fix multinet model for v2 (#1208)

main/audio/wake_words/custom_wake_word.cc

@@ -1,11 +1,13 @@
 #include "custom_wake_word.h"
 #include "audio_service.h"
 #include "system_info.h"
+#include "assets.h"
 
 #include <esp_log.h>
-#include "esp_mn_iface.h"
-#include "esp_mn_models.h"
-#include "esp_mn_speech_commands.h"
+#include <esp_mn_iface.h>
+#include <esp_mn_models.h>
+#include <esp_mn_speech_commands.h>
+#include <cJSON.h>
 
 
 #define TAG "CustomWakeWord"
@@ -34,13 +36,66 @@ CustomWakeWord::~CustomWakeWord() {
     }
 }
 
+void CustomWakeWord::ParseWakenetModelConfig() {
+    // Read index.json
+    auto& assets = Assets::GetInstance();
+    void* ptr = nullptr;
+    size_t size = 0;
+    if (!assets.GetAssetData("index.json", ptr, size)) {
+        ESP_LOGE(TAG, "Failed to read index.json");
+        return;
+    }
+    cJSON* root = cJSON_ParseWithLength(static_cast<char*>(ptr), size);
+    if (root == nullptr) {
+        ESP_LOGE(TAG, "Failed to parse index.json");
+        return;
+    }
+    cJSON* multinet_model = cJSON_GetObjectItem(root, "multinet_model");
+    if (cJSON_IsObject(multinet_model)) {
+        cJSON* language = cJSON_GetObjectItem(multinet_model, "language");
+        cJSON* duration = cJSON_GetObjectItem(multinet_model, "duration");
+        cJSON* threshold = cJSON_GetObjectItem(multinet_model, "threshold");
+        cJSON* commands = cJSON_GetObjectItem(multinet_model, "commands");
+        if (cJSON_IsString(language)) {
+            language_ = language->valuestring;
+        }
+        if (cJSON_IsNumber(duration)) {
+            duration_ = duration->valueint;
+        }
+        if (cJSON_IsNumber(threshold)) {
+            threshold_ = threshold->valuedouble;
+        }
+        if (cJSON_IsArray(commands)) {
+            for (int i = 0; i < cJSON_GetArraySize(commands); i++) {
+                cJSON* command = cJSON_GetArrayItem(commands, i);
+                if (cJSON_IsObject(command)) {
+                    cJSON* command_name = cJSON_GetObjectItem(command, "command");
+                    cJSON* text = cJSON_GetObjectItem(command, "text");
+                    cJSON* action = cJSON_GetObjectItem(command, "action");
+                    if (cJSON_IsString(command_name) && cJSON_IsString(text) && cJSON_IsString(action)) {
+                        commands_.push_back({command_name->valuestring, text->valuestring, action->valuestring});
+                        ESP_LOGI(TAG, "Command: %s, Text: %s, Action: %s", command_name->valuestring, text->valuestring, action->valuestring);
+                    }
+                }
+            }
+        }
+    }
+    cJSON_Delete(root);
+}
+
+
 bool CustomWakeWord::Initialize(AudioCodec* codec, srmodel_list_t* models_list) {
     codec_ = codec;
+    commands_.clear();
 
     if (models_list == nullptr) {
+        language_ = "cn";
         models_ = esp_srmodel_init("model");
+        threshold_ = CONFIG_CUSTOM_WAKE_WORD_THRESHOLD / 100.0f;
+        commands_.push_back({CONFIG_CUSTOM_WAKE_WORD, CONFIG_CUSTOM_WAKE_WORD_DISPLAY, "wake"});
     } else {
         models_ = models_list;
+        ParseWakenetModelConfig();
     }
 
     if (models_ == nullptr || models_->num == -1) {
@@ -49,19 +104,20 @@ bool CustomWakeWord::Initialize(AudioCodec* codec, srmodel_list_t* models_list)
     }
 
     // 初始化 multinet (命令词识别)
-    mn_name_ = esp_srmodel_filter(models_, ESP_MN_PREFIX, ESP_MN_CHINESE);
+    mn_name_ = esp_srmodel_filter(models_, ESP_MN_PREFIX, language_.c_str());
     if (mn_name_ == nullptr) {
         ESP_LOGE(TAG, "Failed to initialize multinet, mn_name is nullptr");
         ESP_LOGI(TAG, "Please refer to https://pcn7cs20v8cr.feishu.cn/wiki/CpQjwQsCJiQSWSkYEvrcxcbVnwh to add custom wake word");
         return false;
     }
 
-    ESP_LOGI(TAG, "multinet: %s", mn_name_);
     multinet_ = esp_mn_handle_from_name(mn_name_);
-    multinet_model_data_ = multinet_->create(mn_name_, 3000);  // 3 秒超时
-    multinet_->set_det_threshold(multinet_model_data_, CONFIG_CUSTOM_WAKE_WORD_THRESHOLD / 100.0f);
+    multinet_model_data_ = multinet_->create(mn_name_, duration_);
+    multinet_->set_det_threshold(multinet_model_data_, threshold_);
     esp_mn_commands_clear();
-    esp_mn_commands_add(1, CONFIG_CUSTOM_WAKE_WORD);
+    for (int i = 0; i < commands_.size(); i++) {
+        esp_mn_commands_add(i + 1, commands_[i].command.c_str());
+    }
     esp_mn_commands_update();
     
     multinet_->print_active_speech_commands(multinet_model_data_);
@@ -104,16 +160,18 @@ void CustomWakeWord::Feed(const std::vector<int16_t>& data) {
         return;
     } else if (mn_state == ESP_MN_STATE_DETECTED) {
         esp_mn_results_t *mn_result = multinet_->get_results(multinet_model_data_);
-        ESP_LOGI(TAG, "Custom wake word detected: command_id=%d, string=%s, prob=%f", 
-                mn_result->command_id[0], mn_result->string, mn_result->prob[0]);
+        for (int i = 0; i < mn_result->num && running_; i++) {
+            ESP_LOGI(TAG, "Custom wake word detected: command_id=%d, string=%s, prob=%f", 
+                    mn_result->command_id[i], mn_result->string, mn_result->prob[i]);
+            auto& command = commands_[mn_result->command_id[i] - 1];
+            if (command.action == "wake") {
+                last_detected_wake_word_ = command.text;
+                running_ = false;
                 
-        if (mn_result->command_id[0] == 1) {
-            last_detected_wake_word_ = CONFIG_CUSTOM_WAKE_WORD_DISPLAY;
-        }
-        running_ = false;
-        
-        if (wake_word_detected_callback_) {
-            wake_word_detected_callback_(last_detected_wake_word_);
+                if (wake_word_detected_callback_) {
+                    wake_word_detected_callback_(last_detected_wake_word_);
+                }
+            }
         }
         multinet_->clean(multinet_model_data_);
     } else if (mn_state == ESP_MN_STATE_TIMEOUT) {

main/audio/wake_words/custom_wake_word.h

@@ -33,11 +33,21 @@ public:
     const std::string& GetLastDetectedWakeWord() const { return last_detected_wake_word_; }
 
 private:
+    struct Command {
+        std::string command;
+        std::string text;
+        std::string action;
+    };
+
     // multinet 相关成员变量
     esp_mn_iface_t* multinet_ = nullptr;
     model_iface_data_t* multinet_model_data_ = nullptr;
     srmodel_list_t *models_ = nullptr;
     char* mn_name_ = nullptr;
+    std::string language_ = "cn";
+    int duration_ = 3000;
+    float threshold_ = 0.2;
+    std::deque<Command> commands_;
  
     std::function<void(const std::string& wake_word)> wake_word_detected_callback_;
     AudioCodec* codec_ = nullptr;
@@ -53,6 +63,7 @@ private:
     std::condition_variable wake_word_cv_;
 
     void StoreWakeWordData(const std::vector<int16_t>& data);
+    void ParseWakenetModelConfig();
 };
 
 #endif

scripts/build_default_assets.py

@@ -156,21 +156,44 @@ def copy_directory(src, dst):
         return False
 
 
-def process_wakenet_model(wakenet_model_dir, build_dir, assets_dir):
-    """Process wakenet_model parameter"""
-    if not wakenet_model_dir:
+def process_sr_models(wakenet_model_dir, multinet_model_dirs, build_dir, assets_dir):
+    """Process SR models (wakenet and multinet) and generate srmodels.bin"""
+    if not wakenet_model_dir and not multinet_model_dirs:
         return None
     
-    # Copy input directory to build directory
-    wakenet_build_dir = os.path.join(build_dir, "wakenet_model")
-    if os.path.exists(wakenet_build_dir):
-        shutil.rmtree(wakenet_build_dir)
-    copy_directory(wakenet_model_dir, os.path.join(wakenet_build_dir, os.path.basename(wakenet_model_dir)))
+    # Create SR models build directory
+    sr_models_build_dir = os.path.join(build_dir, "srmodels")
+    if os.path.exists(sr_models_build_dir):
+        shutil.rmtree(sr_models_build_dir)
+    os.makedirs(sr_models_build_dir)
+    
+    models_processed = 0
+    
+    # Copy wakenet model if available
+    if wakenet_model_dir:
+        wakenet_name = os.path.basename(wakenet_model_dir)
+        wakenet_dst = os.path.join(sr_models_build_dir, wakenet_name)
+        if copy_directory(wakenet_model_dir, wakenet_dst):
+            models_processed += 1
+            print(f"Added wakenet model: {wakenet_name}")
+    
+    # Copy multinet models if available
+    if multinet_model_dirs:
+        for multinet_model_dir in multinet_model_dirs:
+            multinet_name = os.path.basename(multinet_model_dir)
+            multinet_dst = os.path.join(sr_models_build_dir, multinet_name)
+            if copy_directory(multinet_model_dir, multinet_dst):
+                models_processed += 1
+                print(f"Added multinet model: {multinet_name}")
+    
+    if models_processed == 0:
+        print("Warning: No SR models were successfully processed")
+        return None
     
     # Use pack_models function to generate srmodels.bin
-    srmodels_output = os.path.join(wakenet_build_dir, "srmodels.bin")
+    srmodels_output = os.path.join(sr_models_build_dir, "srmodels.bin")
     try:
-        pack_models(wakenet_build_dir, "srmodels.bin")
+        pack_models(sr_models_build_dir, "srmodels.bin")
         print(f"Generated: {srmodels_output}")
         # Copy srmodels.bin to assets directory
         copy_file(srmodels_output, os.path.join(assets_dir, "srmodels.bin"))
@@ -180,6 +203,11 @@ def process_wakenet_model(wakenet_model_dir, build_dir, assets_dir):
         return None
 
 
+def process_wakenet_model(wakenet_model_dir, build_dir, assets_dir):
+    """Process wakenet_model parameter (legacy compatibility function)"""
+    return process_sr_models(wakenet_model_dir, None, build_dir, assets_dir)
+
+
 def process_text_font(text_font_file, assets_dir):
     """Process text_font parameter"""
     if not text_font_file:
@@ -250,7 +278,7 @@ def process_extra_files(extra_files_dir, assets_dir):
     return extra_files_list
 
 
-def generate_index_json(assets_dir, srmodels, text_font, emoji_collection, extra_files=None):
+def generate_index_json(assets_dir, srmodels, text_font, emoji_collection, extra_files=None, multinet_model_info=None):
     """Generate index.json file"""
     index_data = {
         "version": 1
@@ -268,6 +296,9 @@ def generate_index_json(assets_dir, srmodels, text_font, emoji_collection, extra
     if extra_files:
         index_data["extra_files"] = extra_files
     
+    if multinet_model_info:
+        index_data["multinet_model"] = multinet_model_info
+    
     # Write index.json
     index_path = os.path.join(assets_dir, "index.json")
     with open(index_path, 'w', encoding='utf-8') as f:
@@ -434,6 +465,132 @@ def read_wakenet_from_sdkconfig(sdkconfig_path):
     return models[0] if models else None
 
 
+def read_multinet_from_sdkconfig(sdkconfig_path):
+    """
+    Read multinet models from sdkconfig (based on movemodel.py logic)
+    Returns a list of multinet model names
+    """
+    if not os.path.exists(sdkconfig_path):
+        print(f"Warning: sdkconfig file not found: {sdkconfig_path}")
+        return []
+        
+    with io.open(sdkconfig_path, "r") as f:
+        models_string = ''
+        for label in f:
+            label = label.strip("\n")
+            if 'CONFIG_SR_MN' in label and label[0] != '#':
+                models_string += label
+
+    models = []
+    if "CONFIG_SR_MN_CN_MULTINET3_SINGLE_RECOGNITION" in models_string:
+        models.append('mn3_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET4_5_SINGLE_RECOGNITION_QUANT8" in models_string:
+        models.append('mn4q8_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET4_5_SINGLE_RECOGNITION" in models_string:
+        models.append('mn4_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET5_RECOGNITION_QUANT8" in models_string:
+        models.append('mn5q8_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET6_QUANT" in models_string:
+        models.append('mn6_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET6_AC_QUANT" in models_string:
+        models.append('mn6_cn_ac')
+    elif "CONFIG_SR_MN_CN_MULTINET7_QUANT" in models_string:
+        models.append('mn7_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET7_AC_QUANT" in models_string:
+        models.append('mn7_cn_ac')
+
+    if "CONFIG_SR_MN_EN_MULTINET5_SINGLE_RECOGNITION_QUANT8" in models_string:
+        models.append('mn5q8_en')
+    elif "CONFIG_SR_MN_EN_MULTINET5_SINGLE_RECOGNITION" in models_string:
+        models.append('mn5_en')
+    elif "CONFIG_SR_MN_EN_MULTINET6_QUANT" in models_string:
+        models.append('mn6_en')
+    elif "CONFIG_SR_MN_EN_MULTINET7_QUANT" in models_string:
+        models.append('mn7_en')
+
+    if "MULTINET6" in models_string or "MULTINET7" in models_string:
+        models.append('fst')
+
+    return models
+
+
+def read_custom_wake_word_from_sdkconfig(sdkconfig_path):
+    """
+    Read custom wake word configuration from sdkconfig
+    Returns a dict with custom wake word info or None if not configured
+    """
+    if not os.path.exists(sdkconfig_path):
+        print(f"Warning: sdkconfig file not found: {sdkconfig_path}")
+        return None
+        
+    config_values = {}
+    with io.open(sdkconfig_path, "r") as f:
+        for line in f:
+            line = line.strip("\n")
+            if line.startswith('#') or '=' not in line:
+                continue
+                
+            # Check for custom wake word configuration
+            if 'CONFIG_USE_CUSTOM_WAKE_WORD=y' in line:
+                config_values['use_custom_wake_word'] = True
+            elif 'CONFIG_CUSTOM_WAKE_WORD=' in line and not line.startswith('#'):
+                # Extract string value (remove quotes)
+                value = line.split('=', 1)[1].strip('"')
+                config_values['wake_word'] = value
+            elif 'CONFIG_CUSTOM_WAKE_WORD_DISPLAY=' in line and not line.startswith('#'):
+                # Extract string value (remove quotes)
+                value = line.split('=', 1)[1].strip('"')
+                config_values['display'] = value
+            elif 'CONFIG_CUSTOM_WAKE_WORD_THRESHOLD=' in line and not line.startswith('#'):
+                # Extract numeric value
+                value = line.split('=', 1)[1]
+                try:
+                    config_values['threshold'] = int(value)
+                except ValueError:
+                    try:
+                        config_values['threshold'] = float(value)
+                    except ValueError:
+                        print(f"Warning: Invalid threshold value: {value}")
+                        config_values['threshold'] = 20  # default (will be converted to 0.2)
+    
+    # Return config only if custom wake word is enabled and required fields are present
+    if (config_values.get('use_custom_wake_word', False) and 
+        'wake_word' in config_values and 
+        'display' in config_values and 
+        'threshold' in config_values):
+        return {
+            'wake_word': config_values['wake_word'],
+            'display': config_values['display'],
+            'threshold': config_values['threshold'] / 100.0  # Convert to decimal (20 -> 0.2)
+        }
+    
+    return None
+
+
+def get_language_from_multinet_models(multinet_models):
+    """
+    Determine language from multinet model names
+    Returns 'cn', 'en', or None
+    """
+    if not multinet_models:
+        return None
+    
+    # Check for Chinese models
+    cn_indicators = ['_cn', 'cn_']
+    en_indicators = ['_en', 'en_']
+    
+    has_cn = any(any(indicator in model for indicator in cn_indicators) for model in multinet_models)
+    has_en = any(any(indicator in model for indicator in en_indicators) for model in multinet_models)
+    
+    # If both or neither, default to cn
+    if has_cn and not has_en:
+        return 'cn'
+    elif has_en and not has_cn:
+        return 'en'
+    else:
+        return 'cn'  # Default to Chinese
+
+
 def get_wakenet_model_path(model_name, esp_sr_model_path):
     """
     Get the full path to the wakenet model directory
@@ -449,6 +606,25 @@ def get_wakenet_model_path(model_name, esp_sr_model_path):
         return None
 
 
+def get_multinet_model_paths(model_names, esp_sr_model_path):
+    """
+    Get the full paths to the multinet model directories
+    Returns a list of valid model paths
+    """
+    if not model_names:
+        return []
+    
+    valid_paths = []
+    for model_name in model_names:
+        multinet_model_path = os.path.join(esp_sr_model_path, 'multinet_model', model_name)
+        if os.path.exists(multinet_model_path):
+            valid_paths.append(multinet_model_path)
+        else:
+            print(f"Warning: Multinet model directory not found: {multinet_model_path}")
+    
+    return valid_paths
+
+
 def get_text_font_path(builtin_text_font, xiaozhi_fonts_path):
     """
     Get the text font path if needed
@@ -485,7 +661,7 @@ def get_emoji_collection_path(default_emoji_collection, xiaozhi_fonts_path):
         return None
 
 
-def build_assets_integrated(wakenet_model_path, text_font_path, emoji_collection_path, extra_files_path, output_path):
+def build_assets_integrated(wakenet_model_path, multinet_model_paths, text_font_path, emoji_collection_path, extra_files_path, output_path, multinet_model_info=None):
     """
     Build assets using integrated functions (no external dependencies)
     """
@@ -503,13 +679,13 @@ def build_assets_integrated(wakenet_model_path, text_font_path, emoji_collection
         print("Starting to build assets...")
         
         # Process each component
-        srmodels = process_wakenet_model(wakenet_model_path, temp_build_dir, assets_dir) if wakenet_model_path else None
+        srmodels = process_sr_models(wakenet_model_path, multinet_model_paths, temp_build_dir, assets_dir) if (wakenet_model_path or multinet_model_paths) else None
         text_font = process_text_font(text_font_path, assets_dir) if text_font_path else None
         emoji_collection = process_emoji_collection(emoji_collection_path, assets_dir) if emoji_collection_path else None
         extra_files = process_extra_files(extra_files_path, assets_dir) if extra_files_path else None
         
         # Generate index.json
-        generate_index_json(assets_dir, srmodels, text_font, emoji_collection, extra_files)
+        generate_index_json(assets_dir, srmodels, text_font, emoji_collection, extra_files, multinet_model_info)
         
         # Generate config.json for packing
         config_path = generate_config_json(temp_build_dir, assets_dir)
@@ -578,9 +754,19 @@ def main():
     print(f"  emoji_collection: {args.emoji_collection}")
     print(f"  output: {args.output}")
     
-    # Read wakenet model from sdkconfig
+    # Read SR models from sdkconfig
     wakenet_model_name = read_wakenet_from_sdkconfig(args.sdkconfig)
+    multinet_model_names = read_multinet_from_sdkconfig(args.sdkconfig)
+    
+    # Get model paths
     wakenet_model_path = get_wakenet_model_path(wakenet_model_name, args.esp_sr_model_path)
+    multinet_model_paths = get_multinet_model_paths(multinet_model_names, args.esp_sr_model_path)
+    
+    # Print model information
+    if wakenet_model_name:
+        print(f"  wakenet model: {wakenet_model_name}")
+    if multinet_model_names:
+        print(f"  multinet models: {', '.join(multinet_model_names)}")
     
     # Get text font path if needed
     text_font_path = get_text_font_path(args.builtin_text_font, args.xiaozhi_fonts_path)
@@ -591,9 +777,34 @@ def main():
     # Get extra files path if provided
     extra_files_path = args.extra_files
     
+    # Read custom wake word configuration
+    custom_wake_word_config = read_custom_wake_word_from_sdkconfig(args.sdkconfig)
+    multinet_model_info = None
+    
+    if custom_wake_word_config and multinet_model_names:
+        # Determine language from multinet models
+        language = get_language_from_multinet_models(multinet_model_names)
+        
+        # Build multinet_model info structure
+        multinet_model_info = {
+            "language": language,
+            "duration": 3000,  # Default duration in ms
+            "threshold": custom_wake_word_config['threshold'],
+            "commands": [
+                {
+                    "command": custom_wake_word_config['wake_word'],
+                    "text": custom_wake_word_config['display'],
+                    "action": "wake"
+                }
+            ]
+        }
+        print(f"  custom wake word: {custom_wake_word_config['wake_word']} ({custom_wake_word_config['display']})")
+        print(f"  wake word language: {language}")
+        print(f"  wake word threshold: {custom_wake_word_config['threshold']}")
+    
     # Check if we have anything to build
-    if not wakenet_model_path and not text_font_path and not emoji_collection_path and not extra_files_path:
-        print("Warning: No assets to build (no wakenet, text font, emoji collection, or extra files)")
+    if not wakenet_model_path and not multinet_model_paths and not text_font_path and not emoji_collection_path and not extra_files_path and not multinet_model_info:
+        print("Warning: No assets to build (no SR models, text font, emoji collection, extra files, or custom wake word)")
         # Create an empty assets.bin file
         os.makedirs(os.path.dirname(args.output), exist_ok=True)
         with open(args.output, 'wb') as f:
@@ -602,8 +813,8 @@ def main():
         return
     
     # Build the assets
-    success = build_assets_integrated(wakenet_model_path, text_font_path, emoji_collection_path, 
-                                     extra_files_path, args.output)
+    success = build_assets_integrated(wakenet_model_path, multinet_model_paths, text_font_path, emoji_collection_path, 
+                                     extra_files_path, args.output, multinet_model_info)
     
     if not success:
         sys.exit(1)