⣬µ«ÊÇ£¬ÑÓ³ÙµÄÓïÒôµþ¼Ó²úÉúÑÚ±ÎЧӦ£¬Õâ¶ÔÓïÒôʶ±ðÊÇÖÂÃüµÄÕϰ¡£
ͼ12. »ìÏ졾À´Ô´¡¿
¶ÔÓÚ»ìÏ죬һ°ã´ÓÁ½¸ö·½ÃæÀ´³¢ÊÔ½â¾ö£º1£©È¥»ìÏì 2£©¶ÔÓïÒôʶ±ðµÄÉùѧģÐͼӻìÏìѵÁ·¡£ÓÉÓÚÕæÊµ»·¾³µÄ¸´ÔÓÐÔ£¬Ò»¶¨µÄǰ¶ËÈ¥»ìÏìËã·¨»¹ÊǷdz£ÓбØÒªµÄ¡£
2.1.5 ÉùÔ´¶¨Î»£¨Direction of Arrival estimation, DOA£©
ÉùÔ´¶¨Î»ÊǸù¾ÝÂóÁÐÊÕ¼¯µÄÉùÒôÓȷ¶¨Ëµ»°È˵ÄλÖá£DOA ÖÁÉÙÓÐÁ½¸öÓÃ;£¬1£©ÓÃÓÚ·½Î»µÆµÄչʾ£¬ÔöÇ¿½»»¥Ð§¹û£»2£©×÷Ϊ²¨ÊøÐγɵÄǰµ¼ÈÎÎñ£¬È·¶¨¿Õ¼äÂ˲¨µÄ²ÎÊý¡£
ÉùÔ´¶¨Î»ÓÐÈçϳ£Ó÷½·¨ÓлùÓÚ²¨ÊøÉ¨ÃèµÄÉùÔ´¶¨Î»¡¢»ùÓÚÆð·Ö±æÂÊÂÊÆ×¹À¼ÆµÄÉùÔ´¶¨Î»ÒÔ¼° »ùÓÚµ½´ïʱ¼ä²î£¨Time Difference of Arrival, TDOA£©µÄÉùÔ´¶¨Î»¡£¿¼Âǵ½Ëã·¨¸´ÔÓÐÔºÍÑÓʱ£¬Ò»°ã²ÉÓÃTDOA·½·¨¡£
2.1.6 ²¨ÊøÐγɣ¨Beam Forming, BF£©
²¨ÊøÐγÉÊÇÀûÓÿռäÂ˲¨µÄ·½·¨£¬½«¶à·ÉùÒôÐźţ¬ÕûºÏΪһ·Ðźš£Í¨¹ý²¨ÊøÐγɣ¬Ò»·½Ãæ¿ÉÒÔÔöÇ¿ÔʼµÄÓïÒôÐźţ¬ÁíÒ»·½ÃæÒÖÖÆÅÔ·Ðźţ¬Æðµ½½µÔëºÍÈ¥»ìÏìµÄ×÷Óã¨Í¼13£©¡£
ͼ13. ²¨ÐγÉʾÒâͼ¡¾À´Ô´¡¿
2.2 »½ÐÑ
³öÓÚ±£»¤Óû§Òþ˽ºÍ¼õÉÙÎóʶ±ðÁ½¸öÒòËØµÄ¿¼ÂÇ£¬ÖÇÄÜÒôÏäÒ»°ãÔÚ¼ì²âµ½»½ÐÑ´ÊÖ®ºó£¬²Å»á¿ªÊ¼½øÒ»²½µÄ¸´ÔÓÐźŴ¦Àí£¨ÉùÔ´¶¨Î»¡¢²¨ÊøÐγɣ©ºÍºóÐøµÄÓïÒô½»»¥¹ý³Ì¡£
Ò»°ã¶øÑÔ£¬»½»½ÐÑÄ£¿éÊÇÒ»¸öСÐÍÓïÒôʶ±ðÒýÇæ¡£ÓÉÓÚÄ¿±êµ¥Ò»£¨¼ì²â ³öÖ¸¶¨µÄ»½ÐÑ´Ê£©£¬»½ÐÑÖ»ÐèÒª½ÏСµÄÉùѧģÐͺÍÓïÑÔÄ£ÐÍ£¨Ö»Ðè񻂿·Ö³öÓÐÎÞ»½ÐѴʳöÏÖ£©£¬Éùѧ´ò·ÖºÍ½âÂë¿ÉÒԺܿ죬¿Õ¼äÕ¼ÓÃÉÙ£¬Äܹ»ÔÚ±¾µØÊµÊ±¡£
Ò²Óл½ÐÑ×öΪ¹Ø¼ü´Ê¼ìË÷£¨key word search£©»òÎı¾Ïà¹ØµÄÉùÎÆÊ¶±ðÎÊÌâÀ´½â¾ö¡£
2.3 ÓïÒô½»»¥
ÓïÒô½»»¥µÄ»ù±¾Á÷³ÌÈçͼ16Ëùʾ¡£ÏÂÃæ·Ö±ð¶Ô¸÷¸ö»·½Ú½øÐмòÒª½éÉÜ¡£
ͼ14. ÓïÒô½»»¥»ù±¾Á÷³Ì
ICASSP¡¯17 ¹ØÓÚ¶Ô»°ÏµÍ³µÄ tutorial £¬²¢¸½ÓÐϵͳÐԵIJο¼ÎÄÏס£
2.3.1 ÓïÒôʶ±ð£¨Automatic Speech Recognition, ASR£©
ÓïÒôʶ±ðµÄÄ¿µÄÊǽ«ÓïÒôÐźÅת»¯ÎªÎı¾¡£ÓïÒôʶ±ð¼¼ÊõÏà¶Ô³ÉÊ졣Ŀǰ£¬»ùÓÚ½ü³¡Ðźŵġ¢Êܿػ·¾³£¨µÍÔëÉù¡¢µÍ»ìÏ죩ϵıê×¼ÒôÓïÒôʶ±ðÄܹ»´ïµ½ºÜµÄˮƽ¡£È»¶øÔÚÖÇÄÜÒôÏ俪·ÅÐÔµÄÕæÊµ»·¾³£¬ÓïÒôʶ±ðÒÀÈ»ÊÇÒ»¸ö²»Ð¡µÄÌôÕ½£¬ÐèÒª½ÓºÏǰ¶ËÐźŴ¦ÀíÒ»ÆðÀ´ÓÅ»¯¡£
2.3.2 ×ÔÈ»ÓïÑÔÀí½â£¨Natural Language Understanding, NLU£©
NLU ×÷Ϊһ¸öÑо¿¿ÎÌ⻹ԶûÓб»½â¾ö¡£µ«ÊÇÔÚÏÞ¶¨ÁìÓòÏ£¬½áºÏÁ¼ºÃµÄ²úÆ·Éè¼Æ£¬ÎÒÃÇ»¹ÊÇÄܹ»ÀûÓÃÏÖÓм¼Êõ£¬×ö³öʵÓõIJúÆ·¡£
¿ÉÒÔ½«»ùÓÚ¿ò¼ÜµÄ£¨frame-based£© NLU ·ÖΪÈý¸ö×ÓÎÊÌâÈ¥½â¾ö£¨Í¼15£©£º
* ÁìÓò·ÖÀࣺʶ±ð³öÓû§ÃüÁîËùÊôÁìÓò¡£ÆäÖУ¬ÁìÓòÊÇÔ¤ÏÈÉè¼ÆµÄ·â±Õ¼¯ºÏ£¨Èç²úÆ·Éè¼ÆÉÏ£¬ÒôÏäÖ»Ö§³ÖÒôÀÖ¡¢ÌìÆøµÈÁìÓò£©£¬¶øÃ¿¸öÁìÓò¶¼Ö»Ö§³ÖÎÞÏÞÔ¤ÉèµÄ²éѯÄÚÈݺͽ»»¥·½Ê½¡£
* Òâͼ·ÖÀࣺÔÚÏàÓ¦ÁìÓò£¬Ê¶±ðÓû§µÄÒâͼ£¨Èç²¥·ÅÒôÀÖ¡¢ÔÝÍ£»òÇл»µÈ£©¡£ÒâͼÍùÍù¶ÔÓ¦×Åʵ¼ÊµÄ²Ù×÷¡£
* ʵÌå³éÈ¡£¨²ÛÌî³ä£©£ºÈ·¶¨Òâͼ£¨²Ù×÷£©µÄ²ÎÊý£¨ÈçÈ·¶¨£¬¾ßÌåÊDz¥·ÅÄÄÊ׸è»òÄÄλ¸èÊֵĸèÇú£©¡£
ͼ15. »ùÓÚ¿ò¼ÜµÄ×ÔÈ»ÓïÑÔÀí½â
2.3.3 ¶Ô»°¹ÜÀí£¨Diaglou Management, DM£©
¶àÂÖ¶Ô»°¶ÔÓÚ×ÔÈ»µÄÈ˹¤½»»¥·Ç³£ÖØÒª¡£±ÈÈ磬µ±ÎÒÃÇѯÎÊ¡°±±¾©Ã÷ÌìµÄÌìÆøÔõô£¿¡±£¬Ö®ºó£¬¸üϰ¹ß×·ÎÊ¡°ÄÇÉîÛÚÄØ£¿¡±¶ø²»ÊÇÖØ¸´µÄ˵¡±**ÉîÛÚÃ÷ÌìµÄÌìÆøÔõô£¿**¡°
ÔÚ NLU ÎÞÓеõ½ºÜºÃ½â¾öµÄÇé¿öÏ£¬¶Ô»°¹ÜÀíËÆºõ²»¿ÉÄÜ¡£ºÃÔÚÏÞ·¶Î§Ï£¬½áºÏ²úÆ·Éè¼Æ£¬»¹ÊÇÄÜ×öµÄ²»´í¡£Ò»°ãµÄ×÷·¨ÊÇ£¬½«ÂÖ¶Ô»°½âÎö³öµÄ²ÎÊý×öΪÉÏÏÂÎÄ£¨È«¾Ö±äÁ¿£©£¬´øÈëµ½ÏÂÒ»ÂÖ¶Ô»°£»µ±Ç°ÂÖ¶Ô»°£¬¸ù¾ÝÒ»¶¨µÄÌõ¼þÅжϣ¬ÊÇ·ñ±£³ÖÔÚÉÏÒ»ÂÖµÄÁìÓò£¬ÊÇ·ñÇå¿ÕÉÏÏÂÎÄ¡£
²»Í¬ÓÚ´¿´âµÄÁÄÌì»úÆ÷µÄ¶Ô»°¹ÜÀí£¬ÖÇÄÜÒôÏäµÄ¶Ô»°¹ÜÀí»¹ÓÐʵ¼ÊµÄ²Ù×÷¹¦ÄÜ£¨²éѯÐÅÏ¢¡¢Ìṩ¿ØÖÆÖ¸Á¡£
2.3.4 ×ÔÈ»ÓïÑÔÉú³É£¨Natural Language Generation, NLG£©
ĿǰÍêÈ«×Ô¶¯»¯µÄ NLG ·½·¨»¹²»³ÉÊ졣ʵ¼Ê²úÆ·ÖУ¬¶à²ÉÓÃÔ¤ÏÈÉè¼ÆµÄÎı¾Ä£°åÀ´Éú³ÉÎı¾Êä³ö¡£±ÈÈ磬²¥·Å¸èÇúʱ£¬Éú³ÉÓï¾äΪ£º¡°¼´½«ÎªÄú²¥·Å¡¾¸èÊÖÃû¡¿µÄ¡¾¸èÇúÃû¡¿¡±¡£
2.3.5 ÓïÒôºÏ³É£¨Speech Synthesis£©
ÓïÒôºÏ³ÉÓÖ½Ð×öÎÄÓïת»»£¨Text-to-Speech£¬TTS£©£¬¸ü³£¼û¿ÉÄÜÊÇ TTS ÕâÒ»³Æºô¡£TTS µÄÖÕ¼«Ä¿±êÊÇ£¬Ê¹»úÆ÷Äܹ»ÏñÈËÒ»ÑùÀʶÁÈÎÒâ¸ø¶¨µÄÎı¾¡£
ÆÀ¼ÛʵÓõÄÓïÒôºÏ³ÉϵͳµÄÁ½¸öÖ÷ÒªµÄ±ê×¼ÊÇ1£©¿É¶®¶È£¨ÈËÄܹ»Ìý¶®£©ºÍ2£©×ÔÈ»¶È£¨Ê¹ÈËÌý×ÅÊæ·þ£©¡£Ä¿Ç°£¬¿É¶®¶ÈµÄÎÊÌâ»ù±¾µÃµ½½â¾ö¡£²ÎÊýºÏ³ÉºÍÆ´½ÓºÏ³ÉÊÇTTSµÄÁ½ÖÖÖ÷ÒªºÏ³É·½·¨£¬ÆäÖУ¬²ÎÊý¼ÆËãÁ¿Ð¡£¬²¿ÊðÁé»î£¬µ«×ÔÈ»½Ï²î£»Æ´½Ó½Ó½üÕæÈË·¢Òô£¬´æ´¢ºÍ¼ÆËã×ÊÔ´¸ß£¬Ò»°ãÖ»ÄÜÔÚÏߺϳɡ£ÀýÈ磬Echo ²ÉÓõĻùÓÚµ¥ÔªÑ¡Ôñ£¨unit selection£©µÄÆ´½ÓºÏ³É¡£
2.4 ÆäËû¼¼Êõ
×îºó£¬ÎÒÃǼòµ¥ÁоÙһЩÏà¶Ô³ÉÊ죬µ«»¹Ã»Óй㷺ӦÓÃÓÚÖÇÄÜÒôÏäµÄ¼¼Êõ¡£
ÉùÎÆÊ¶±ð
ÉùÎÆÊ¶±ðÊǾÝÓïÒô²¨Ðη´Ó³Ëµ»°ÈËÉúÀíºÍÐÐÎªÌØÕ÷µÄÓïÒô²ÎÊý£¬×Ô¶¯Ê¶±ð˵»°ÈËÉí·ÝµÄÒ»Ïî¼¼Êõ¡£Î¢ÐÅÖеÄÉùÒôËø¾ÍÊÇÉùÎÆ¼¼ÊõµÄÒ»Ïî¾ßÌåÓ¦Óá£
ͨ¹ýÉùÎÆÊ¶±ð£¬¿ÉÒÔÉè¼Æ³ö¸ü¼Ó¸öÐÔ»¯µÄ·þÎñ¡£
ÈËÁ³¼ì²â
Èç¹ûÒôÏäÅäÖÃΪÉãÏñÍ·£¬¿ÉÒÔͨÈËÁ³¼ì²â£¬È·¶¨Óû§µÄλÖá£Ò»·½Ãæ¿ÉÒÔÓиüºÃµÄ½»»¥Éè¼Æ£¬ÁíÒ»·½Ãæ¿ÉÒÔ¸¨ÖúÉùÔ´¶¨Î»¡£
ÈËÁ³Ê¶±ð
ͬÉùÎÆÊ¶±ðÀàËÆ£¬ÈËÁ³Ê¶±ðÒ²¿ÉÒÔÓÃÀ´È·¶¨Óû§µÄÉí·Ý¡£
|