misakiotoha/Bionic_sphere
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

1. 进一步拓展了语音识别，目前可以从sd卡导入模型，避免了model文件占用flash分区大小

Browse Source 
2. 进一步修改了语音识别，关闭了关键词唤醒功能，只保留了指令识别功能
3. 构建了业务层的基本框架(增加了底层驱动对于的C++兼容)，业务代码采用C++编写，启用了RTTI(运行时类型识别)
This commit is contained in:
Misaki
2025-09-03 00:19:14 +08:00
parent ce0998c1c6
commit 5d79f88918
42 changed files with 822 additions and 100 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Lib/MIC_Driver/MIC_Speech.c
+144 -29
View File
@@ -107,6 +107,117 @@ static void detect_hander(AppSpeech *self)
break;
}
// if (res->wakeup_state == WAKENET_DETECTED) {
// ESP_LOGI(TAG, "WAKEWORD DETECTED\n");
// multinet->clean(model_data); // clean all status of multinet
// LCD_Backlight_original = LCD_Backlight;
// } else if (res->wakeup_state == WAKENET_CHANNEL_VERIFIED) {
// ESP_LOGI(TAG, "AFE_FETCH_CHANNEL_VERIFIED, channel index: %d\n", res->trigger_channel_id);
// ESP_LOGI(TAG, ">>> Say your command <<<");
// self->detected = true;
// self->afe_handle->disable_wakenet(afe_data);
// LCD_Backlight = 35;
//
// }
esp_mn_state_t mn_state = multinet->detect(model_data, res->data);
if (mn_state == ESP_MN_STATE_DETECTING) {
self->command = COMMAND_NOT_DETECTED;
continue;
} else if (mn_state == ESP_MN_STATE_DETECTED) {
esp_mn_results_t *mn_result = multinet->get_results(model_data);
// for (int i = 0; i < mn_result->num; i++) {
// ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
// i+1, mn_result->command_id[i], mn_result->phrase_id[i], mn_result->string, mn_result->prob[i]);
// }
ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
1, mn_result->command_id[0], mn_result->phrase_id[0], mn_result->string, mn_result->prob[0]);
switch (mn_result->command_id[0]) {
case 0:
LCD_Backlight = 100;
break;
case 1:
LCD_Backlight = 30;
break;
case 2:
LCD_Backlight = 0;
break;
case 3:
LCD_Backlight = 100;
break;
case 4:
play_Music_Flag = 1;
break;
default: printf("Unknown Command!\r\n"); break;
}
self->command = (command_word_t)mn_result->command_id[0];
// self->afe_handle->enable_wakenet(afe_data);
// self->detected = false;
// self->afe_handle->disable_wakenet(afe_data); // 停止唤醒
self->detected = true;
ESP_LOGI(TAG, ">>> Say your command <<<");
self->command = COMMAND_TIMEOUT;
} else if (mn_state == ESP_MN_STATE_TIMEOUT) {
esp_mn_results_t *mn_result = multinet->get_results(model_data);
ESP_LOGI(TAG, "timeout, string:%s\n", mn_result->string);
self->command = COMMAND_TIMEOUT;
// self->afe_handle->enable_wakenet(afe_data);
self->detected = false;
ESP_LOGI(TAG, ">>> Waiting to be waken up <<<");
LCD_Backlight = LCD_Backlight_original;
if(play_Music_Flag){
play_Music_Flag = 0;
if(ACTIVE_TRACK_CNT)
_lv_demo_music_resume();
else
printf("No MP3 file found in SD card!\r\n");
}
}
}
if (model_data) {
multinet->destroy(model_data);
model_data = NULL;
}
self->afe_handle->destroy(afe_data);
vTaskDelete(NULL);
}
// 下面的函数是上面的备份，使用前需要在idf.py menuconfig中先配置打开唤醒模型
static void detect_handler_continuous(AppSpeech *self)
{
esp_afe_sr_data_t *afe_data = self->afe_data;
int afe_chunksize = self->afe_handle->get_fetch_chunksize(afe_data);
#if defined(CONFIG_SR_MN_CN_MULTINET5_RECOGNITION_QUANT8) || defined(CONFIG_SR_MN_CN_MULTINET6_QUANT) || defined(CONFIG_SR_MN_CN_MULTINET6_AC_QUANT)
char *mn_name = esp_srmodel_filter(self->models, ESP_MN_PREFIX, ESP_MN_CHINESE);
#else
char *mn_name = esp_srmodel_filter(self->models, ESP_MN_PREFIX, ESP_MN_ENGLISH);
#endif // CONFIG_IDF_TARGET_ESP32S3
ESP_LOGI(TAG, "multinet:%s\n", mn_name);
esp_mn_iface_t *multinet = esp_mn_handle_from_name(mn_name);
model_iface_data_t *model_data = multinet->create(mn_name, 6000);
esp_mn_commands_update_from_sdkconfig(multinet, model_data); // Add speech commands from sdkconfig
int mu_chunksize = multinet->get_samp_chunksize(model_data);
assert(mu_chunksize == afe_chunksize);
// FILE *fp = fopen("/sdcard/out", "w");
// if (fp == NULL) ESP_LOGE(TAG,"can not open file\n");
//print active speech commands
multinet->print_active_speech_commands(model_data);
ESP_LOGI(TAG, "Ready");
self->detected = false;
while (true)
{
afe_fetch_result_t* res = self->afe_handle->fetch(afe_data);
if (!res || res->ret_value == ESP_FAIL) {
ESP_LOGE(TAG, "fetch error!\n");
break;
}
if (res->wakeup_state == WAKENET_DETECTED) {
ESP_LOGI(TAG, "WAKEWORD DETECTED\n");
multinet->clean(model_data); // clean all status of multinet
@@ -117,7 +228,7 @@ static void detect_hander(AppSpeech *self)
self->detected = true;
self->afe_handle->disable_wakenet(afe_data);
LCD_Backlight = 35;
}
if (self->detected) {
@@ -129,33 +240,33 @@ static void detect_hander(AppSpeech *self)
} else if (mn_state == ESP_MN_STATE_DETECTED) {
esp_mn_results_t *mn_result = multinet->get_results(model_data);
// for (int i = 0; i < mn_result->num; i++) {
// ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
// ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
// i+1, mn_result->command_id[i], mn_result->phrase_id[i], mn_result->string, mn_result->prob[i]);
// }
ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
ESP_LOGI(TAG, "TOP %d, command_id: %d, phrase_id: %d, string:%s prob: %f\n",
1, mn_result->command_id[0], mn_result->phrase_id[0], mn_result->string, mn_result->prob[0]);
switch (mn_result->command_id[0]) {
case 0:
LCD_Backlight = 100;
case 0:
LCD_Backlight = 100;
break;
case 1:
LCD_Backlight = 30;
case 1:
LCD_Backlight = 30;
break;
case 2:
LCD_Backlight = 0;
case 2:
LCD_Backlight = 0;
break;
case 3:
LCD_Backlight = 100;
case 3:
LCD_Backlight = 100;
break;
case 4:
play_Music_Flag = 1;
case 4:
play_Music_Flag = 1;
break;
default: printf("Unknown Command!\r\n"); break;
}
self->command = (command_word_t)mn_result->command_id[0];
// self->afe_handle->enable_wakenet(afe_data);
// self->detected = false;
self->afe_handle->disable_wakenet(afe_data);
self->detected = true;
ESP_LOGI(TAG, ">>> Say your command <<<");
@@ -171,9 +282,9 @@ static void detect_hander(AppSpeech *self)
if(play_Music_Flag){
play_Music_Flag = 0;
if(ACTIVE_TRACK_CNT)
_lv_demo_music_resume();
_lv_demo_music_resume();
else
printf("No MP3 file found in SD card!\r\n");
printf("No MP3 file found in SD card!\r\n");
}
}
}
@@ -187,32 +298,32 @@ static void detect_hander(AppSpeech *self)
}
// 初始化
void MIC_Speech_init()
{
MIC_Speech.afe_handle = &ESP_AFE_SR_HANDLE;
MIC_Speech.detected = false;
MIC_Speech.command = COMMAND_TIMEOUT;
MIC_Speech.models = esp_srmodel_init("model"); // 这边配置为SD卡当中的文件路径
MIC_Speech.models = esp_srmodel_init("/sdcard/srmodels"); // 这边配置为SD卡当中的文件路径
i2s_init(I2S_NUM_1, 16000, 2, 32);
// sd_card_mount("/sdcard");
afe_config_t afe_config = {
.aec_init = true,
.se_init = true,
.vad_init = true,
.wakenet_init = true,
.aec_init = true, // 回声消除(当用户在播放音频的时候使用语音识别可以有效提告识别率)
.se_init = true, // 降噪
.vad_init = true, // VDA(语音活动检测)，用于检测当前是否处于说话状态，如果是，就将音频数据发送给 multinet
.wakenet_init = false, // 关闭唤醒词
.voice_communication_init = false,
.voice_communication_agc_init = false,
.voice_communication_agc_gain = 15,
.vad_mode = VAD_MODE_3,
.wakenet_model_name = NULL,
.vad_mode = VAD_MODE_0, /*VAD_MODE_3,*/ // VAD 灵敏度等级
.wakenet_model_name = NULL, // 不再指定 wakenet
.wakenet_model_name_2 = NULL,
.wakenet_mode = DET_MODE_2CH_90,
.wakenet_mode = DET_MODE_2CH_90, // 0 = 关闭
.afe_mode = SR_MODE_LOW_COST,
.afe_perferred_core = 0,
.afe_perferred_priority = 5,
.afe_ringbuf_size = 50,
.afe_ringbuf_size = 50, // AFE ringbuffer 环形缓冲区大小
.memory_alloc_mode = AFE_MEMORY_ALLOC_MORE_PSRAM,
.afe_linear_gain = 1.0,
.agc_mode = AFE_MN_PEAK_AGC_MODE_2,
@@ -222,7 +333,7 @@ void MIC_Speech_init()
.ref_num = 1,
.sample_rate = 16000,
},
.debug_init = false,
.debug_init = false, // afe内部调试
.debug_hook = {{AFE_DEBUG_HOOK_MASE_TASK_IN, NULL}, {AFE_DEBUG_HOOK_FETCH_TASK_IN, NULL}},
};
afe_config.aec_init = false;
@@ -235,6 +346,10 @@ void MIC_Speech_init()
afe_config.pcm_config.sample_rate = 16000;
afe_config.wakenet_model_name = esp_srmodel_filter(MIC_Speech.models, ESP_WN_PREFIX, NULL);
MIC_Speech.afe_data = MIC_Speech.afe_handle->create_from_config(&afe_config);
xTaskCreatePinnedToCore((TaskFunction_t)feed_handler, "App/SR/Feed", 4 * 1024, &MIC_Speech, 5, NULL, 1);
xTaskCreatePinnedToCore((TaskFunction_t)detect_hander, "App/SR/Detect", 5 * 1024, &MIC_Speech, 5, NULL, 1);
// 注意两个任务被分配了不同的核心与优先级，这是为了防止AFE（Audio Front-End）内部环形缓冲区溢出
// 也就是“喂数据线程” 比 “取数据线程” 跑得快，生产 > 消费，经典的生产者消费者问题
// 但即使这么做了，由于i2s在开始读取数据的时候，识别模型还没加载完成，因此在开始阶段必然会出现环形缓冲区满的警告，问题不大
xTaskCreatePinnedToCore((TaskFunction_t)feed_handler, "App/SR/Feed", 4 * 1024, &MIC_Speech, 4, NULL, 0);
xTaskCreatePinnedToCore((TaskFunction_t)detect_hander, "App/SR/Detect", 5 * 1024, &MIC_Speech, 6, NULL, 1);
}