fix(audio): prevent empty session replay

ql-application/threadx/interface/custom_app/vad_det.c

@@ -29,6 +29,9 @@
 // 在文件顶部添加状态变量
 static uint8_t speech_detect_count = 0;  // 连续语音帧计数器
 #define SPEECH_DETECT_THRESHOLD  20        // 连续检测到语音的阈值
+#define PLAYBACK_VAD_GUARD_MS     800       // 屏蔽播放结束后的尾音和音频通路切换噪声
+#define VAD_FRAME_BYTES           640       // 20ms @ 16kHz, 16-bit mono
+#define VAD_PREROLL_FRAMES        25        // 保留500ms预录音，避免吞掉用户首字
 uint8_t no_speech_timeout = 0;
 extern int g_mic_working;
 
@@ -57,6 +60,9 @@ static int64_t no_sound_start_time = 0;
 static int64_t cmd_wait_start_time = 0;
 static bool first_sound_detected = false;
 static int64_t recording_start_time = 0;  // 记录录音开始时间（首次检测到声音时）
+static uint8_t vad_preroll[VAD_PREROLL_FRAMES][VAD_FRAME_BYTES];
+static size_t vad_preroll_next = 0;
+static size_t vad_preroll_count = 0;
 
 static int vad_tts_playback_finished_status = 0;
 // 统计
@@ -101,6 +107,9 @@ void set_state(system_state_t new_state) {
                 no_sound_start_time = now;
                 first_sound_detected = false;
                 speech_frame_count = 0;
+				speech_detect_count = 0;
+				vad_preroll_next = 0;
+				vad_preroll_count = 0;
 				g_mic_working = 0;
                 break;
             case STATE_RECORDING_ACTIVE:
@@ -290,26 +299,63 @@ void vad_leave_message(void)
 	audio_packet_type = 4;
 	
 	vad_leave_message_flag = 1;
+	/* A previous conversation may still be parked in WAITING_COMMAND.  NFC
+	 * leave-message recording must enter IDLE first, otherwise g_triggered is
+	 * left pending forever and its 10-second no-speech timer never starts. */
+	set_state(STATE_IDLE);
 	vad_audio_trigger_record();
 }
 
+int vad_leave_message_is_active(void)
+{
+	return vad_leave_message_flag != 0 || audio_packet_type == 4;
+}
+
+/* The conversation key has priority over an unfinished NFC leave-message
+ * transaction.  Clear every local latch before the normal conversation prompt
+ * is queued so that its recording can never inherit packetType=4. */
+int vad_cancel_leave_message(void)
+{
+	int was_active = vad_leave_message_is_active();
+
+	if (!was_active) return 0;
+
+	printf("cancel pending leave message\r\n");
+	vad_leave_message_flag = 0;
+	g_triggered = false;
+	audio_packet_type = 0;
+	no_speech_timeout = 0;
+	speech_detect_count = 0;
+	vad_preroll_next = 0;
+	vad_preroll_count = 0;
+	key_start_leave = 0;
+
+	if (vad_audio_is_recording()) {
+		set_state(STATE_IDLE);
+		vad_audio_stop_no_wait_resp();
+	} else {
+		set_state(STATE_IDLE);
+	}
+
+	return 1;
+}
+
 // 主处理函数 - 需要在循环中定期调用
 void vad_audio_process(void) {
     size_t bytes_read = 0;
     int64_t now = hal_os_up_time_ms();
     
     static size_t buffer_offset = 0;
-	uint8_t mic_buf[640]={0};
-	int record_len = 640;
+	uint8_t mic_buf[VAD_FRAME_BYTES]={0};
+	int record_len = VAD_FRAME_BYTES;
 	uint16 frame[320];
 	int index = 0, i = 0;
     
     // 状态机处理
     switch (g_state) {
-        case STATE_IDLE:
-			no_speech_timeout = 0;
-			vad_leave_message_flag = 0;
-            speech_detect_count = 0;  // 空闲时清零计数器
+        case STATE_IDLE:
+			no_speech_timeout = 0;
+            speech_detect_count = 0;  // 空闲时清零计数器
             vad_tts_playback_finished_status = 0;
 			
             if (g_triggered) {
@@ -326,15 +372,19 @@ void vad_audio_process(void) {
 			//每次读取640字节音频数据
 			ql_audio_record_read(record_hd, mic_buf, &record_len); 
 
-			//如果检测到声音再去传输
-		//	if(first_sound_detected){
-			//实际上每次编码需要输入1152字节的PCM数据
-			if(hal_os_rb_write(ring_sender, (uint8*)mic_buf, record_len, 25)>0){
-
-				//quec_usbcdc_write((uint8*)mic_buf, record_len);
-				ql_rtos_semaphore_release(sender_sem); //告诉线程去获取录音数据
+			if (g_state == STATE_RECORDING_WAIT_SOUND) {
+				/* 首次开口确认前不向服务端发送静音。只在本地保留最近500ms，
+				 * 确认开口时再回填，防止截掉用户首字。 */
+				memcpy(vad_preroll[vad_preroll_next], mic_buf, VAD_FRAME_BYTES);
+				vad_preroll_next = (vad_preroll_next + 1) % VAD_PREROLL_FRAMES;
+				if (vad_preroll_count < VAD_PREROLL_FRAMES) {
+					vad_preroll_count++;
+				}
+			} else {
+				if(hal_os_rb_write(ring_sender, (uint8*)mic_buf, record_len, 25)>0){
+					ql_rtos_semaphore_release(sender_sem);
+				}
 			}
-		//	}
 		#if 1
 			for(index=0; index<320; index++)
 			{
@@ -347,6 +397,12 @@ void vad_audio_process(void) {
             
             // 首次声音检测
             if (!first_sound_detected) {
+				/* 播放刚结束时扬声器尾音和PA切换噪声容易被FVAD认为用户开口。 */
+				if ((now - no_sound_start_time) < PLAYBACK_VAD_GUARD_MS) {
+					speech_detect_count = 0;
+					buffer_offset = 0;
+					break;
+				}
                 if (current_vad == VAD_STATE_ACTIVE) {
                     speech_detect_count++;
                     printf("Speech frame count: %d\r\n", speech_detect_count);
@@ -363,8 +419,23 @@ void vad_audio_process(void) {
 						}
 						#endif
 						
-                        first_sound_detected = true;
-                        set_state(STATE_RECORDING_ACTIVE);
+	                        first_sound_detected = true;
+						{
+							size_t first = (vad_preroll_next + VAD_PREROLL_FRAMES - vad_preroll_count) % VAD_PREROLL_FRAMES;
+							size_t n;
+							int preroll_queued = 0;
+							for (n = 0; n < vad_preroll_count; n++) {
+								size_t frame_index = (first + n) % VAD_PREROLL_FRAMES;
+								if (hal_os_rb_write(ring_sender, vad_preroll[frame_index], VAD_FRAME_BYTES, 25) > 0) {
+									preroll_queued = 1;
+								}
+							}
+							if (preroll_queued) {
+								ql_rtos_semaphore_release(sender_sem);
+							}
+							vad_preroll_count = 0;
+						}
+	                        set_state(STATE_RECORDING_ACTIVE);
                         printf("First speech confirmed by FVAD (continuous %d frames)", SPEECH_DETECT_THRESHOLD);
                     }
                 } else {
@@ -418,15 +489,20 @@ void vad_audio_process(void) {
             buffer_offset = 0;
             break;
             
-        case STATE_WAITING_COMMAND:
-        	if(vad_tts_playback_finished_status == 1) {
-                printf("finished_status:%d, no_speech_timeout:%d\n", vad_tts_playback_finished_status, no_speech_timeout);
-                set_state(STATE_IDLE);
-                ql_rtos_task_sleep_ms(50);
-                if(no_speech_timeout == 0&&vad_leave_message_flag==0) {
-                    vad_audio_trigger_record();
-                }
-            }
+        case STATE_WAITING_COMMAND:
+			if(vad_tts_playback_finished_status == 1) {
+				int leave_message_finished = vad_leave_message_flag != 0;
+                printf("finished_status:%d, no_speech_timeout:%d\n", vad_tts_playback_finished_status, no_speech_timeout);
+                set_state(STATE_IDLE);
+                ql_rtos_task_sleep_ms(50);
+                if(no_speech_timeout == 0&&vad_leave_message_flag==0) {
+                    vad_audio_trigger_record();
+                }
+				if(leave_message_finished) {
+					vad_leave_message_flag = 0;
+					audio_packet_type = 0;
+				}
+            }
 
 			key_start_leave = 0;
             break;

ql-application/threadx/interface/custom_app/ws_talkingq.c

@@ -30,9 +30,9 @@
 
 
 #define PCM_FRAME_LENGTH    1152
-#define CLOUD_TTS_ERROR_URL_FORMAT \
-    "http://101.35.224.118:8080/assets/roles/banban/tts_error/tts_error%02u.mp3"
-#define LOCAL_TTS_ERROR_PATH "U:/tts_error.mp3"
+#define CLOUD_TTS_ERROR_URL_FORMAT \
+    "http://101.35.224.118:8080/assets/roles/banban/tts_error/tts_error%02u.mp3"
+#define LOCAL_TTS_ERROR_PATH "U:/tts_error.mp3"
 
 static int g_sim_ready = 0;
 
@@ -50,14 +50,14 @@ static uint8_t sender_buf[PCM_FRAME_LENGTH]={0};
 static int16 mp3_buff[2*PCM_FRAME_LENGTH];
 static int16 *ppbuff=mp3_buff;
 
-static int onmessage_status = -1;
-volatile int tts_playing = 0;
-unsigned char wss_connect_start = 0;
-static volatile int player_mq_couter = 0;
-/* Set while a barge-in prompt bridges the old and new sessions.  The old
- * server response can still deliver URLs after its player queue was cleared. */
-static volatile int barge_in_transition = 0;
-static unsigned char audio_in_task_cancel = 0;
+static int onmessage_status = -1;
+volatile int tts_playing = 0;
+unsigned char wss_connect_start = 0;
+static volatile int player_mq_couter = 0;
+/* Set while a barge-in prompt bridges the old and new sessions.  The old
+ * server response can still deliver URLs after its player queue was cleared. */
+static volatile int barge_in_transition = 0;
+static volatile unsigned char audio_in_task_cancel = 0;
 
 static unsigned char xiaoice_wss_connect_status = 0; 
 static unsigned char wss_connect_counter = 0;
@@ -66,22 +66,24 @@ int fota_start = 0;
 int mqtt_task_run_flag = 0;
 
 extern int sub_topic_status;
-extern int mqtt_task_run_flag;
-static unsigned char shine_init = 0;
-extern int fisrt_net_start;
-extern u8 power_down_status;
-extern int aofei_mp3_player_get_result(void);
-
-void sendAuthentication(void);
-void vad_start_func(void);
-void xiaoice_close_all_audio(void);
+extern int mqtt_task_run_flag;
+static unsigned char shine_init = 0;
+extern int fisrt_net_start;
+extern u8 power_down_status;
+extern int aofei_mp3_player_get_result(void);
+extern int vad_leave_message_is_active(void);
+extern int vad_cancel_leave_message(void);
+
+void sendAuthentication(void);
+void vad_start_func(void);
+void xiaoice_close_all_audio(void);
 
 int audio_packet_type = 0;
 
 
-typedef struct ice_work_item_s
-{
-    int type;  //0-tts_out 1-session_continue 2-session_end 3-云端超时提示（失败时播放本地兜底）
+typedef struct ice_work_item_s
+{
+    int type;  //0-tts_out 1-session_continue 2-session_end 3-云端超时提示（失败时播放本地兜底）
     int error;
 	int del_url;
     void *data; 
@@ -117,11 +119,11 @@ static librws_app_t app = &_app;
 static ql_sem_t ws_connect_sem = NULL;
 extern int fota_start;
 
-#define OTA_AUDIO_STOP_TIMEOUT_MS 3000
-#define OTA_AUDIO_STOP_POLL_MS      20
-
-void xiaoice_clear_player_mq(void);
-
+#define OTA_AUDIO_STOP_TIMEOUT_MS 3000
+#define OTA_AUDIO_STOP_POLL_MS      20
+
+void xiaoice_clear_player_mq(void);
+
 
 static int tts_play_status = 0;//1播放中  2播放完成
 
@@ -204,12 +206,12 @@ static void remove_underscores(char *str) {
     *dst = '\0'; // 添加字符串结束符
 }
 
-/* 生成 banban 角色的随机云端对话超时提示音 URL。 */
-static int build_random_tts_error_url(char *url, size_t url_size)
-{
-    static uint32_t random_sequence = 0;
-    static unsigned int last_random_index = 10;
-    uint32_t random_value;
+/* 生成 banban 角色的随机云端对话超时提示音 URL。 */
+static int build_random_tts_error_url(char *url, size_t url_size)
+{
+    static uint32_t random_sequence = 0;
+    static unsigned int last_random_index = 10;
+    uint32_t random_value;
     unsigned int random_index;
     int written;
 
@@ -238,32 +240,32 @@ static int build_random_tts_error_url(char *url, size_t url_size)
         return -1;
     }
 
-    printf("random cloud tts error index:%02u, url:%s\n", random_index, url);
-    return 1;
-}
-
-/*
- * 后台下发任意角色的对话超时 URL 时，统一替换成 banban 的随机云端资源。
- * 非对话超时 URL 保持不变。
- */
-static int randomize_tts_error_url(char *url, size_t url_size)
-{
-    char *file_name;
-
-    if(url == NULL || url_size == 0){
-        return 0;
-    }
-
-    file_name = strrchr(url, '/');
-    file_name = (file_name == NULL) ? url : file_name + 1;
-
-    if(strncmp(file_name, "tts_error", strlen("tts_error")) != 0 &&
-       strncmp(file_name, "tts_err", strlen("tts_err")) != 0){
-        return 0;
-    }
-
-    return build_random_tts_error_url(url, url_size);
-}
+    printf("random cloud tts error index:%02u, url:%s\n", random_index, url);
+    return 1;
+}
+
+/*
+ * 后台下发任意角色的对话超时 URL 时，统一替换成 banban 的随机云端资源。
+ * 非对话超时 URL 保持不变。
+ */
+static int randomize_tts_error_url(char *url, size_t url_size)
+{
+    char *file_name;
+
+    if(url == NULL || url_size == 0){
+        return 0;
+    }
+
+    file_name = strrchr(url, '/');
+    file_name = (file_name == NULL) ? url : file_name + 1;
+
+    if(strncmp(file_name, "tts_error", strlen("tts_error")) != 0 &&
+       strncmp(file_name, "tts_err", strlen("tts_err")) != 0){
+        return 0;
+    }
+
+    return build_random_tts_error_url(url, url_size);
+}
 
 void add_file_to_play_player_queue(char *path, int type)
 {
@@ -312,21 +314,21 @@ void add_file_to_play_player_queue(char *path, int type)
 
 
 //如果录音后 超过10s 还没有收到信息 提示本轮对话结束
-static void onmessage_text(rws_socket socket, const char *text, const unsigned int len)
-{
-		if (fota_start == 1) {
-			printf("ignore websocket audio during OTA\r\n");
-			return;
-		}
-
-		/* The new session id is generated only when recording starts.  Until
-		 * then, every TTS message on this socket belongs to the interrupted
-		 * session and must not be allowed to refill the cleared player queue. */
-		if (barge_in_transition) {
-			printf("drop stale websocket message during barge-in\r\n");
-			return;
-		}
-
+static void onmessage_text(rws_socket socket, const char *text, const unsigned int len)
+{
+		if (fota_start == 1) {
+			printf("ignore websocket audio during OTA\r\n");
+			return;
+		}
+
+		/* The new session id is generated only when recording starts.  Until
+		 * then, every TTS message on this socket belongs to the interrupted
+		 * session and must not be allowed to refill the cleared player queue. */
+		if (barge_in_transition) {
+			printf("drop stale websocket message during barge-in\r\n");
+			return;
+		}
+
 	//	printf("message(txt), %d(byte): %s ", len, text);
 		
 		char url_buff[128]={0x00};
@@ -351,6 +353,7 @@ static void onmessage_text(rws_socket socket, const char *text, const unsigned i
 			char *ret_str2 = NULL;
 			size_t url_len = 0;
 			int is_tts_error_url = 0;
+			int is_message_ok_url = 0;
 			if(ret_str1 == NULL){
 				ret_str1 = strstr(text, "http://");
 			}
@@ -367,6 +370,14 @@ static void onmessage_text(rws_socket socket, const char *text, const unsigned i
 				}
 				memcpy(url_buff, ret_str1, url_len);
 				url_buff[url_len] = '\0';
+				is_message_ok_url = strstr(url_buff, "message_ok") != NULL;
+
+				/* A cancelled NFC transaction may still have a response in flight.
+				 * Never let that acknowledgement enter a normal conversation. */
+				if(is_message_ok_url && !vad_leave_message_is_active()){
+					printf("drop stale leave message ack\r\n");
+					return;
+				}
 
 				/* 两个角色统一使用 banban 的 HTTP 随机超时提示音 */
 				is_tts_error_url = randomize_tts_error_url(url_buff, sizeof(url_buff));
@@ -391,7 +402,7 @@ static void onmessage_text(rws_socket socket, const char *text, const unsigned i
 					tts_play_status = 2;
 				}
 
-				if(is_tts_error_url == 1 || strstr(url_buff, "message_ok")!=NULL){
+				if(is_tts_error_url == 1 || is_message_ok_url){
 					printf("tts_error or message_ok\n");
 					vad_set_speech_timeout_flag(1);
 				}
@@ -400,8 +411,8 @@ static void onmessage_text(rws_socket socket, const char *text, const unsigned i
 				if(tts_playing==2) return;
 				
 		
-				/* 对话超时提示使用 type=3，云端播放失败时由播放线程切换到本地兜底。 */
-				add_file_to_play_player_queue(url_buff, is_tts_error_url == 1 ? 3 : 0);
+				/* 对话超时提示使用 type=3，云端播放失败时由播放线程切换到本地兜底。 */
+				add_file_to_play_player_queue(url_buff, is_tts_error_url == 1 ? 3 : 0);
 			}
 		}
 
@@ -772,12 +783,12 @@ void generateSessionId(void)
 
 
 #if 1
-void sendStartPacket(void) {
-    generateSessionId();
-	/* sessionId now identifies the new conversation.  Late URLs from the old
-	 * conversation will fail the existing session-id check below. */
-	barge_in_transition = 0;
-	generateDeviceId();
+void sendStartPacket(void) {
+    generateSessionId();
+	/* sessionId now identifies the new conversation.  Late URLs from the old
+	 * conversation will fail the existing session-id check below. */
+	barge_in_transition = 0;
+	generateDeviceId();
     gSequenceNumber = 0;
 
     AudioPacket startPacket = {0};
@@ -894,23 +905,23 @@ void sendRecordingAudioPacket(AudioPacket *packet) {
 
 extern const char *mp3_file[];
 
-int app_play_tips_file(int index)
-{
-	int file_valid_check = 0;
-	int access_ret = 0;
-
-	printf("xiaoice_play_index:%d\n", index);
-
-		//正在录音过程中不允许播放
-//	if(xiaoice_get_tts_play_status() || get_start_mic_flag()==1) return -1;
-    if(power_down_status!=1){
-		if(vad_audio_is_recording() || tts_playing!=0 || fota_start==1 || power_down_status==2 || get_sleep_mode()==1) {
-			printf("tips blocked:index=%d vad=%d tts=%d fota=%d power=%d sleep=%d\r\n",
-			       index, vad_audio_is_recording(), tts_playing, fota_start,
-			       power_down_status, get_sleep_mode());
-			return -1;
-		}
-    }
+int app_play_tips_file(int index)
+{
+	int file_valid_check = 0;
+	int access_ret = 0;
+
+	printf("xiaoice_play_index:%d\n", index);
+
+		//正在录音过程中不允许播放
+//	if(xiaoice_get_tts_play_status() || get_start_mic_flag()==1) return -1;
+    if(power_down_status!=1){
+		if(vad_audio_is_recording() || tts_playing!=0 || fota_start==1 || power_down_status==2 || get_sleep_mode()==1) {
+			printf("tips blocked:index=%d vad=%d tts=%d fota=%d power=%d sleep=%d\r\n",
+			       index, vad_audio_is_recording(), tts_playing, fota_start,
+			       power_down_status, get_sleep_mode());
+			return -1;
+		}
+    }
 	
 	char filename_buff[64]={0x00};
 	
@@ -919,18 +930,18 @@ int app_play_tips_file(int index)
 	if(index==0) {
 		file_valid_check = 0;
 	}else{ 
-		access_ret = ql_access(filename_buff, 0);
-		if(access_ret==0){
-			file_valid_check = 1;
-		}else{
-			printf("tips file missing:index=%d path=%s access=%d\r\n",
-			       index, filename_buff, access_ret);
-			file_valid_check = 2;
-		}
+		access_ret = ql_access(filename_buff, 0);
+		if(access_ret==0){
+			file_valid_check = 1;
+		}else{
+			printf("tips file missing:index=%d path=%s access=%d\r\n",
+			       index, filename_buff, access_ret);
+			file_valid_check = 2;
+		}
 	}
 
 	if(file_valid_check!=2){
-		if(index==18 || index==19){
+		if(index==18 || index==19){
 			add_file_to_play_player_queue(filename_buff, 2);
 		}else{
 			add_file_to_play_player_queue(filename_buff, 0);
@@ -942,53 +953,53 @@ int app_play_tips_file(int index)
 	return 0;
 }
 
-int app_play_ota_tips_file(void)
-{
-	char filename_buff[64] = {0};
-	uint32_t timeout;
-
-	/* OTA owns the audio path from this point on. Stop the active conversation
-	 * before queuing the prompt, otherwise recording/TTS rejects or masks it. */
-	if (vad_audio_is_recording()) {
-		vad_audio_stop_record();
-	}
-
-	if (get_xiaoice_connected_status() == 2) {
-		talkingq_ws_disconnect();
-	}
-
-	tts_playing = 2;
-	xiaoice_clear_player_mq();
-
-	timeout = hal_os_up_time_ms();
-	while (player_mq_couter > 0) {
-		ql_rtos_task_sleep_ms(OTA_AUDIO_STOP_POLL_MS);
-		if (hal_os_up_time_ms() - timeout > OTA_AUDIO_STOP_TIMEOUT_MS) {
-			printf("OTA wait previous audio timeout, counter:%d\r\n",
-			       player_mq_couter);
-			return -3;
-		}
-	}
-
-	tts_play_status = 0;
-	tts_playing = 0;
-
-	/* Normal prompts are blocked while fota_start is set. Queue the dedicated
-	 * OTA prompt directly after the old conversation has fully stopped. */
-	snprintf(filename_buff, sizeof(filename_buff), "U:/%s", mp3_file[3]);
-	printf("xiaoice_play_index:3\n");
-	if (ql_access(filename_buff, 0) != 0) {
-		return -2;
-	}
-
-	add_file_to_play_player_queue(filename_buff, 0);
-	return player_mq_couter > 0 ? 0 : -1;
-}
-
-void app_player_thread(void *arvg)
-{
-	ice_work_item *player_item=NULL;
-	int play_ret;
+int app_play_ota_tips_file(void)
+{
+	char filename_buff[64] = {0};
+	uint32_t timeout;
+
+	/* OTA owns the audio path from this point on. Stop the active conversation
+	 * before queuing the prompt, otherwise recording/TTS rejects or masks it. */
+	if (vad_audio_is_recording()) {
+		vad_audio_stop_record();
+	}
+
+	if (get_xiaoice_connected_status() == 2) {
+		talkingq_ws_disconnect();
+	}
+
+	tts_playing = 2;
+	xiaoice_clear_player_mq();
+
+	timeout = hal_os_up_time_ms();
+	while (player_mq_couter > 0) {
+		ql_rtos_task_sleep_ms(OTA_AUDIO_STOP_POLL_MS);
+		if (hal_os_up_time_ms() - timeout > OTA_AUDIO_STOP_TIMEOUT_MS) {
+			printf("OTA wait previous audio timeout, counter:%d\r\n",
+			       player_mq_couter);
+			return -3;
+		}
+	}
+
+	tts_play_status = 0;
+	tts_playing = 0;
+
+	/* Normal prompts are blocked while fota_start is set. Queue the dedicated
+	 * OTA prompt directly after the old conversation has fully stopped. */
+	snprintf(filename_buff, sizeof(filename_buff), "U:/%s", mp3_file[3]);
+	printf("xiaoice_play_index:3\n");
+	if (ql_access(filename_buff, 0) != 0) {
+		return -2;
+	}
+
+	add_file_to_play_player_queue(filename_buff, 0);
+	return player_mq_couter > 0 ? 0 : -1;
+}
+
+void app_player_thread(void *arvg)
+{
+	ice_work_item *player_item=NULL;
+	int play_ret;
 	
     while(1){
       ql_rtos_queue_wait(player_queue, (u8*)(&player_item), sizeof(ice_work_item *), 0xFFFFFFFF);
@@ -998,29 +1009,29 @@ void app_player_thread(void *arvg)
 			  if(player_item->data){
 			  	 // printf("xiaoice player! %s\n",player_item->data);		  
 
-				  app_set_app_operation(1);
-				  aofei_mp3_player(player_item->data);
-				  play_ret = aofei_mp3_player_get_result();
-				  app_set_app_operation(1);
-
-				  if(player_mq_couter>0) player_mq_couter--;
-
-				  /*
-				   * type=3 只用于对话超时提示。云端连接、HTTP、读取或播放器初始化
-				   * 失败时才排入本地兜底；如果用户已经开始下一轮录音，则丢弃旧提示。
-				   */
-				  if(play_ret != 0 && player_item->type == 3 &&
-				     !vad_audio_is_recording() && tts_playing_ret() != 2){
-					  printf("cloud tts error play failed:%d, fallback:%s\r\n",
-					         play_ret, LOCAL_TTS_ERROR_PATH);
-					  if(ql_access(LOCAL_TTS_ERROR_PATH, 0) == 0){
-						  add_file_to_play_player_queue(LOCAL_TTS_ERROR_PATH, 0);
-					  }else{
-						  printf("local tts error fallback missing:%s\r\n", LOCAL_TTS_ERROR_PATH);
-					  }
-				  }
-
-				  if(player_mq_couter==0){
+				  app_set_app_operation(1);
+				  aofei_mp3_player(player_item->data);
+				  play_ret = aofei_mp3_player_get_result();
+				  app_set_app_operation(1);
+
+				  if(player_mq_couter>0) player_mq_couter--;
+
+				  /*
+				   * type=3 只用于对话超时提示。云端连接、HTTP、读取或播放器初始化
+				   * 失败时才排入本地兜底；如果用户已经开始下一轮录音，则丢弃旧提示。
+				   */
+				  if(play_ret != 0 && player_item->type == 3 &&
+				     !vad_audio_is_recording() && tts_playing_ret() != 2){
+					  printf("cloud tts error play failed:%d, fallback:%s\r\n",
+					         play_ret, LOCAL_TTS_ERROR_PATH);
+					  if(ql_access(LOCAL_TTS_ERROR_PATH, 0) == 0){
+						  add_file_to_play_player_queue(LOCAL_TTS_ERROR_PATH, 0);
+					  }else{
+						  printf("local tts error fallback missing:%s\r\n", LOCAL_TTS_ERROR_PATH);
+					  }
+				  }
+
+				  if(player_mq_couter==0){
 					  printf("player_mq_couter==0\r\n");
 
 					   if(tts_play_status==2){
@@ -1060,46 +1071,69 @@ void app_player_thread(void *arvg)
 
 int tts_play_interrupt_func(void)
 {
-	int playback_interrupted = 0;
-
+	int playback_interrupted = 0;
+	int leave_message_cancelled = 0;
+
+	/* A short press explicitly starts a normal conversation.  Cancel an NFC
+	 * leave-message before the generic "already recording" guard, otherwise a
+	 * pending/recording leave-message can leak packetType=4 into this session. */
+	leave_message_cancelled = vad_cancel_leave_message();
+	if(leave_message_cancelled){
+		uint64_t cancel_start;
+
+		/* Let the sender task discard buffered PCM and its encoder without
+		 * emitting END.  Only then close the old leave-message socket. */
+		audio_in_task_cancel = 1;
+		ql_rtos_semaphore_release(sender_sem);
+		cancel_start = hal_os_up_time_ms();
+		while(audio_in_task_cancel &&
+		      hal_os_up_time_ms() - cancel_start < OTA_AUDIO_STOP_TIMEOUT_MS){
+			ql_rtos_task_sleep_ms(20);
+		}
+		if(audio_in_task_cancel){
+			printf("wait leave message sender cancel timeout\r\n");
+		}
+		talkingq_ws_disconnect();
+	}
+
 	//如果在vad状态，则直接返回
 	if(vad_audio_is_recording()||g_mic_working) return -1;
 
 	g_mic_working = 1;//等待转为vad状态就清0 或者拾音超时 或者ws没有正常
-
-	/* Remember whether this key press is interrupting TTS.  tts_playing == 2
-	 * is only a cancellation signal for the old player; it must not leak into
-	 * the next conversation, otherwise prompts and error audio are discarded. */
-	playback_interrupted = (tts_playing != 0 || player_mq_couter > 0);
-	barge_in_transition = playback_interrupted;
-	xiaoice_close_all_audio();
+
+	/* Remember whether this key press is interrupting TTS.  tts_playing == 2
+	 * is only a cancellation signal for the old player; it must not leak into
+	 * the next conversation, otherwise prompts and error audio are discarded. */
+	playback_interrupted = (tts_playing != 0 || player_mq_couter > 0);
+	barge_in_transition = playback_interrupted;
+	xiaoice_close_all_audio();
 
 	talkingq_ws_connect();
 
-	if(get_xiaoice_connected_status()!=2) {
-		/* Let the caller play its local network-error prompt. */
-		tts_play_status = 0;
-		tts_playing = 0;
-		barge_in_transition = 0;
-		g_mic_working = 0;
-		return -2;
-	}
-
-	if(playback_interrupted){
-		printf("barge_in_prompt\n");
-		/* The old player has finished consuming the cancellation state. */
-		tts_play_status = 0;
-		tts_playing = 0;
-		/* Use the same local-prompt flow as ai_active_zh.mp3. Queue type 2
-		 * starts VAD after playback; fall back to VAD if the file is absent. */
-		if(app_play_tips_file(19) != 0){
-			vad_start_func();
-		}
+	if(get_xiaoice_connected_status()!=2) {
+		/* Let the caller play its local network-error prompt. */
+		tts_play_status = 0;
+		tts_playing = 0;
+		barge_in_transition = 0;
+		g_mic_working = 0;
+		return -2;
+	}
+
+	if(playback_interrupted){
+		printf("barge_in_prompt\n");
+		/* The old player has finished consuming the cancellation state. */
+		tts_play_status = 0;
+		tts_playing = 0;
+		/* Use the same local-prompt flow as ai_active_zh.mp3. Queue type 2
+		 * starts VAD after playback; fall back to VAD if the file is absent. */
+		if(app_play_tips_file(19) != 0){
+			vad_start_func();
+		}
 	}else{
 		app_play_tips_file(18);
 	}
-
-	return 0;
+
+	return 0;
 }
 
 void stop_player_func(void)
@@ -1107,7 +1141,7 @@ void stop_player_func(void)
 	vad_audio_stop_record();
 	xiaoice_close_all_audio();
 
-	tts_play_status = 0;
+	tts_play_status = 0;
 	tts_playing = 0;
 }
 
@@ -1227,19 +1261,19 @@ static void check_connect_thread_entry(void *parameter)
 			printf("tts_error_set_speech1\n");
 			vad_set_speech_timeout_flag(1);
 			
-			/*
-			 * 后台没有返回任何音频 URL 时，也先由设备生成随机云端提示音。
-			 * type=3 会在 HTTP 下载或播放初始化失败时自动切换到本地音频。
-			 */
-			{
-				char cloud_tts_error_url[128] = {0};
-				if(build_random_tts_error_url(cloud_tts_error_url,
-				                              sizeof(cloud_tts_error_url)) == 1){
-					add_file_to_play_player_queue(cloud_tts_error_url, 3);
-				}else{
-					app_play_tips_file(9);
-				}
-			}
+			/*
+			 * 后台没有返回任何音频 URL 时，也先由设备生成随机云端提示音。
+			 * type=3 会在 HTTP 下载或播放初始化失败时自动切换到本地音频。
+			 */
+			{
+				char cloud_tts_error_url[128] = {0};
+				if(build_random_tts_error_url(cloud_tts_error_url,
+				                              sizeof(cloud_tts_error_url)) == 1){
+					add_file_to_play_player_queue(cloud_tts_error_url, 3);
+				}else{
+					app_play_tips_file(9);
+				}
+			}
 		}
 	#endif
 	
@@ -1258,7 +1292,7 @@ void xiaoice_clear_player_mq(void)
 			 free(player_item->data);
 			 player_item->data = NULL;
 		  }
-		  if(player_mq_couter>0) player_mq_couter--;
+		  if(player_mq_couter>0) player_mq_couter--;
 		  free(player_item);
 		  player_item = NULL;
 	  }
@@ -1284,16 +1318,16 @@ void xiaoice_close_all_audio(void)
 
 		timeout = hal_os_up_time_ms();
 
-		/* The queue counter also includes the item currently owned by the
-		 * player task.  Wait for that item to observe tts_playing == 2 and
-		 * finish cleanup before the cancellation state is reused/reset. */
-		while(player_mq_couter>0){
+		/* The queue counter also includes the item currently owned by the
+		 * player task.  Wait for that item to observe tts_playing == 2 and
+		 * finish cleanup before the cancellation state is reused/reset. */
+		while(player_mq_couter>0){
 			ql_rtos_task_sleep_ms(20);
-			if(hal_os_up_time_ms()-timeout>OTA_AUDIO_STOP_TIMEOUT_MS){
-				printf("wait player stop timeout, counter:%d\r\n",
-				       player_mq_couter);
-				break;
-			}
+			if(hal_os_up_time_ms()-timeout>OTA_AUDIO_STOP_TIMEOUT_MS){
+				printf("wait player stop timeout, counter:%d\r\n",
+				       player_mq_couter);
+				break;
+			}
 		}
 	}
 #endif
@@ -1313,10 +1347,35 @@ void app_sender_thread(void *arvg)
     		ql_rtos_semaphore_wait(sender_sem, 0xFFFFFFFF);
 
 sender_loop:
+			if(audio_in_task_cancel){
+				left = rb_get_space_used(ring_sender);
+				while(left > 0){
+					int discard_len = left > PCM_FRAME_LENGTH ? PCM_FRAME_LENGTH : left;
+					hal_os_rb_read(ring_sender, sender_buf, discard_len, 0);
+					left = rb_get_space_used(ring_sender);
+				}
+				if(shine_init != 0){
+					shine_close(sh_shine);
+					sh_shine = NULL;
+					shine_init = 0;
+				}
+				audio_packet_type = 0;
+				audio_in_task_cancel = 0;
+				printf("leave message sender cancelled\r\n");
+				continue;
+			}
 
-			if(!app_get_data_call_net_status()) continue;
-
-			if(shine_init==0){
+			if(!app_get_data_call_net_status()) continue;
+
+			/* One recording can release the semaphore many times while the sender
+			 * drains the ring buffer.  A stale token must never create an empty
+			 * session after the previous encoder has already been closed. */
+			left = rb_get_space_used(ring_sender);
+			if(shine_init == 0 && left <= 0){
+				continue;
+			}
+
+			if(shine_init==0){
 				shine_set_config_mpeg_defaults(&(sh_config.mpeg)); //--init sh_config struct
 				sh_config.wave.channels=PCM_MONO; //PCM_STEREO;//PCM_MONO;
 				sh_config.wave.samplerate=16000;//valid samplerates: 44100,48000,32000,22050,24000,16000,11025,12000,8000
@@ -1366,10 +1425,13 @@ sender_loop:
 					sh_shine = NULL;
 				}
 
-				shine_init = 0;
-				
-				audio_packet_type = 0;
-			}
+				shine_init = 0;
+
+				audio_packet_type = 0;
+				/* Discard semaphore tokens already represented by the PCM that this
+				 * session drained.  Recording has stopped, so no new frames race here. */
+				while(ql_rtos_semaphore_wait(sender_sem, 0) == 0) {}
+			}
 	}
 }