本文分享自華為雲社群《unimrcp對接華為雲ASR（Linux版）》，作者： ASR-beginer。
本篇文章提供了unimrcp對接華為雲ASR的保姆級教程，根據第一到四章，可從頭逐步編譯+整合基於華為雲ASR的unimrcp系統（授人以漁）。同時，本文第五章（直接給條魚）提供了作者修改好的原始碼，直接一鍵編譯即可。
一、安裝unimrcp
1.提前安裝必要的依賴項（系統級）
2.下載unimrcp
3.下載unimrcp的依賴庫，並安裝
進入https://www.unimrcp.org/downloads/dependencies，直接選擇最新版本下載
解壓後，進入unimrcp-deps-1.6.0，執行安裝命令：
./build-dep-libs.sh
4.安裝unimrcp
之後可以在/usr/local/unimrcp 路徑下，看到以下內容：
5.ASR測試
先啟動服務端./unimrcpserver
(base) [root@asr-build bin]# ./unimrcpserver 2024-03-09 10:23:39:729352 [NOTICE] UniMRCP Server [1.8.0] 2024-03-09 10:23:39:729427 [INFO] APR [1.5.2] 2024-03-09 10:23:39:729436 [NOTICE] Create MRCP Server 2024-03-09 10:23:39:729459 [NOTICE] Open Config File [/usr/local/unimrcp/conf/unimrcpserver.xml] 2024-03-09 10:23:39:729874 [INFO] Set Property ip:127.0.0.1 2024-03-09 10:23:39:729888 [INFO] Register Codec [PCMU] 2024-03-09 10:23:39:729890 [INFO] Register Codec [PCMA] 2024-03-09 10:23:39:729893 [INFO] Register Codec [L16] 2024-03-09 10:23:39:729895 [INFO] Register Codec [G722] 2024-03-09 10:23:39:729911 [NOTICE] Register Resource [speechsynth] 2024-03-09 10:23:39:729916 [NOTICE] Register Resource [speechrecog] 2024-03-09 10:23:39:729918 [NOTICE] Register Resource [recorder] 2024-03-09 10:23:39:729922 [NOTICE] Register Resource [speakverify] 2024-03-09 10:23:39:729925 [INFO] Register Resource Factory 2024-03-09 10:23:39:729938 [NOTICE] Create SofiaSIP Agent [SIP-Agent-1] [1.12.11-239-g54ef3e2] sip:127.0.0.1:8060;transport=udp,tcp 2024-03-09 10:23:39:729943 [INFO] Register Signaling Agent [SIP-Agent-1] 2024-03-09 10:23:39:729954 [NOTICE] Create RTSP Server [RTSP-Agent-1] 127.0.0.1:1554 [100] connection timeout [600 sec] 2024-03-09 10:23:39:730021 [INFO] Register Signaling Agent [RTSP-Agent-1] 2024-03-09 10:23:39:730027 [NOTICE] Create MRCPv2 Agent [MRCPv2-Agent-1] 127.0.0.1:1544 [100] 2024-03-09 10:23:39:730047 [INFO] Register Connection Agent [MRCPv2-Agent-1] 2024-03-09 10:23:39:730051 [NOTICE] Create Media Engine [Media-Engine-1] 2024-03-09 10:23:39:730058 [INFO] Register Media Engine [Media-Engine-1] 2024-03-09 10:23:39:730062 [NOTICE] Create RTP Termination Factory 127.0.0.1:[5000,6000] 2024-03-09 10:23:39:730064 [INFO] Register RTP Termination Factory [RTP-Factory-1] 2024-03-09 10:23:39:730069 [INFO] Load Plugin [Demo-Synth-1] [/usr/local/unimrcp/plugin/demosynth.so] 2024-03-09 10:23:39:730166 [INFO] Register MRCP Engine [Demo-Synth-1] 2024-03-09 10:23:39:730171 [INFO] Load Plugin [Huawei-Recog-1] [/usr/local/unimrcp/plugin/huaweirecog.so] 2024-03-09 10:23:39:730223 [INFO] Register MRCP Engine [Huawei-Recog-1] 2024-03-09 10:23:39:730226 [INFO] Load Plugin [Demo-Verifier-1] [/usr/local/unimrcp/plugin/demoverifier.so] 2024-03-09 10:23:39:730276 [INFO] Register MRCP Engine [Demo-Verifier-1] 2024-03-09 10:23:39:730279 [INFO] Load Plugin [Recorder-1] [/usr/local/unimrcp/plugin/mrcprecorder.so] 2024-03-09 10:23:39:730323 [INFO] Register MRCP Engine [Recorder-1] 2024-03-09 10:23:39:730333 [INFO] Register RTP Settings [RTP-Settings-1] 2024-03-09 10:23:39:730338 [NOTICE] Create MRCPv2 Profile [uni2] 2024-03-09 10:23:39:730344 [INFO] Associate Resource [speechsynth] to Engine [Demo-Synth-1] in Profile [uni2] 2024-03-09 10:23:39:730347 [INFO] Associate Resource [speechrecog] to Engine [Huawei-Recog-1] in Profile [uni2] 2024-03-09 10:23:39:730349 [INFO] Associate Resource [recorder] to Engine [Recorder-1] in Profile [uni2] 2024-03-09 10:23:39:730351 [INFO] Associate Resource [speakverify] to Engine [Demo-Verifier-1] in Profile [uni2] 2024-03-09 10:23:39:730354 [INFO] Register Profile [uni2] 2024-03-09 10:23:39:730356 [NOTICE] Create MRCPv1 Profile [uni1] 2024-03-09 10:23:39:730358 [INFO] Associate Resource [speechsynth] to Engine [Demo-Synth-1] in Profile [uni1] 2024-03-09 10:23:39:730360 [INFO] Associate Resource [speechrecog] to Engine [Huawei-Recog-1] in Profile [uni1] 2024-03-09 10:23:39:730362 [INFO] Associate Resource [recorder] to Engine [Recorder-1] in Profile [uni1] 2024-03-09 10:23:39:730364 [INFO] Associate Resource [speakverify] to Engine [Demo-Verifier-1] in Profile [uni1] 2024-03-09 10:23:39:730367 [INFO] Register Profile [uni1] 2024-03-09 10:23:39:730370 [INFO] Start Task [MRCP Server] >2024-03-09 10:23:39:730456 [INFO] Open Engine [Demo-Verifier-1] 2024-03-09 10:23:39:730471 [INFO] Start Task [Demo Verifier Engine] 2024-03-09 10:23:39:730526 [INFO] Open Engine [Huawei-Recog-1] 2024-03-09 10:23:39:730530 [INFO] Start Task [Huawei Recog Engine] 2024-03-09 10:23:39:730552 [INFO] Open Engine [Recorder-1] 2024-03-09 10:23:39:730555 [INFO] Open Engine [Demo-Synth-1] 2024-03-09 10:23:39:730557 [INFO] Start Task [Demo Synth Engine] 2024-03-09 10:23:39:730580 [INFO] Start Task [SIP-Agent-1] 2024-03-09 10:23:39:730599 [INFO] Start Task [RTSP-Agent-1] 2024-03-09 10:23:39:730619 [INFO] Start Task [MRCPv2-Agent-1] 2024-03-09 10:23:39:730642 [INFO] Start Task [Media-Engine-1] sres: /etc/resolv.conf: unknown option 2024-03-09 10:23:39:731672 [NOTICE] MRCP Server Started
再啟動客戶端，測試ASR
>run recog >[1] 2024-03-09 10:24:46:654383 [NOTICE] Create MRCP Handle 0x7f902c035630 [uni2] 2024-03-09 10:24:46:654420 [INFO] Create Channel umc-1 <new> 2024-03-09 10:24:46:654466 [INFO] Receive App Request umc-1 <new> [2] 2024-03-09 10:24:46:654535 [INFO] Add MRCP Handle umc-1 <new> 2024-03-09 10:24:46:654573 [NOTICE] Add Control Channel umc-1 <new@speechrecog> 2024-03-09 10:24:46:656847 [INFO] Send Offer umc-1 <new> [c:1 a:1 v:0] to 127.0.0.1:8060 2024-03-09 10:24:46:656879 [INFO] Local SDP umc-1 <new> v=0 o=UniMRCPClient 0 0 IN IP4 127.0.0.1 s=- c=IN IP4 127.0.0.1 t=0 0 m=application 9 TCP/MRCPv2 1 a=setup:active a=connection:new a=resource:speechrecog a=cmid:1 m=audio 4000 RTP/AVP 0 8 9 96 101 a=rtpmap:0 PCMU/8000 a=rtpmap:8 PCMA/8000 a=rtpmap:9 G722/8000 a=rtpmap:96 L16/8000 a=rtpmap:101 telephone-event/8000 a=fmtp:101 0-15 a=sendonly a=ptime:20 a=mid:1 2024-03-09 10:24:46:657606 [INFO] Receive SIP Event [nua_i_state] Status 0 INVITE sent [SIP-Agent-1] 2024-03-09 10:24:46:657624 [NOTICE] SIP Call State umc-1 [calling] 2024-03-09 10:24:46:661600 [INFO] Receive SIP Event [nua_r_invite] Status 200 OK [SIP-Agent-1] 2024-03-09 10:24:46:661658 [INFO] Receive SIP Event [nua_i_state] Status 200 OK [SIP-Agent-1] 2024-03-09 10:24:46:661664 [NOTICE] SIP Call State umc-1 [ready] 2024-03-09 10:24:46:661668 [INFO] Remote SDP umc-1 <new> v=0 o=UniMRCPServer 8207428304472510678 6899681545498287106 IN IP4 192.168.34.32 s=- c=IN IP4 127.0.0.1 t=0 0 m=application 1544 TCP/MRCPv2 1 a=setup:passive a=connection:new a=channel:328a56e4ddbc11ee@speechrecog a=cmid:1 m=audio 5000 RTP/AVP 0 101 a=rtpmap:0 PCMU/8000 a=rtpmap:101 telephone-event/8000 a=fmtp:101 0-15 a=recvonly a=ptime:20 a=mid:1 2024-03-09 10:24:46:661738 [INFO] Receive SIP Event [nua_i_active] Status 200 Call active [SIP-Agent-1] 2024-03-09 10:24:46:661756 [INFO] Receive Answer umc-1 <new> [c:1 a:1 v:0] Status 200 2024-03-09 10:24:46:661892 [NOTICE] Established TCP/MRCPv2 Connection 127.0.0.1:59830 <-> 127.0.0.1:1544 2024-03-09 10:24:46:661906 [INFO] Add Control Channel <328a56e4ddbc11ee@speechrecog> 127.0.0.1:59830 <-> 127.0.0.1:1544 [1] 2024-03-09 10:24:46:666844 [INFO] Enable RTP Session 127.0.0.1:4000 2024-03-09 10:24:46:666873 [INFO] Open RTP Transmitter 127.0.0.1:4000 -> 127.0.0.1:5000 2024-03-09 10:24:46:666878 [INFO] Media Path umc-1 Source->[LPCM/8000/1]->Bridge->[LPCM/8000/1]->Encoder->[PCMU/8000/1]->Sink 2024-03-09 10:24:46:666900 [INFO] Raise App Response umc-1 <328a56e4ddbc11ee> [2] SUCCESS [0] 2024-03-09 10:24:46:666975 [INFO] Receive App MRCP Request umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:46:666979 [INFO] Send MRCP Request umc-1 <328a56e4ddbc11ee@speechrecog> [1] 2024-03-09 10:24:46:667028 [INFO] Send MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [442 bytes] MRCP/2.0 442 DEFINE-GRAMMAR 1 Channel-Identifier: 328a56e4ddbc11ee@speechrecog Content-Type: application/srgs+xml Content-Id: request1@form-level Content-Length: 269 <?xml version="1.0"?> <grammar xmlns="http://www.w3.org/2001/06/grammar" xml:lang="en-US" version="1.0" mode="voice" root="digit"> <rule id="digit"> <one-of> <item>one</item> <item>two</item> <item>three</item> </one-of> </rule> </grammar> 2024-03-09 10:24:46:667298 [INFO] Receive MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [112 bytes] MRCP/2.0 112 1 200 COMPLETE Channel-Identifier: 328a56e4ddbc11ee@speechrecog Completion-Cause: 000 success 2024-03-09 10:24:46:667336 [INFO] Raise App MRCP Response umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:46:667403 [INFO] Receive App MRCP Request umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:46:667406 [INFO] Send MRCP Request umc-1 <328a56e4ddbc11ee@speechrecog> [2] 2024-03-09 10:24:46:667414 [INFO] Set [/usr/local/unimrcp/data/one-8kHz.pcm] as Speech Source 2024-03-09 10:24:46:667428 [INFO] Send MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [304 bytes] MRCP/2.0 304 RECOGNIZE 2 Channel-Identifier: 328a56e4ddbc11ee@speechrecog Content-Type: text/uri-list Cancel-If-Queue: false No-Input-Timeout: 5000 Recognition-Timeout: 10000 Start-Input-Timers: true Confidence-Threshold: 0.5 Save-Waveform: true Content-Length: 27 session:request1@form-level 2024-03-09 10:24:46:667664 [INFO] Receive MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [83 bytes] MRCP/2.0 83 2 200 IN-PROGRESS Channel-Identifier: 328a56e4ddbc11ee@speechrecog 2024-03-09 10:24:46:667686 [INFO] Raise App MRCP Response umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:47:541082 [INFO] Receive MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [94 bytes] MRCP/2.0 94 START-OF-INPUT 2 IN-PROGRESS Channel-Identifier: 328a56e4ddbc11ee@speechrecog 2024-03-09 10:24:47:541165 [INFO] Raise App MRCP Event umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:48:011172 [INFO] Receive MRCPv2 Data 127.0.0.1:59830 <-> 127.0.0.1:1544 [393 bytes] MRCP/2.0 393 RECOGNITION-COMPLETE 2 COMPLETE Channel-Identifier: 328a56e4ddbc11ee@speechrecog Completion-Cause: 000 success Content-Type: application/x-nlsml Content-Length: 208 <?xml version="1.0"?> <result> <interpretation grammar="session:request1@form-level.store" confidence="0.97"> <instance>one</instance> <input mode="speech">one</input> </interpretation> </result> 2024-03-09 10:24:48:011247 [INFO] Raise App MRCP Event umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:48:011354 [INFO] Interpretation[0].confidence: 0.97 2024-03-09 10:24:48:011363 [INFO] Interpretation[0].grammar: session:request1@form-level.store 2024-03-09 10:24:48:011380 [INFO] Interpretation[0].instance[0]: one 2024-03-09 10:24:48:011386 [INFO] Interpretation[0].input: one 2024-03-09 10:24:48:011388 [INFO] Interpretation[0].input.mode: speech 2024-03-09 10:24:48:011391 [INFO] Interpretation[0].input.confidence: 1.00 2024-03-09 10:24:48:011412 [INFO] Receive App Request umc-1 <328a56e4ddbc11ee> [1] 2024-03-09 10:24:48:011424 [INFO] Terminate Session umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:48:011454 [INFO] Remove Control Channel <328a56e4ddbc11ee@speechrecog> [0] 2024-03-09 10:24:48:011459 [INFO] Close TCP/MRCPv2 Connection 127.0.0.1:59830 <-> 127.0.0.1:1544 2024-03-09 10:24:48:011791 [INFO] Receive SIP Event [nua_r_bye] Status 200 OK [SIP-Agent-1] 2024-03-09 10:24:48:011805 [INFO] Receive SIP Event [nua_i_state] Status 200 to BYE [SIP-Agent-1] 2024-03-09 10:24:48:011810 [NOTICE] SIP Call State umc-1 [terminated] 2024-03-09 10:24:48:011871 [INFO] Session Terminated umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:48:016774 [INFO] Close RTP Transmitter 127.0.0.1:4000 -> 127.0.0.1:5000 [s:67 o:10720] 2024-03-09 10:24:48:016791 [INFO] Remove RTP Session 127.0.0.1:4000 2024-03-09 10:24:48:016831 [NOTICE] Destroy TCP/MRCPv2 Connection 127.0.0.1:59830 <-> 127.0.0.1:1544 2024-03-09 10:24:48:016851 [INFO] Remove MRCP Handle umc-1 <328a56e4ddbc11ee> 2024-03-09 10:24:48:016855 [INFO] Raise App Response umc-1 <328a56e4ddbc11ee> [1] SUCCESS [0] 2024-03-09 10:24:48:016953 [NOTICE] Destroy MRCP Handle umc-1
第124~130行即為模擬的識別結果
二、安裝華為ASR SDK
華為ASR C++（Linux版）SDK安裝過程參考之前的文章：https://bbs.huaweicloud.com/blogs/392949
這裡推薦一鍵安裝，因為本SDK將所有依賴均打包在倉庫中，可以在任意Linux機器上編譯透過（g++ 4.8.5及以上，cmake版本至少是3.14）
三、unimrcp整合華為ASR SDK
本章節用從底到高的順序進行記錄，即先修改底層程式碼/配置，再修改頂層程式碼
1.將（上個步驟 安裝華為ASR SDK）編譯好的SDK打包起來，並命名為huaweicloud_asr，作為備用，其程式碼結構如下：
2.進入unimrcp/plugins，調整程式碼結構，如下圖所示
cd umimrcp/plugins mv demo-recog huawei-recog # 將原始的demo-recog重新命名成huawei-recog mv huawei-recog/src/demo_recog_engine.c huawei-recog/src/huawei_recog_engine.cc cp -r path_to_sdk/huaweicloud_asr huawei-recog/huaweicloud_asr # 將華為asr sdk複製到umimrcp/plugins/huawei-recog/huaweicloud_asr處
3.修改huawei-recog/src/huawei_recog_engine.cc檔案，具體內容如下
/** Process START-INPUT-TIMERS request */ static apt_bool_t huawei_recog_channel_timers_start( mrcp_engine_channel_t *channel, mrcp_message_t *request, mrcp_message_t *response) { huawei_recog_channel_t *recog_channel = (huawei_recog_channel_t *)channel->method_obj; recog_channel->timers_started = TRUE; return mrcp_engine_channel_message_send(channel, response); } /** Dispatch MRCP request */ static apt_bool_t huawei_recog_channel_request_dispatch(mrcp_engine_channel_t *channel, mrcp_message_t *request) { apt_bool_t processed = FALSE; mrcp_message_t *response = mrcp_response_create(request, request->pool); switch (request->start_line.method_id) { case RECOGNIZER_SET_PARAMS: break; case RECOGNIZER_GET_PARAMS: break; case RECOGNIZER_DEFINE_GRAMMAR: break; case RECOGNIZER_RECOGNIZE: processed = huawei_recog_channel_recognize(channel, request, response); break; case RECOGNIZER_GET_RESULT: break; case RECOGNIZER_START_INPUT_TIMERS: processed = huawei_recog_channel_timers_start(channel, request, response); break; case RECOGNIZER_STOP: processed = huawei_recog_channel_stop(channel, request, response); break; default: break; } if (processed == FALSE) { /* send asynchronous response for not handled request */ mrcp_engine_channel_message_send(channel, response); } return TRUE; } /** Callback is called from MPF engine context to destroy any additional data associated with audio stream */ static apt_bool_t huawei_recog_stream_destroy(mpf_audio_stream_t *stream) { return TRUE; } /** Callback is called from MPF engine context to perform any action before open */ static apt_bool_t huawei_recog_stream_open(mpf_audio_stream_t *stream, mpf_codec_t *codec) { return TRUE; } /** Callback is called from MPF engine context to perform any action after close */ static apt_bool_t huawei_recog_stream_close(mpf_audio_stream_t *stream) { return TRUE; } /* Raise huawei START-OF-INPUT event */ static apt_bool_t huawei_recog_start_of_input(huawei_recog_channel_t *recog_channel) { /* create START-OF-INPUT event */ mrcp_message_t *message = mrcp_event_create(recog_channel->recog_request, RECOGNIZER_START_OF_INPUT, recog_channel->recog_request->pool); if (!message) { return FALSE; } /* set request state */ message->start_line.request_state = MRCP_REQUEST_STATE_INPROGRESS; /* send asynch event */ return mrcp_engine_channel_message_send(recog_channel->channel, message); } /* Load demo recognition result */ static apt_bool_t huawei_recog_result_load(huawei_recog_channel_t *recog_channel, mrcp_message_t *message) { mrcp_engine_channel_t *channel = recog_channel->channel; const apt_dir_layout_t *dir_layout = channel->engine->dir_layout; /* read the demo result from file 這裡是從asr真實結果中讀取*/ mrcp_generic_header_t *generic_header; apt_string_assign_n(&message->body, recog_channel->call_back->Result().c_str(), recog_channel->call_back->Result().size(), message->pool); /* get/allocate generic header */ generic_header = mrcp_generic_header_prepare(message); if (generic_header) { /* set content types */ apt_string_assign(&generic_header->content_type, "application/x-nlsml", message->pool); mrcp_generic_header_property_add(message, GENERIC_HEADER_CONTENT_TYPE); } return TRUE; } /* Raise huawei RECOGNITION-COMPLETE event */ static apt_bool_t huawei_recog_recognition_complete( huawei_recog_channel_t *recog_channel, mrcp_recog_completion_cause_e cause) { mrcp_recog_header_t *recog_header; /* create RECOGNITION-COMPLETE event */ mrcp_message_t *message = mrcp_event_create( recog_channel->recog_request, RECOGNIZER_RECOGNITION_COMPLETE, recog_channel->recog_request->pool); if (!message) { return FALSE; } /* get/allocate recognizer header */ recog_header = (mrcp_recog_header_t *)mrcp_resource_header_prepare(message); if (recog_header) { /* set completion cause */ recog_header->completion_cause = cause; mrcp_resource_header_property_add(message, RECOGNIZER_HEADER_COMPLETION_CAUSE); } /* set request state */ message->start_line.request_state = MRCP_REQUEST_STATE_COMPLETE; if (cause == RECOGNIZER_COMPLETION_CAUSE_SUCCESS) { huawei_recog_result_load(recog_channel, message); } recog_channel->recog_request = NULL; /* send asynch event */ return mrcp_engine_channel_message_send(recog_channel->channel, message); } /** Callback is called from MPF engine context to write/send new frame */ static apt_bool_t huawei_recog_stream_write(mpf_audio_stream_t *stream, const mpf_frame_t *frame) { huawei_recog_channel_t *recog_channel = (huawei_recog_channel_t *)stream->obj; if (recog_channel->stop_response) { /* send asynchronous response to STOP request */ mrcp_engine_channel_message_send(recog_channel->channel, recog_channel->stop_response); recog_channel->stop_response = NULL; recog_channel->recog_request = NULL; return TRUE; } bool finish = false; // 接收的資料幀轉給SDK if (recog_channel->recog_request) { // 第622行至660行 是unimrcp提供的斷句規則，這裡禁用中間兩個case（因為效果差），直接用華為雲的斷句規則 // mpf_detector_event_e det_event = mpf_activity_detector_process(recog_channel->detector, frame); // switch (det_event) { // case MPF_DETECTOR_EVENT_ACTIVITY: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Voice Activity " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // huawei_recog_start_of_input(recog_channel); // break; // case MPF_DETECTOR_EVENT_INACTIVITY: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Voice Inactivity " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // if (recog_channel->rasr_client) { // recog_channel->rasr_client->SendEnd(); // recog_channel->rasr_client->Close(); // } // finish = true; // huawei_recog_recognition_complete(recog_channel, RECOGNIZER_COMPLETION_CAUSE_SUCCESS); // apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "get RECOGNIZER_COMPLETION_CAUSE_SUCCESS stop recog"); // break; // case MPF_DETECTOR_EVENT_NOINPUT: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Noinput " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // if (recog_channel->timers_started == TRUE) { // if (recog_channel->rasr_client) { // recog_channel->rasr_client->SendEnd(); // recog_channel->rasr_client->Close(); // } // huawei_recog_recognition_complete(recog_channel,RECOGNIZER_COMPLETION_CAUSE_NO_INPUT_TIMEOUT); // } // finish = true; // break; // default: // break; // } Status cur_status = recog_channel->call_back->GetStatus(); switch (cur_status) { case Status::voice_start: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Voice Activity " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); huawei_recog_start_of_input(recog_channel); recog_channel->call_back->SetStatus(Status::recognizing); break; case Status::voice_end: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Voice Inactivity " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); if (recog_channel->rasr_client) { recog_channel->rasr_client->SendEnd(); recog_channel->rasr_client->Close(); } finish = true; huawei_recog_recognition_complete(recog_channel, RECOGNIZER_COMPLETION_CAUSE_SUCCESS); apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "get RECOGNIZER_COMPLETION_CAUSE_SUCCESS stop recog"); break; case Status::exceeded_silence: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Noinput " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); if (recog_channel->timers_started == TRUE) { if (recog_channel->rasr_client) { recog_channel->rasr_client->SendEnd(); recog_channel->rasr_client->Close(); } huawei_recog_recognition_complete(recog_channel,RECOGNIZER_COMPLETION_CAUSE_NO_INPUT_TIMEOUT); } finish = true; break; default: break; } if (recog_channel->recog_request) { if ((frame->type & MEDIA_FRAME_TYPE_EVENT) == MEDIA_FRAME_TYPE_EVENT) { if (frame->marker == MPF_MARKER_START_OF_EVENT) { apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Start of Event " APT_SIDRES_FMT " id:%d", MRCP_MESSAGE_SIDRES(recog_channel->recog_request), frame->event_frame.event_id); } else if (frame->marker == MPF_MARKER_END_OF_EVENT) { apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected End of Event " APT_SIDRES_FMT " id:%d duration:%d ts", MRCP_MESSAGE_SIDRES(recog_channel->recog_request), frame->event_frame.event_id, frame->event_frame.duration); } } } if (recog_channel->audio_out) { fwrite(frame->codec_frame.buffer, 1, frame->codec_frame.size, recog_channel->audio_out); } if (recog_channel->rasr_client && !finish) { recog_channel->rasr_client->SendBinary((unsigned char*)frame->codec_frame.buffer, frame->codec_frame.size); apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Sending [%d] bytes", frame->codec_frame.size); } } return TRUE; } static apt_bool_t huawei_recog_msg_signal( huawei_recog_msg_type_e type, mrcp_engine_channel_t *channel, mrcp_message_t *request) { apt_bool_t status = FALSE; huawei_recog_channel_t *huawei_channel = (huawei_recog_channel_t *)channel->method_obj; huawei_recog_engine_t *huawei_engine = huawei_channel->huawei_engine; apt_task_t *task = apt_consumer_task_base_get(huawei_engine->task); apt_task_msg_t *msg = apt_task_msg_get(task); if (msg) { huawei_recog_msg_t *huawei_msg; msg->type = TASK_MSG_USER; huawei_msg = (huawei_recog_msg_t *)msg->data; huawei_msg->type = type; huawei_msg->channel = channel; huawei_msg->request = request; status = apt_task_msg_signal(task, msg); } return status; } static apt_bool_t huawei_recog_msg_process(apt_task_t *task, apt_task_msg_t *msg) { huawei_recog_msg_t *huawei_msg = (huawei_recog_msg_t *)msg->data; switch (huawei_msg->type) { case HUAWEI_RECOG_MSG_OPEN_CHANNEL: /* open channel and send asynch response */ mrcp_engine_channel_open_respond(huawei_msg->channel, TRUE); apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "HUAWEI_RECOG_MSG_OPEN_CHANNEL"); break; case HUAWEI_RECOG_MSG_CLOSE_CHANNEL: { /* close channel, make sure there is no activity and send asynch response */ huawei_recog_channel_t *recog_channel = (huawei_recog_channel_t *)huawei_msg->channel->method_obj; if (recog_channel->audio_out) { fclose(recog_channel->audio_out); recog_channel->audio_out = NULL; } apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "HUAWEI_RECOG_MSG_CLOSE_CHANNEL"); mrcp_engine_channel_close_respond(huawei_msg->channel); break; } case HUAWEI_RECOG_MSG_REQUEST_PROCESS: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "HUAWEI_RECOG_MSG_REQUEST_PROCESS"); huawei_recog_channel_request_dispatch(huawei_msg->channel, huawei_msg->request); break; default: break; } return TRUE; }
4.修改asr外掛的makefile.am檔案（進入路徑：unimrcp/plugins）: vim huawei-recog/Makefile.am
AM_CPPFLAGS = $(UNIMRCP_PLUGIN_INCLUDES) plugin_LTLIBRARIES = huaweirecog.la huaweirecog_la_SOURCES = src/huawei_recog_engine.cc huaweirecog_la_LDFLAGS = $(UNIMRCP_PLUGIN_OPTS) \ -L$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party \ -L$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/jsoncpp-install/lib64 \ -L$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/gflags-install/lib \ -L$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/glog-install/lib64 \ ${CMAKE_SOURCE_DIR}/third_party/openssl-install/lib \ -lhuawei_rasr -ssl -lcrypto -ljsoncpp -lgflags_nothreads -stdc++=14 huaweirecog_ladir = $(libdir) huaweirecog_la_DATA = $(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/libhuawei_rasr.so \ $(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/openssl-install/lib/libcrypto.so \ $(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/openssl-install/lib/libssl.so \ $(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/jsoncpp-install/lib64/libjsoncpp.so \ $(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/gflags-install/lib/libgflags_nothreads.so include $(top_srcdir)/build/rules/uniplugin.am UNIMRCP_PLUGIN_INCLUDES += -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/include \ -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/websocketpp-src \ -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/jsoncpp-install/include \ -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/boost-src/ \ -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/openssl-install/include \ -I$(top_srcdir)/plugins/huawei-recog/huaweicloud_asr/third_party/gflags-install/include
5.修改外掛的Makefile.am（進入路徑：unimrcp/plugins） vim Makefile.am
6.修改最頂層的configure.ac（進入unimrcp根目錄）：
dnl dnl Autoconf configuration file for UniMRCP. dnl dnl Use ./bootstrap to produce a configure script. dnl AC_PREREQ(2.59) AC_INIT([unimrcp],[1.8.0]) AC_CONFIG_AUX_DIR([build]) AC_CONFIG_MACRO_DIR([build/acmacros]) dnl Set ac_macro_dir variable manually for autoconf 2.61 and above. ac_macro_dir="build/acmacros" AC_SUBST(ac_aux_dir) AC_SUBST(ac_macro_dir) dnl Include m4 macros for libtool 2. sinclude(build/acmacros/libtool.m4) sinclude(build/acmacros/ltoptions.m4) sinclude(build/acmacros/ltsugar.m4) sinclude(build/acmacros/ltversion.m4) sinclude(build/acmacros/lt~obsolete.m4) AC_PREFIX_DEFAULT(/usr/local/unimrcp) dnl Define the directory layout. APR_ENABLE_LAYOUT(classic, [plugindir logdir vardir]) AC_SUBST(plugindir) AC_SUBST(logdir) AC_SUBST(vardir) dnl Reparse the configure arguments. APR_PARSE_ARGUMENTS dnl Generate ./config.nice to reuse ./configure command-line. APR_CONFIG_NICE(config.nice) AM_INIT_AUTOMAKE([no-define nostdinc foreign subdir-objects]) dnl Enable silent build rules available since automake 1.11. m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])]) dnl Set default language. AC_LANG_C AC_PROG_CC AC_PROG_CXX AM_PROG_CC_C_O AC_PROG_INSTALL dnl Skip detection of Fortran. m4_undefine([AC_PROG_F77]) m4_defun([AC_PROG_F77],[]) AC_PROG_LIBTOOL dnl Suppress warning: ar: 'u' modifier ignored since 'D' is the default AC_SUBST(AR_FLAGS, [cr]) dnl Do not use autoconf generated compiler DEFS. rm confdefs.h touch confdefs.h dnl Check for C compiler vendor. AX_COMPILER_VENDOR dnl Search for pkg-config. AC_PATH_PROG(PKG_CONFIG, pkg-config, no) if test "x$PKG_CONFIG" = "xno"; then AC_MSG_ERROR([pkg-config is missing]) fi dnl Get version information. get_version="$srcdir/build/get-version.sh" version_hdr="$srcdir/build/uni_version.h" plugin_version_hdr="$srcdir/libs/mrcp-engine/include/mrcp_engine_plugin.h" UNI_DOTTED_VERSION="`$get_version all $version_hdr UNI`" UNI_LT_VERSION="-version-info `$get_version libtool $version_hdr UNI`" PLUGIN_LT_VERSION="-version-info `$get_version libtool $plugin_version_hdr PLUGIN`" AC_SUBST(UNI_DOTTED_VERSION) AC_SUBST(UNI_LT_VERSION) AC_SUBST(PLUGIN_LT_VERSION) echo "UniMRCP Version: ${UNI_DOTTED_VERSION}" dnl Check for the APR and APR-util libraries. UNIMRCP_CHECK_APR dnl Check for the Sofia-SIP library. UNIMRCP_CHECK_SOFIA dnl Enable AMR codec. AC_ARG_ENABLE(amr-codec, [AC_HELP_STRING([--enable-amr-codec ],[enable AMR codec])], [enable_amr_codec="$enableval"], [enable_amr_codec="no"]) AC_MSG_NOTICE([enable AMR codec: $enable_amr_codec]) if test "${enable_amr_codec}" != "no"; then dnl Check for the OpenCORE AMR library. UNIMRCP_CHECK_OPENCORE_AMR dnl Check for the VO AMRWBENC library. UNIMRCP_CHECK_VO_AMRWBENC fi AM_CONDITIONAL([UNIMRCP_AMR_CODEC],[test "${enable_amr_codec}" != "no"]) dnl Enable inter-library dependencies. AC_ARG_ENABLE(interlib-deps, [AC_HELP_STRING([--disable-interlib-deps ],[disable inter-library dependencies (might break builds)])], [enable_interlib_deps="$enableval"], [enable_interlib_deps="yes"]) AC_MSG_NOTICE([enable inter-library dependencies: $enable_interlib_deps]) if test "${enable_interlib_deps}" == "yes"; then link_all_deplibs=yes link_all_deplibs_CXX=yes else link_all_deplibs=no link_all_deplibs_CXX=no fi dnl Enable maintainer mode. AC_ARG_ENABLE(maintainer-mode, [AC_HELP_STRING([--enable-maintainer-mode ],[turn on debugging and compile time warnings])], [enable_maintainer_mode="$enableval"], [enable_maintainer_mode="no"]) AC_MSG_NOTICE([enable maintainer mode: $enable_maintainer_mode]) if test "${enable_maintainer_mode}" != "no"; then APR_ADDTO(CFLAGS,-g) if test "x${ax_cv_c_compiler_vendor}" = "xgnu" ; then APR_ADDTO(CFLAGS,-Wall -Werror) fi fi dnl UniMRCP client library. AC_ARG_ENABLE(client-lib, [AC_HELP_STRING([--disable-client-lib ],[exclude unimrcpclient lib from build])], [enable_client_lib="$enableval"], [enable_client_lib="yes"]) AM_CONDITIONAL([UNIMRCP_CLIENT_LIB],[test "${enable_client_lib}" = "yes"]) dnl Sample UniMRCP client application in C. AC_ARG_ENABLE(client-app, [AC_HELP_STRING([--disable-client-app ],[exclude sample unimrcpclient app from build])], [enable_client_app="$enableval"], [enable_client_app="yes"]) AM_CONDITIONAL([UNIMRCP_CLIENT_APP],[test "${enable_client_lib}" = "yes" && test "${enable_client_app}" = "yes"]) dnl Sample UniMRCP client application in C++. AC_ARG_ENABLE(umc, [AC_HELP_STRING([--disable-umc ],[exclude sample unimrcpclient C++ app from build])], [enable_umc="$enableval"], [enable_umc="yes"]) AM_CONDITIONAL([UMC],[test "${enable_client_lib}" = "yes" && test "${enable_umc}" = "yes"]) dnl Miscellaneous ASR client library and application. AC_ARG_ENABLE(asr-client, [AC_HELP_STRING([--disable-asr-client ],[exclude misc ASR client lib and app from build])], [enable_asr_client="$enableval"], [enable_asr_client="yes"]) AM_CONDITIONAL([ASR_CLIENT],[test "${enable_client_lib}" = "yes" && test "${enable_asr_client}" = "yes"]) AM_CONDITIONAL([COMMON_CLIENT_DATA],[test "${enable_client_app}" = "yes" || test "${enable_umc}" = "yes" ||test "${enable_asr_client}" = "yes"]) dnl UniMRCP server library. AC_ARG_ENABLE(server-lib, [AC_HELP_STRING([--disable-server-lib ],[exclude unimrcpserver lib from build])], [enable_server_lib="$enableval"], [enable_server_lib="yes"]) AM_CONDITIONAL([UNIMRCP_SERVER_LIB],[test "${enable_server_lib}" = "yes"]) dnl UniMRCP server application. AC_ARG_ENABLE(server-app, [AC_HELP_STRING([--disable-server-app ],[exclude unimrcpserver app from build])], [enable_server_app="$enableval"], [enable_server_app="yes"]) AM_CONDITIONAL([UNIMRCP_SERVER_APP],[test "${enable_server_lib}" = "yes" && test "${enable_server_app}" = "yes"]) dnl Demo synthesizer plugin. UNI_PLUGIN_ENABLED(demosynth) AM_CONDITIONAL([DEMOSYNTH_PLUGIN],[test "${enable_demosynth_plugin}" = "yes"]) dnl Huawei recognizer plugin. UNI_PLUGIN_ENABLED(huaweirecog) AM_CONDITIONAL([HUAWEIRECOG_PLUGIN],[test "${enable_huaweirecog_plugin}" = "yes"]) dnl Demo verifier plugin. UNI_PLUGIN_ENABLED(demoverifier) AM_CONDITIONAL([DEMOVERIFIER_PLUGIN],[test "${enable_demoverifier_plugin}" = "yes"]) dnl Recorder plugin. UNI_PLUGIN_ENABLED(recorder) AM_CONDITIONAL([RECORDER_PLUGIN],[test "${enable_recorder_plugin}" = "yes"]) dnl Enable test suites. AC_ARG_ENABLE(test-suites, [AC_HELP_STRING([--enable-test-suites ],[build test suites])], [enable_test_suites="$enableval"], [enable_test_suites="no"]) AM_CONDITIONAL([TEST_SUITES],[test "${enable_test_suites}" != "no"]) AM_CONDITIONAL(ISMAC, [test `uname -s` = Darwin]) AC_CONFIG_FILES([ Makefile libs/Makefile libs/apr-toolkit/Makefile libs/mpf/Makefile libs/mrcp/Makefile libs/mrcp-signaling/Makefile libs/mrcpv2-transport/Makefile libs/mrcp-engine/Makefile libs/mrcp-server/Makefile libs/mrcp-client/Makefile libs/uni-rtsp/Makefile modules/Makefile modules/mrcp-sofiasip/Makefile modules/mrcp-unirtsp/Makefile plugins/Makefile plugins/mrcp-recorder/Makefile plugins/demo-synth/Makefile plugins/huawei-recog/Makefile plugins/demo-verifier/Makefile platforms/Makefile platforms/libunimrcp-server/Makefile platforms/libunimrcp-client/Makefile platforms/unimrcp-server/Makefile platforms/unimrcp-client/Makefile platforms/libasr-client/Makefile platforms/asr-client/Makefile platforms/umc/Makefile tests/Makefile tests/apttest/Makefile tests/mpftest/Makefile tests/mrcptest/Makefile tests/rtsptest/Makefile tests/strtablegen/Makefile build/Makefile build/pkgconfig/Makefile build/pkgconfig/unimrcpclient.pc build/pkgconfig/unimrcpserver.pc build/pkgconfig/unimrcpplugin.pc conf/Makefile data/Makefile docs/doxygen.conf ]) AC_OUTPUT echo echo '****************************** REPORT ******************************' echo echo UniMRCP version............... : $UNI_DOTTED_VERSION echo echo APR version................... : $apr_version echo APR-util version.............. : $apu_version echo Sofia-SIP version............. : $sofia_version if test "${enable_amr_codec}" != "no"; then echo echo OpenCORE AMR version.......... : $opencore_amr_version echo VO AMRWBENC version........... : $vo_amrwbenc_version fi echo echo Compiler...................... : $CC echo Compiler flags................ : $CFLAGS echo Preprocessor definitions...... : $CPPFLAGS echo Linker flags.................. : $LDFLAGS echo echo UniMRCP client lib............ : $enable_client_lib echo Sample UniMRCP client app..... : $enable_client_app echo Sample UMC C++ client app..... : $enable_umc echo Misc ASR client lib and app... : $enable_asr_client echo echo UniMRCP server lib............ : $enable_server_lib echo UniMRCP server app............ : $enable_server_app echo echo Demo synthesizer plugin....... : $enable_demosynth_plugin echo Huawei recognizer plugin........ : $enable_huaweirecog_plugin echo Demo verifier plugin.......... : $enable_demoverifier_plugin echo Recorder plugin............... : $enable_recorder_plugin echo echo Installation layout........... : $layout_name echo Installation directory........ : $prefix echo echo '********************************************************************'
7.新增asr配置檔案（進入unimrcp根目錄），這些都是必填項，按照備註獲取相應的asr鑑權資訊：vim conf/asr_info.conf
8.重新編譯unimrcp，預設會安裝在/usr/local/unimrcp中
./bootstrap make -j make install
9.unimrcp程式碼結構展示
/usr/local/unimrcp ├── bin # 可執行檔案 ├── conf # 配置檔案，可以設定訪問IP和埠號 ├── data # 預設提供的一些音訊檔案 ├── include # 程式必須的標頭檔案 ├── lib # 依賴的庫檔案 ├── log # 日誌路徑 ├── plugin # 外掛所依賴的庫檔案 └── var # 儲存的音訊檔案
10.執行結果展示
啟動服務端 /usr/local/unimrcp/bin/unimrcpserver
(base) [root@asr-build bin]# ./unimrcpserver 2024-03-11 17:17:14:704844 [NOTICE] UniMRCP Server [1.8.0] 2024-03-11 17:17:14:704913 [INFO] APR [1.5.2] 2024-03-11 17:17:14:704924 [NOTICE] Create MRCP Server 2024-03-11 17:17:14:704946 [NOTICE] Open Config File [/usr/local/unimrcp/conf/unimrcpserver.xml] 2024-03-11 17:17:14:705077 [INFO] Set Property ip:127.0.0.1 2024-03-11 17:17:14:705086 [INFO] Register Codec [PCMU] 2024-03-11 17:17:14:705088 [INFO] Register Codec [PCMA] 2024-03-11 17:17:14:705090 [INFO] Register Codec [L16] 2024-03-11 17:17:14:705092 [INFO] Register Codec [G722] 2024-03-11 17:17:14:705127 [NOTICE] Register Resource [speechsynth] 2024-03-11 17:17:14:705132 [NOTICE] Register Resource [speechrecog] 2024-03-11 17:17:14:705135 [NOTICE] Register Resource [recorder] 2024-03-11 17:17:14:705138 [NOTICE] Register Resource [speakverify] 2024-03-11 17:17:14:705141 [INFO] Register Resource Factory 2024-03-11 17:17:14:705157 [NOTICE] Create SofiaSIP Agent [SIP-Agent-1] [1.12.11-239-g54ef3e2] sip:127.0.0.1:8060;transport=udp,tcp 2024-03-11 17:17:14:705163 [INFO] Register Signaling Agent [SIP-Agent-1] 2024-03-11 17:17:14:705178 [NOTICE] Create RTSP Server [RTSP-Agent-1] 127.0.0.1:1554 [100] connection timeout [600 sec] 2024-03-11 17:17:14:705258 [INFO] Register Signaling Agent [RTSP-Agent-1] 2024-03-11 17:17:14:705266 [NOTICE] Create MRCPv2 Agent [MRCPv2-Agent-1] 127.0.0.1:1544 [100] 2024-03-11 17:17:14:705290 [INFO] Register Connection Agent [MRCPv2-Agent-1] 2024-03-11 17:17:14:705297 [NOTICE] Create Media Engine [Media-Engine-1] 2024-03-11 17:17:14:705305 [INFO] Register Media Engine [Media-Engine-1] 2024-03-11 17:17:14:705309 [NOTICE] Create RTP Termination Factory 127.0.0.1:[5000,6000] 2024-03-11 17:17:14:705311 [INFO] Register RTP Termination Factory [RTP-Factory-1] 2024-03-11 17:17:14:705316 [INFO] Load Plugin [Demo-Synth-1] [/usr/local/unimrcp/plugin/demosynth.so] 2024-03-11 17:17:14:705393 [INFO] Register MRCP Engine [Demo-Synth-1] 2024-03-11 17:17:14:705397 [INFO] Load Plugin [Huawei-Recog-1] [/usr/local/unimrcp/plugin/huaweirecog.so] 2024-03-11 17:17:14:712580 [INFO] Register MRCP Engine [Huawei-Recog-1] 2024-03-11 17:17:14:712599 [INFO] Load Plugin [Demo-Verifier-1] [/usr/local/unimrcp/plugin/demoverifier.so] 2024-03-11 17:17:14:712677 [INFO] Register MRCP Engine [Demo-Verifier-1] 2024-03-11 17:17:14:712682 [INFO] Load Plugin [Recorder-1] [/usr/local/unimrcp/plugin/mrcprecorder.so] 2024-03-11 17:17:14:712739 [INFO] Register MRCP Engine [Recorder-1] 2024-03-11 17:17:14:712755 [INFO] Register RTP Settings [RTP-Settings-1] 2024-03-11 17:17:14:712764 [NOTICE] Create MRCPv2 Profile [uni2] 2024-03-11 17:17:14:712773 [INFO] Associate Resource [speechsynth] to Engine [Demo-Synth-1] in Profile [uni2] 2024-03-11 17:17:14:712776 [INFO] Associate Resource [speechrecog] to Engine [Huawei-Recog-1] in Profile [uni2] 2024-03-11 17:17:14:712779 [INFO] Associate Resource [recorder] to Engine [Recorder-1] in Profile [uni2] 2024-03-11 17:17:14:712782 [INFO] Associate Resource [speakverify] to Engine [Demo-Verifier-1] in Profile [uni2] 2024-03-11 17:17:14:712786 [INFO] Register Profile [uni2] 2024-03-11 17:17:14:712790 [NOTICE] Create MRCPv1 Profile [uni1] 2024-03-11 17:17:14:712793 [INFO] Associate Resource [speechsynth] to Engine [Demo-Synth-1] in Profile [uni1] 2024-03-11 17:17:14:712796 [INFO] Associate Resource [speechrecog] to Engine [Huawei-Recog-1] in Profile [uni1] 2024-03-11 17:17:14:712799 [INFO] Associate Resource [recorder] to Engine [Recorder-1] in Profile [uni1] 2024-03-11 17:17:14:712802 [INFO] Associate Resource [speakverify] to Engine [Demo-Verifier-1] in Profile [uni1] 2024-03-11 17:17:14:712805 [INFO] Register Profile [uni1] 2024-03-11 17:17:14:712810 [INFO] Start Task [MRCP Server] >2024-03-11 17:17:14:712891 [INFO] Open Engine [Demo-Verifier-1] 2024-03-11 17:17:14:712906 [INFO] Start Task [Demo Verifier Engine] 2024-03-11 17:17:14:712963 [INFO] Open Engine [Huawei-Recog-1] 2024-03-11 17:17:14:712966 [INFO] Start Task [Huawei Recog Engine] 2024-03-11 17:17:14:712990 [INFO] Open Engine [Recorder-1] 2024-03-11 17:17:14:712993 [INFO] Open Engine [Demo-Synth-1] 2024-03-11 17:17:14:712995 [INFO] Start Task [Demo Synth Engine] 2024-03-11 17:17:14:713019 [INFO] Start Task [SIP-Agent-1] 2024-03-11 17:17:14:713037 [INFO] Start Task [RTSP-Agent-1] 2024-03-11 17:17:14:713159 [INFO] Start Task [MRCPv2-Agent-1] 2024-03-11 17:17:14:713313 [INFO] Start Task [Media-Engine-1] sres: /etc/resolv.conf: unknown option 2024-03-11 17:17:14:714108 [NOTICE] MRCP Server Started
啟動客戶端：/usr/local/unimrcp/umc，然後在互動視窗執行run recog，獲取識別結果
四、unimrcp定製化改動
1. 修改xml檔案（在conf資料夾中），指定訪問地址和埠號。埠號我們採用預設埠，IP地址指定為本地IP
unimrcpclient.xml修改：
unimrcpserver.xml修改：
2. 利用unimrcp調整ASR斷句引數
unimrcp的header裡面提供了一組斷句引數，來控制響應時間/句子長段（比如，說完一段話之後，等待多少毫秒，給出結果）。各引數名及其含義如下（在路徑data/params_default.txt下）
No-Input-Timeout: 5000 -----> 使用者接到電話後超過5000ms沒有聲音，就結束識別 等價於華為雲ASR的vad_head Recognition-Timeout: 20000 ------> 使用者一直說話，最長能說多久（預設20s），結束識別。等價於華為雲ASR的max_seconds Speech-Complete-Timeout: 400 -------> 使用者開始說話後，停頓多久(預設400ms)就結束說話
2.1 透過請求引數傳遞斷句引數
以上引數均是透過請求體的header進行傳遞，參考文件：https://www.unimrcp.org/manuals/html/GoogleSRUsageManual.html
可以透過unimrcp列印的日誌，檢視引數設定是否生效：
2.2 強制指定斷句引數
為了方便管理，建議對所有通話保留相同的斷句引數，具體設定如下：
if(recog_header) { if(mrcp_resource_header_property_check(request,RECOGNIZER_HEADER_START_INPUT_TIMERS) == TRUE) { recog_channel->timers_started = recog_header->start_input_timers; } if(mrcp_resource_header_property_check(request,RECOGNIZER_HEADER_NO_INPUT_TIMEOUT) == TRUE) { mpf_activity_detector_noinput_timeout_set(recog_channel->detector,10000); // 強制指定 如果10s中無人說話，就結束 } if(mrcp_resource_header_property_check(request,RECOGNIZER_HEADER_SPEECH_COMPLETE_TIMEOUT) == TRUE) { mpf_activity_detector_silence_timeout_set(recog_channel->detector, 600); // 強制指定600ms } }
3. 利用華為雲ASR做智慧短句
華為雲ASR提供的sentence模式，可適用於外呼、數字人場景，透過設定vad_head/vad_tail/max_seconds引數可以智慧短句，相關程式碼變動如下
if (recog_channel->recog_request) { // 第622行至660行 是unimrcp提供的斷句規則，這裡禁用中間兩個case（因為效果差），直接用華為雲的斷句規則 // mpf_detector_event_e det_event = mpf_activity_detector_process(recog_channel->detector, frame); // switch (det_event) { // case MPF_DETECTOR_EVENT_ACTIVITY: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Voice Activity " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // huawei_recog_start_of_input(recog_channel); // break; // case MPF_DETECTOR_EVENT_INACTIVITY: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Voice Inactivity " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // if (recog_channel->rasr_client) { // recog_channel->rasr_client->SendEnd(); // recog_channel->rasr_client->Close(); // } // finish = true; // huawei_recog_recognition_complete(recog_channel, RECOGNIZER_COMPLETION_CAUSE_SUCCESS); // apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "get RECOGNIZER_COMPLETION_CAUSE_SUCCESS stop recog"); // break; // case MPF_DETECTOR_EVENT_NOINPUT: // apt_log(RECOG_LOG_MARK, // APT_PRIO_INFO, // "Detected Noinput " APT_SIDRES_FMT, // MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); // if (recog_channel->timers_started == TRUE) { // if (recog_channel->rasr_client) { // recog_channel->rasr_client->SendEnd(); // recog_channel->rasr_client->Close(); // } // huawei_recog_recognition_complete(recog_channel,RECOGNIZER_COMPLETION_CAUSE_NO_INPUT_TIMEOUT); // } // finish = true; // break; // default: // break; // } Status cur_status = recog_channel->call_back->GetStatus(); switch (cur_status) { case Status::voice_start: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Voice Activity " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); huawei_recog_start_of_input(recog_channel); recog_channel->call_back->SetStatus(Status::recognizing); break; case Status::voice_end: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Voice Inactivity " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); if (recog_channel->rasr_client) { recog_channel->rasr_client->SendEnd(); recog_channel->rasr_client->Close(); } finish = true; huawei_recog_recognition_complete(recog_channel, RECOGNIZER_COMPLETION_CAUSE_SUCCESS); apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "get RECOGNIZER_COMPLETION_CAUSE_SUCCESS stop recog"); break; case Status::exceeded_silence: apt_log(RECOG_LOG_MARK, APT_PRIO_INFO, "Detected Noinput " APT_SIDRES_FMT, MRCP_MESSAGE_SIDRES(recog_channel->recog_request)); if (recog_channel->timers_started == TRUE) { if (recog_channel->rasr_client) { recog_channel->rasr_client->SendEnd(); recog_channel->rasr_client->Close(); } huawei_recog_recognition_complete(recog_channel,RECOGNIZER_COMPLETION_CAUSE_NO_INPUT_TIMEOUT); } finish = true; break; default: break; }
五、快速搭建unimrcp服務
作者按照前四章逐步執行，並在GitHub和Gitee上釋出了修改後的程式碼，大家可以直接拿來使用。
- Github連結：戳這裡
- Gitee連結： 戳這裡
本程式碼倉提供兩種安裝方式供使用者選擇
方法一： 原始碼安裝
- 先下載安裝華為雲實時語音識別SDK，安裝指南參考https://bbs.huaweicloud.com/blogs/392949
- 將安裝的華為雲ASR SDK放入plugins/huawei_recog/huaweicloud_asr 】
- 進入根目錄/unimcrp-deps-1.6.0，安裝安裝unimrcp的依賴庫
./build-dep-libs.sh
4.進入根目錄，安裝unimrcp
./bootstrap ./configure make make install
5.unimrcp預設安裝在/usr/local/unimrcp中,使用前需要手動修改asr_info.conf中的配置項
方法二：直接使用release庫
wget https://gitee.com/computervisionlearner/unimrcp_with_huaweicloud_asr/releases/download/v0.0.1/unimrcp-1.8.0_unimrcp-deps-1.6.0_binary.tar.gz tar -xzvf unimrcp-1.8.0_unimrcp-deps-1.6.0_binary.tar.gz
點選關注，第一時間瞭解華為雲新鮮技術~