refactor(voice): split ws transport protocol audio marker modules

main/CMakeLists.txt

@@ -21,6 +21,9 @@ idf_component_register(
         "domain/src/voice_interaction.c"
         "domain/src/voice_interaction_common.c"
         "domain/src/voice_interaction_ws.c"
+        "domain/src/voice_interaction_ws_protocol.c"
+        "domain/src/voice_interaction_ws_audio.c"
+        "domain/src/voice_interaction_marker.c"
         "domain/src/voice_interaction_tasks.c"
         "platform/src/platform_bootstrap.c"
         "platform/src/wifi_manager.c"

main/domain/internal/voice_interaction_internal.h

@@ -117,6 +117,10 @@ esp_err_t voice_send_directive(const char *action,
                                const char *directive,
                                bool with_dialog_id);
 
+void voice_ws_protocol_handle_text_message(const char *text, size_t len);
+void voice_ws_audio_handle_downstream_packet(const uint8_t *data, size_t len);
+void voice_marker_try_trigger_image_generation(const char *dialog_id, const char *final_text);
+
 void voice_websocket_event_handler(void *handler_args,
                                    esp_event_base_t base,
                                    int32_t event_id,

main/domain/src/voice_interaction_marker.c

@@ -0,0 +1,480 @@
+#include "voice_interaction_internal.h"
+
+#include <ctype.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "esp_log.h"
+#include "image_generation.h"
+#include "printer_protocol.h"
+#include "raster_tools.h"
+
+static const char *TAG = "voice_interaction";
+
+#define VOICE_IMAGE_MARKER_PROMPT_MAX_LEN 800
+#define VOICE_IMAGE_MARKER_FIRST_SENTENCE_MAX_LEN 1024
+#define VOICE_IMAGE_GEN_TASK_STACK 16384
+#define VOICE_IMAGE_GEN_TASK_PRIO 3
+#define VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS 45000
+#define VOICE_IMAGE_PRINT_THRESHOLD 160
+#define VOICE_IMAGE_PRINT_MAX_HEIGHT 2200
+#define VOICE_IMAGE_MARKER_TEMPLATE \
+    "%s，黑白线稿风格，白色背景，主体居中，边缘清晰，细节简洁，高对比，避免大面积阴影和灰度渐变，适配热敏打印机打印。"
+#define VOICE_IMAGE_MARKER_TEMPLATE_FALLBACK_SUBJECT "简洁卡通图案"
+
+typedef struct {
+    char prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1];
+    char dialog_id[40];
+} voice_image_generation_task_arg_t;
+
+static portMUX_TYPE s_marker_image_gen_lock = portMUX_INITIALIZER_UNLOCKED;
+static bool s_marker_image_gen_running;
+static int64_t s_marker_last_success_ms;
+static char s_marker_last_success_prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1];
+static char s_marker_last_success_dialog_id[40];
+
+static bool voice_marker_image_generation_try_acquire(const char *prompt,
+                                                      const char *dialog_id,
+                                                      int64_t now_ms,
+                                                      uint32_t *out_remain_ms) {
+    bool acquired = false;
+    uint32_t remain_ms = 0;
+    const char *safe_prompt = prompt != NULL ? prompt : "";
+    const char *safe_dialog = dialog_id != NULL ? dialog_id : "";
+
+    portENTER_CRITICAL(&s_marker_image_gen_lock);
+    if (s_marker_image_gen_running) {
+        acquired = false;
+    } else if (s_marker_last_success_ms > 0 &&
+               strcmp(safe_prompt, s_marker_last_success_prompt) == 0 &&
+               strcmp(safe_dialog, s_marker_last_success_dialog_id) == 0) {
+        int64_t elapsed_ms = now_ms - s_marker_last_success_ms;
+        if (elapsed_ms >= 0 && elapsed_ms < VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS) {
+            remain_ms = (uint32_t)(VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS - elapsed_ms);
+            acquired = false;
+        } else {
+            s_marker_image_gen_running = true;
+            acquired = true;
+        }
+    } else {
+        s_marker_image_gen_running = true;
+        acquired = true;
+    }
+    portEXIT_CRITICAL(&s_marker_image_gen_lock);
+
+    if (out_remain_ms != NULL) {
+        *out_remain_ms = remain_ms;
+    }
+    return acquired;
+}
+
+static void voice_marker_image_generation_release(void) {
+    portENTER_CRITICAL(&s_marker_image_gen_lock);
+    s_marker_image_gen_running = false;
+    portEXIT_CRITICAL(&s_marker_image_gen_lock);
+}
+
+static void voice_marker_image_generation_record_success(const char *prompt,
+                                                         const char *dialog_id,
+                                                         int64_t now_ms) {
+    const char *safe_prompt = prompt != NULL ? prompt : "";
+    const char *safe_dialog = dialog_id != NULL ? dialog_id : "";
+
+    portENTER_CRITICAL(&s_marker_image_gen_lock);
+    strlcpy(s_marker_last_success_prompt, safe_prompt, sizeof(s_marker_last_success_prompt));
+    strlcpy(s_marker_last_success_dialog_id, safe_dialog, sizeof(s_marker_last_success_dialog_id));
+    s_marker_last_success_ms = now_ms;
+    portEXIT_CRITICAL(&s_marker_image_gen_lock);
+}
+
+static void voice_log_text_preview(const char *stage, const char *text) {
+    if (stage == NULL || text == NULL || text[0] == '\0') {
+        return;
+    }
+
+    size_t len = strlen(text);
+    const int preview = 240;
+    ESP_LOGI(TAG,
+             "[stage] %s: len=%u text=%.*s%s",
+             stage,
+             (unsigned)len,
+             preview,
+             text,
+             (len > (size_t)preview) ? "..." : "");
+}
+
+static bool voice_is_space_char(char c) {
+    return isspace((unsigned char)c) != 0;
+}
+
+static size_t voice_find_first_sentence_end(const char *text, size_t len) {
+    if (text == NULL || len == 0) {
+        return 0;
+    }
+
+    for (size_t i = 0; i < len; ++i) {
+        unsigned char c = (unsigned char)text[i];
+        if (c == '.' || c == '!' || c == '?' || c == '\n' || c == '\r') {
+            return i + 1;
+        }
+
+        if (i + 2 < len) {
+            unsigned char b1 = (unsigned char)text[i];
+            unsigned char b2 = (unsigned char)text[i + 1];
+            unsigned char b3 = (unsigned char)text[i + 2];
+
+            if (b1 == 0xE3 && b2 == 0x80 && b3 == 0x82) {
+                return i + 3;
+            }
+            if (b1 == 0xEF && b2 == 0xBC &&
+                (b3 == 0x81 || b3 == 0x9F || b3 == 0x9B)) {
+                return i + 3;
+            }
+        }
+    }
+
+    return len;
+}
+
+static bool voice_extract_first_sentence_text(const char *text,
+                                              char *out_sentence,
+                                              size_t out_len) {
+    if (text == NULL || out_sentence == NULL || out_len == 0) {
+        return false;
+    }
+
+    size_t len = strlen(text);
+    size_t begin = 0;
+    while (begin < len && voice_is_space_char(text[begin])) {
+        begin++;
+    }
+    if (begin >= len) {
+        return false;
+    }
+
+    size_t end = begin + voice_find_first_sentence_end(text + begin, len - begin);
+    while (end > begin && voice_is_space_char(text[end - 1])) {
+        end--;
+    }
+    if (end <= begin) {
+        return false;
+    }
+
+    size_t copy_len = end - begin;
+    if (copy_len >= out_len) {
+        copy_len = out_len - 1;
+    }
+    memcpy(out_sentence, text + begin, copy_len);
+    out_sentence[copy_len] = '\0';
+    return copy_len > 0;
+}
+
+static bool voice_extract_marker_prompt(const char *text, char *out_prompt, size_t out_len) {
+    if (text == NULL || out_prompt == NULL || out_len == 0) {
+        return false;
+    }
+
+    const char *open = strstr(text, "/*");
+    if (open == NULL) {
+        return false;
+    }
+
+    const char *close = strstr(open + 2, "*/");
+    if (close == NULL) {
+        return false;
+    }
+
+    const char *begin = open + 2;
+    const char *end = close;
+    while (begin < end && voice_is_space_char(*begin)) {
+        begin++;
+    }
+    while (end > begin && voice_is_space_char(*(end - 1))) {
+        end--;
+    }
+    if (end <= begin) {
+        return false;
+    }
+
+    size_t prompt_len = (size_t)(end - begin);
+    if (prompt_len >= out_len) {
+        prompt_len = out_len - 1;
+    }
+    memcpy(out_prompt, begin, prompt_len);
+    out_prompt[prompt_len] = '\0';
+    return prompt_len > 0;
+}
+
+static void voice_build_marker_image_prompt(const char *subject,
+                                            char *out_prompt,
+                                            size_t out_len) {
+    if (out_prompt == NULL || out_len == 0) {
+        return;
+    }
+
+    const char *safe_subject = (subject != NULL && subject[0] != '\0') ?
+                                   subject :
+                                   VOICE_IMAGE_MARKER_TEMPLATE_FALLBACK_SUBJECT;
+    int written = snprintf(out_prompt,
+                           out_len,
+                           VOICE_IMAGE_MARKER_TEMPLATE,
+                           safe_subject);
+    if (written < 0) {
+        out_prompt[0] = '\0';
+        return;
+    }
+
+    if ((size_t)written >= out_len) {
+        out_prompt[out_len - 1] = '\0';
+    }
+}
+
+static esp_err_t voice_submit_marker_generated_image(const voice_image_generation_task_arg_t *task_arg,
+                                                     const image_generation_result_t *result,
+                                                     char *err,
+                                                     size_t err_len) {
+    if (task_arg == NULL || result == NULL || result->png == NULL || result->png_len == 0) {
+        voice_fill_err(err, err_len, "invalid image result");
+        return ESP_ERR_INVALID_ARG;
+    }
+
+    uint16_t raster_width = 0;
+    uint16_t raster_height = 0;
+    uint8_t *raster = NULL;
+    size_t raster_len = 0;
+    char decode_err[128] = {0};
+    ESP_LOGI(TAG,
+             "[stage] marker_image_print_decode_start: dialog_id=%s request_id=%s png_bytes=%u",
+             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+             result->request_id[0] != '\0' ? result->request_id : "-",
+             (unsigned)result->png_len);
+    esp_err_t decode_rc = raster_tools_convert_png_to_raster_384(result->png,
+                                                                  result->png_len,
+                                                                  true,
+                                                                  VOICE_IMAGE_PRINT_THRESHOLD,
+                                                                  false,
+                                                                  VOICE_IMAGE_PRINT_MAX_HEIGHT,
+                                                                  &raster_width,
+                                                                  &raster_height,
+                                                                  &raster,
+                                                                  &raster_len,
+                                                                  decode_err,
+                                                                  sizeof(decode_err));
+    if (decode_rc != ESP_OK) {
+        voice_fill_err(err,
+                       err_len,
+                       decode_err[0] != '\0' ? decode_err : "decode generated image failed");
+        ESP_LOGW(TAG,
+                 "[stage] marker_image_print_decode_failed: dialog_id=%s rc=0x%x msg=%s",
+                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                 (unsigned)decode_rc,
+                 err != NULL && err[0] != '\0' ? err : "decode generated image failed");
+        return decode_rc;
+    }
+
+    ESP_LOGI(TAG,
+             "[stage] marker_image_print_decode_done: dialog_id=%s raster_size=%ux%u raster_bytes=%u",
+             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+             (unsigned)raster_width,
+             (unsigned)raster_height,
+             (unsigned)raster_len);
+
+    uint32_t job_id = 0;
+    char submit_err[128] = {0};
+    esp_err_t submit_rc = printer_protocol_submit_raster_job(raster,
+                                                              raster_len,
+                                                              raster_width,
+                                                              raster_height,
+                                                              NULL,
+                                                              &job_id,
+                                                              submit_err,
+                                                              sizeof(submit_err));
+    if (submit_rc != ESP_OK) {
+        voice_fill_err(err,
+                       err_len,
+                       submit_err[0] != '\0' ? submit_err : "submit print job failed");
+        ESP_LOGW(TAG,
+                 "[stage] marker_image_print_submit_failed: dialog_id=%s rc=0x%x msg=%s",
+                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                 (unsigned)submit_rc,
+                 err != NULL && err[0] != '\0' ? err : "submit print job failed");
+    } else {
+        ESP_LOGI(TAG,
+                 "[stage] marker_image_print_submit_ok: dialog_id=%s request_id=%s job_id=%u",
+                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                 result->request_id[0] != '\0' ? result->request_id : "-",
+                 (unsigned)job_id);
+    }
+
+    free(raster);
+    return submit_rc;
+}
+
+static void voice_image_generation_task(void *arg) {
+    voice_image_generation_task_arg_t *task_arg = (voice_image_generation_task_arg_t *)arg;
+    if (task_arg == NULL) {
+        vTaskDelete(NULL);
+        return;
+    }
+
+    image_generation_request_t req = {
+        .prompt = task_arg->prompt,
+        .size = NULL,
+        .prompt_extend = false,
+        .has_seed = false,
+        .seed = 0,
+        .generation_timeout_ms = 0,
+        .download_timeout_ms = 0,
+    };
+
+    image_generation_result_t result = {0};
+    image_generation_result_reset(&result);
+
+    bool status_poll_paused = false;
+    char gen_err[160] = {0};
+    ESP_LOGI(TAG,
+             "[stage] marker_image_gen_start: dialog_id=%s prompt_len=%u",
+             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+             (unsigned)strlen(task_arg->prompt));
+    printer_protocol_set_status_poll_paused(true);
+    status_poll_paused = true;
+    esp_err_t rc = image_generation_generate_png(&req, &result, gen_err, sizeof(gen_err));
+    printer_protocol_set_status_poll_paused(false);
+    status_poll_paused = false;
+    if (rc == ESP_OK) {
+        ESP_LOGI(TAG,
+                 "[stage] marker_image_gen_done: dialog_id=%s request_id=%s size=%ux%u png_bytes=%u",
+                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                 result.request_id[0] != '\0' ? result.request_id : "-",
+                 (unsigned)result.width,
+                 (unsigned)result.height,
+                 (unsigned)result.png_len);
+
+        char submit_err[160] = {0};
+        esp_err_t submit_rc = voice_submit_marker_generated_image(task_arg,
+                                                                  &result,
+                                                                  submit_err,
+                                                                  sizeof(submit_err));
+        if (submit_rc == ESP_OK) {
+            voice_marker_image_generation_record_success(task_arg->prompt,
+                                                         task_arg->dialog_id,
+                                                         voice_now_ms());
+        } else {
+            ESP_LOGW(TAG,
+                     "[stage] marker_image_print_failed: dialog_id=%s rc=0x%x msg=%s",
+                     task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                     (unsigned)submit_rc,
+                     submit_err[0] != '\0' ? submit_err : "marker image print failed");
+        }
+    } else {
+        ESP_LOGW(TAG,
+                 "[stage] marker_image_gen_failed: dialog_id=%s rc=0x%x msg=%s",
+                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+                 (unsigned)rc,
+                 gen_err[0] != '\0' ? gen_err : "image generation failed");
+    }
+
+    if (status_poll_paused) {
+        printer_protocol_set_status_poll_paused(false);
+    }
+    image_generation_result_free(&result);
+    voice_marker_image_generation_release();
+    image_generation_schedule_prewarm();
+    free(task_arg);
+    vTaskDelete(NULL);
+}
+
+static void voice_schedule_marker_image_generation(const char *prompt, const char *dialog_id) {
+    if (prompt == NULL || prompt[0] == '\0') {
+        return;
+    }
+
+    size_t prompt_len = strlen(prompt);
+    if (prompt_len > VOICE_IMAGE_MARKER_PROMPT_MAX_LEN) {
+        ESP_LOGW(TAG,
+                 "[stage] marker_image_gen_skip: prompt too long, len=%u",
+                 (unsigned)prompt_len);
+        return;
+    }
+
+    uint32_t repeat_remain_ms = 0;
+    if (!voice_marker_image_generation_try_acquire(prompt,
+                                                   dialog_id,
+                                                   voice_now_ms(),
+                                                   &repeat_remain_ms)) {
+        if (repeat_remain_ms > 0) {
+            ESP_LOGI(TAG,
+                     "[stage] marker_image_gen_skip: duplicate prompt suppressed, remain_ms=%u",
+                     (unsigned)repeat_remain_ms);
+        } else {
+            ESP_LOGI(TAG, "[stage] marker_image_gen_skip: previous marker task running");
+        }
+        return;
+    }
+
+    voice_image_generation_task_arg_t *task_arg =
+        (voice_image_generation_task_arg_t *)calloc(1, sizeof(*task_arg));
+    if (task_arg == NULL) {
+        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: no memory for task arg");
+        voice_marker_image_generation_release();
+        return;
+    }
+
+    strlcpy(task_arg->prompt, prompt, sizeof(task_arg->prompt));
+    if (dialog_id != NULL) {
+        strlcpy(task_arg->dialog_id, dialog_id, sizeof(task_arg->dialog_id));
+    }
+
+    BaseType_t ok = xTaskCreate(voice_image_generation_task,
+                                "voice_img_gen",
+                                VOICE_IMAGE_GEN_TASK_STACK,
+                                task_arg,
+                                VOICE_IMAGE_GEN_TASK_PRIO,
+                                NULL);
+    if (ok != pdPASS) {
+        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: create task failed");
+        voice_marker_image_generation_release();
+        free(task_arg);
+        return;
+    }
+
+    ESP_LOGI(TAG,
+             "[stage] marker_image_gen_queued: dialog_id=%s prompt=%s",
+             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
+             task_arg->prompt);
+}
+
+void voice_marker_try_trigger_image_generation(const char *dialog_id, const char *final_text) {
+    if (final_text == NULL || final_text[0] == '\0') {
+        return;
+    }
+
+    char first_sentence[VOICE_IMAGE_MARKER_FIRST_SENTENCE_MAX_LEN] = {0};
+    if (!voice_extract_first_sentence_text(final_text, first_sentence, sizeof(first_sentence))) {
+        return;
+    }
+
+    char marker_subject[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1] = {0};
+    if (!voice_extract_marker_prompt(first_sentence, marker_subject, sizeof(marker_subject))) {
+        return;
+    }
+
+    char templated_prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1] = {0};
+    voice_build_marker_image_prompt(marker_subject,
+                                    templated_prompt,
+                                    sizeof(templated_prompt));
+    if (templated_prompt[0] == '\0') {
+        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: build prompt failed");
+        return;
+    }
+
+    ESP_LOGI(TAG,
+             "[stage] marker_image_gen_detected: first_sentence=%s",
+             first_sentence);
+    ESP_LOGI(TAG,
+             "[stage] marker_image_gen_subject: keyword=%s",
+             marker_subject);
+    voice_log_text_preview("marker_image_gen_prompt_templated", templated_prompt);
+    voice_schedule_marker_image_generation(templated_prompt, dialog_id);
+}

main/domain/src/voice_interaction_ws.c

@@ -1,8 +1,5 @@
 #include "voice_interaction_internal.h"
 
-#include <ctype.h>
-#include <inttypes.h>
-#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
@@ -11,92 +8,10 @@
 #include "cJSON.h"
 #undef cJSON_realloc
 #undef cJSON_malloc
-#include "esp_audio_dec.h"
-#include "esp_audio_enc.h"
 #include "esp_log.h"
-#include "image_generation.h"
-#include "printer_protocol.h"
-#include "raster_tools.h"
-#include "voice_audio.h"
 
 static const char *TAG = "voice_interaction";
 
-#define VOICE_IMAGE_MARKER_PROMPT_MAX_LEN 800
-#define VOICE_IMAGE_MARKER_FIRST_SENTENCE_MAX_LEN 1024
-#define VOICE_IMAGE_GEN_TASK_STACK 16384
-#define VOICE_IMAGE_GEN_TASK_PRIO 3
-#define VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS 45000
-#define VOICE_IMAGE_PRINT_THRESHOLD 160
-#define VOICE_IMAGE_PRINT_MAX_HEIGHT 2200
-#define VOICE_IMAGE_MARKER_TEMPLATE \
-    "%s，黑白线稿风格，白色背景，主体居中，边缘清晰，细节简洁，高对比，避免大面积阴影和灰度渐变，适配热敏打印机打印。"
-#define VOICE_IMAGE_MARKER_TEMPLATE_FALLBACK_SUBJECT "简洁卡通图案"
-
-typedef struct {
-    char prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1];
-    char dialog_id[40];
-} voice_image_generation_task_arg_t;
-
-static portMUX_TYPE s_marker_image_gen_lock = portMUX_INITIALIZER_UNLOCKED;
-static bool s_marker_image_gen_running;
-static int64_t s_marker_last_success_ms;
-static char s_marker_last_success_prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1];
-static char s_marker_last_success_dialog_id[40];
-
-static bool voice_marker_image_generation_try_acquire(const char *prompt,
-                                                      const char *dialog_id,
-                                                      int64_t now_ms,
-                                                      uint32_t *out_remain_ms) {
-    bool acquired = false;
-    uint32_t remain_ms = 0;
-    const char *safe_prompt = prompt != NULL ? prompt : "";
-    const char *safe_dialog = dialog_id != NULL ? dialog_id : "";
-
-    portENTER_CRITICAL(&s_marker_image_gen_lock);
-    if (s_marker_image_gen_running) {
-        acquired = false;
-    } else if (s_marker_last_success_ms > 0 &&
-               strcmp(safe_prompt, s_marker_last_success_prompt) == 0 &&
-               strcmp(safe_dialog, s_marker_last_success_dialog_id) == 0) {
-        int64_t elapsed_ms = now_ms - s_marker_last_success_ms;
-        if (elapsed_ms >= 0 && elapsed_ms < VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS) {
-            remain_ms = (uint32_t)(VOICE_IMAGE_MARKER_REPEAT_SUPPRESS_MS - elapsed_ms);
-            acquired = false;
-        } else {
-            s_marker_image_gen_running = true;
-            acquired = true;
-        }
-    } else {
-        s_marker_image_gen_running = true;
-        acquired = true;
-    }
-    portEXIT_CRITICAL(&s_marker_image_gen_lock);
-
-    if (out_remain_ms != NULL) {
-        *out_remain_ms = remain_ms;
-    }
-    return acquired;
-}
-
-static void voice_marker_image_generation_release(void) {
-    portENTER_CRITICAL(&s_marker_image_gen_lock);
-    s_marker_image_gen_running = false;
-    portEXIT_CRITICAL(&s_marker_image_gen_lock);
-}
-
-static void voice_marker_image_generation_record_success(const char *prompt,
-                                                         const char *dialog_id,
-                                                         int64_t now_ms) {
-    const char *safe_prompt = prompt != NULL ? prompt : "";
-    const char *safe_dialog = dialog_id != NULL ? dialog_id : "";
-
-    portENTER_CRITICAL(&s_marker_image_gen_lock);
-    strlcpy(s_marker_last_success_prompt, safe_prompt, sizeof(s_marker_last_success_prompt));
-    strlcpy(s_marker_last_success_dialog_id, safe_dialog, sizeof(s_marker_last_success_dialog_id));
-    s_marker_last_success_ms = now_ms;
-    portEXIT_CRITICAL(&s_marker_image_gen_lock);
-}
-
 static void *voice_cjson_malloc_hook(size_t size) {
     return voice_malloc_prefer_psram(size);
 }
@@ -253,679 +168,6 @@ esp_err_t voice_send_directive(const char *action,
     return err;
 }
 
-static void voice_handle_downstream_packet(const uint8_t *data, size_t len) {
-    if (data == NULL || len == 0) {
-        return;
-    }
-
-    if (!voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-        return;
-    }
-
-    void *opus_dec = s_voice.opus_dec;
-    uint8_t *pcm_buf = s_voice.pcm_rx_buf;
-    size_t pcm_buf_size = s_voice.pcm_rx_buf_size;
-    voice_unlock();
-
-    if (opus_dec == NULL || pcm_buf == NULL || pcm_buf_size == 0) {
-        return;
-    }
-
-    esp_audio_dec_in_raw_t raw = {
-        .buffer = (uint8_t *)data,
-        .len = (uint32_t)len,
-        .consumed = 0,
-    };
-
-    while (raw.len > 0) {
-        esp_audio_dec_out_frame_t out = {
-            .buffer = pcm_buf,
-            .len = (uint32_t)pcm_buf_size,
-            .needed_size = 0,
-            .decoded_size = 0,
-        };
-        esp_audio_dec_info_t dec_info = {0};
-
-        esp_audio_err_t ret = esp_opus_dec_decode(opus_dec, &raw, &out, &dec_info);
-        if (ret == ESP_AUDIO_ERR_BUFF_NOT_ENOUGH) {
-            if (out.needed_size > pcm_buf_size) {
-                uint8_t *new_buf = (uint8_t *)voice_realloc_prefer_psram(pcm_buf, out.needed_size);
-                if (new_buf == NULL) {
-                    ESP_LOGE(TAG, "No memory to extend pcm rx buffer to %u", out.needed_size);
-                    break;
-                }
-                if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-                    s_voice.pcm_rx_buf = new_buf;
-                    s_voice.pcm_rx_buf_size = out.needed_size;
-                    voice_unlock();
-                }
-                pcm_buf = new_buf;
-                pcm_buf_size = out.needed_size;
-                continue;
-            }
-            break;
-        }
-
-        if (ret != ESP_AUDIO_ERR_OK) {
-            ESP_LOGW(TAG, "opus decode failed: %d", ret);
-            break;
-        }
-
-        if (out.decoded_size > 0) {
-            char audio_err[64] = {0};
-            size_t samples = (size_t)out.decoded_size / VOICE_SAMPLE_BYTES;
-            esp_err_t write_err = voice_audio_write_pcm((const int16_t *)out.buffer,
-                                                        samples,
-                                                        VOICE_AUDIO_IO_TIMEOUT_MS,
-                                                        audio_err,
-                                                        sizeof(audio_err));
-            if (write_err == ESP_OK && voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-                int64_t now_ms = voice_now_ms();
-                int64_t pcm_ms = ((int64_t)samples * 1000 +
-                                  (int64_t)CONFIG_TQ_VOICE_SAMPLE_RATE - 1) /
-                                 (int64_t)CONFIG_TQ_VOICE_SAMPLE_RATE;
-                int64_t base_ms = s_voice.playback_deadline_ms > now_ms ?
-                                  s_voice.playback_deadline_ms : now_ms;
-                s_voice.playback_deadline_ms = base_ms + pcm_ms + VOICE_AUDIO_DRAIN_MARGIN_MS;
-                s_voice.last_downstream_ms = now_ms;
-                voice_unlock();
-            }
-        }
-
-        if (raw.consumed == 0 || raw.consumed > raw.len) {
-            break;
-        }
-        raw.buffer += raw.consumed;
-        raw.len -= raw.consumed;
-    }
-
-    if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-        s_voice.downstream_packets++;
-        s_voice.last_event_ms = voice_now_ms();
-        voice_unlock();
-    }
-}
-
-static const char *voice_get_json_str(cJSON *obj, const char *name) {
-    if (obj == NULL || name == NULL) {
-        return NULL;
-    }
-    cJSON *item = cJSON_GetObjectItemCaseSensitive(obj, name);
-    if (cJSON_IsString(item) && item->valuestring != NULL) {
-        return item->valuestring;
-    }
-    return NULL;
-}
-
-static bool voice_get_json_bool(cJSON *obj, const char *name, bool *out_value) {
-    if (obj == NULL || name == NULL || out_value == NULL) {
-        return false;
-    }
-    cJSON *item = cJSON_GetObjectItemCaseSensitive(obj, name);
-    if (!cJSON_IsBool(item)) {
-        return false;
-    }
-    *out_value = cJSON_IsTrue(item);
-    return true;
-}
-
-static void voice_log_final_text(const char *stage, const char *text) {
-    if (stage == NULL || text == NULL || text[0] == '\0') {
-        return;
-    }
-    size_t len = strlen(text);
-    const int preview = 240;
-    ESP_LOGI(TAG,
-             "[stage] %s: len=%u text=%.*s%s",
-             stage,
-             (unsigned)len,
-             preview,
-             text,
-             (len > (size_t)preview) ? "..." : "");
-}
-
-static bool voice_is_space_char(char c) {
-    return isspace((unsigned char)c) != 0;
-}
-
-static size_t voice_find_first_sentence_end(const char *text, size_t len) {
-    if (text == NULL || len == 0) {
-        return 0;
-    }
-
-    for (size_t i = 0; i < len; ++i) {
-        unsigned char c = (unsigned char)text[i];
-        if (c == '.' || c == '!' || c == '?' || c == '\n' || c == '\r') {
-            return i + 1;
-        }
-
-        if (i + 2 < len) {
-            unsigned char b1 = (unsigned char)text[i];
-            unsigned char b2 = (unsigned char)text[i + 1];
-            unsigned char b3 = (unsigned char)text[i + 2];
-
-            if (b1 == 0xE3 && b2 == 0x80 && b3 == 0x82) {
-                return i + 3;
-            }
-            if (b1 == 0xEF && b2 == 0xBC &&
-                (b3 == 0x81 || b3 == 0x9F || b3 == 0x9B)) {
-                return i + 3;
-            }
-        }
-    }
-
-    return len;
-}
-
-static bool voice_extract_first_sentence_text(const char *text,
-                                              char *out_sentence,
-                                              size_t out_len) {
-    if (text == NULL || out_sentence == NULL || out_len == 0) {
-        return false;
-    }
-
-    size_t len = strlen(text);
-    size_t begin = 0;
-    while (begin < len && voice_is_space_char(text[begin])) {
-        begin++;
-    }
-    if (begin >= len) {
-        return false;
-    }
-
-    size_t end = begin + voice_find_first_sentence_end(text + begin, len - begin);
-    while (end > begin && voice_is_space_char(text[end - 1])) {
-        end--;
-    }
-    if (end <= begin) {
-        return false;
-    }
-
-    size_t copy_len = end - begin;
-    if (copy_len >= out_len) {
-        copy_len = out_len - 1;
-    }
-    memcpy(out_sentence, text + begin, copy_len);
-    out_sentence[copy_len] = '\0';
-    return copy_len > 0;
-}
-
-static bool voice_extract_marker_prompt(const char *text, char *out_prompt, size_t out_len) {
-    if (text == NULL || out_prompt == NULL || out_len == 0) {
-        return false;
-    }
-
-    const char *open = strstr(text, "/*");
-    if (open == NULL) {
-        return false;
-    }
-
-    const char *close = strstr(open + 2, "*/");
-    if (close == NULL) {
-        return false;
-    }
-
-    const char *begin = open + 2;
-    const char *end = close;
-    while (begin < end && voice_is_space_char(*begin)) {
-        begin++;
-    }
-    while (end > begin && voice_is_space_char(*(end - 1))) {
-        end--;
-    }
-    if (end <= begin) {
-        return false;
-    }
-
-    size_t prompt_len = (size_t)(end - begin);
-    if (prompt_len >= out_len) {
-        prompt_len = out_len - 1;
-    }
-    memcpy(out_prompt, begin, prompt_len);
-    out_prompt[prompt_len] = '\0';
-    return prompt_len > 0;
-}
-
-static void voice_build_marker_image_prompt(const char *subject,
-                                            char *out_prompt,
-                                            size_t out_len) {
-    if (out_prompt == NULL || out_len == 0) {
-        return;
-    }
-
-    const char *safe_subject = (subject != NULL && subject[0] != '\0') ?
-                                   subject :
-                                   VOICE_IMAGE_MARKER_TEMPLATE_FALLBACK_SUBJECT;
-    int written = snprintf(out_prompt,
-                           out_len,
-                           VOICE_IMAGE_MARKER_TEMPLATE,
-                           safe_subject);
-    if (written < 0) {
-        out_prompt[0] = '\0';
-        return;
-    }
-
-    if ((size_t)written >= out_len) {
-        out_prompt[out_len - 1] = '\0';
-    }
-}
-
-static esp_err_t voice_submit_marker_generated_image(const voice_image_generation_task_arg_t *task_arg,
-                                                     const image_generation_result_t *result,
-                                                     char *err,
-                                                     size_t err_len) {
-    if (task_arg == NULL || result == NULL || result->png == NULL || result->png_len == 0) {
-        voice_fill_err(err, err_len, "invalid image result");
-        return ESP_ERR_INVALID_ARG;
-    }
-
-    uint16_t raster_width = 0;
-    uint16_t raster_height = 0;
-    uint8_t *raster = NULL;
-    size_t raster_len = 0;
-    char decode_err[128] = {0};
-    ESP_LOGI(TAG,
-             "[stage] marker_image_print_decode_start: dialog_id=%s request_id=%s png_bytes=%u",
-             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-             result->request_id[0] != '\0' ? result->request_id : "-",
-             (unsigned)result->png_len);
-    esp_err_t decode_rc = raster_tools_convert_png_to_raster_384(result->png,
-                                                                  result->png_len,
-                                                                  true,
-                                                                  VOICE_IMAGE_PRINT_THRESHOLD,
-                                                                  false,
-                                                                  VOICE_IMAGE_PRINT_MAX_HEIGHT,
-                                                                  &raster_width,
-                                                                  &raster_height,
-                                                                  &raster,
-                                                                  &raster_len,
-                                                                  decode_err,
-                                                                  sizeof(decode_err));
-    if (decode_rc != ESP_OK) {
-        voice_fill_err(err,
-                       err_len,
-                       decode_err[0] != '\0' ? decode_err : "decode generated image failed");
-        ESP_LOGW(TAG,
-                 "[stage] marker_image_print_decode_failed: dialog_id=%s rc=0x%x msg=%s",
-                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                 (unsigned)decode_rc,
-                 err != NULL && err[0] != '\0' ? err : "decode generated image failed");
-        return decode_rc;
-    }
-
-    ESP_LOGI(TAG,
-             "[stage] marker_image_print_decode_done: dialog_id=%s raster_size=%ux%u raster_bytes=%u",
-             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-             (unsigned)raster_width,
-             (unsigned)raster_height,
-             (unsigned)raster_len);
-
-    uint32_t job_id = 0;
-    char submit_err[128] = {0};
-    esp_err_t submit_rc = printer_protocol_submit_raster_job(raster,
-                                                              raster_len,
-                                                              raster_width,
-                                                              raster_height,
-                                                              NULL,
-                                                              &job_id,
-                                                              submit_err,
-                                                              sizeof(submit_err));
-    if (submit_rc != ESP_OK) {
-        voice_fill_err(err,
-                       err_len,
-                       submit_err[0] != '\0' ? submit_err : "submit print job failed");
-        ESP_LOGW(TAG,
-                 "[stage] marker_image_print_submit_failed: dialog_id=%s rc=0x%x msg=%s",
-                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                 (unsigned)submit_rc,
-                 err != NULL && err[0] != '\0' ? err : "submit print job failed");
-    } else {
-        ESP_LOGI(TAG,
-                 "[stage] marker_image_print_submit_ok: dialog_id=%s request_id=%s job_id=%u",
-                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                 result->request_id[0] != '\0' ? result->request_id : "-",
-                 (unsigned)job_id);
-    }
-
-    free(raster);
-    return submit_rc;
-}
-
-static void voice_image_generation_task(void *arg) {
-    voice_image_generation_task_arg_t *task_arg = (voice_image_generation_task_arg_t *)arg;
-    if (task_arg == NULL) {
-        vTaskDelete(NULL);
-        return;
-    }
-
-    image_generation_request_t req = {
-        .prompt = task_arg->prompt,
-        .size = NULL,
-        .prompt_extend = false,
-        .has_seed = false,
-        .seed = 0,
-        .generation_timeout_ms = 0,
-        .download_timeout_ms = 0,
-    };
-
-    image_generation_result_t result = {0};
-    image_generation_result_reset(&result);
-
-    bool status_poll_paused = false;
-    char gen_err[160] = {0};
-    ESP_LOGI(TAG,
-             "[stage] marker_image_gen_start: dialog_id=%s prompt_len=%u",
-             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-             (unsigned)strlen(task_arg->prompt));
-    printer_protocol_set_status_poll_paused(true);
-    status_poll_paused = true;
-    esp_err_t rc = image_generation_generate_png(&req, &result, gen_err, sizeof(gen_err));
-    printer_protocol_set_status_poll_paused(false);
-    status_poll_paused = false;
-    if (rc == ESP_OK) {
-        ESP_LOGI(TAG,
-                 "[stage] marker_image_gen_done: dialog_id=%s request_id=%s size=%ux%u png_bytes=%u",
-                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                 result.request_id[0] != '\0' ? result.request_id : "-",
-                 (unsigned)result.width,
-                 (unsigned)result.height,
-                 (unsigned)result.png_len);
-
-        char submit_err[160] = {0};
-        esp_err_t submit_rc = voice_submit_marker_generated_image(task_arg,
-                                                                  &result,
-                                                                  submit_err,
-                                                                  sizeof(submit_err));
-        if (submit_rc == ESP_OK) {
-            voice_marker_image_generation_record_success(task_arg->prompt,
-                                                         task_arg->dialog_id,
-                                                         voice_now_ms());
-        } else {
-            ESP_LOGW(TAG,
-                     "[stage] marker_image_print_failed: dialog_id=%s rc=0x%x msg=%s",
-                     task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                     (unsigned)submit_rc,
-                     submit_err[0] != '\0' ? submit_err : "marker image print failed");
-        }
-    } else {
-        ESP_LOGW(TAG,
-                 "[stage] marker_image_gen_failed: dialog_id=%s rc=0x%x msg=%s",
-                 task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-                 (unsigned)rc,
-                 gen_err[0] != '\0' ? gen_err : "image generation failed");
-    }
-
-    if (status_poll_paused) {
-        printer_protocol_set_status_poll_paused(false);
-    }
-    image_generation_result_free(&result);
-    voice_marker_image_generation_release();
-    image_generation_schedule_prewarm();
-    free(task_arg);
-    vTaskDelete(NULL);
-}
-
-static void voice_schedule_marker_image_generation(const char *prompt, const char *dialog_id) {
-    if (prompt == NULL || prompt[0] == '\0') {
-        return;
-    }
-
-    size_t prompt_len = strlen(prompt);
-    if (prompt_len > VOICE_IMAGE_MARKER_PROMPT_MAX_LEN) {
-        ESP_LOGW(TAG,
-                 "[stage] marker_image_gen_skip: prompt too long, len=%u",
-                 (unsigned)prompt_len);
-        return;
-    }
-
-    uint32_t repeat_remain_ms = 0;
-    if (!voice_marker_image_generation_try_acquire(prompt,
-                                                   dialog_id,
-                                                   voice_now_ms(),
-                                                   &repeat_remain_ms)) {
-        if (repeat_remain_ms > 0) {
-            ESP_LOGI(TAG,
-                     "[stage] marker_image_gen_skip: duplicate prompt suppressed, remain_ms=%u",
-                     (unsigned)repeat_remain_ms);
-        } else {
-            ESP_LOGI(TAG, "[stage] marker_image_gen_skip: previous marker task running");
-        }
-        return;
-    }
-
-    voice_image_generation_task_arg_t *task_arg =
-        (voice_image_generation_task_arg_t *)calloc(1, sizeof(*task_arg));
-    if (task_arg == NULL) {
-        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: no memory for task arg");
-        voice_marker_image_generation_release();
-        return;
-    }
-
-    strlcpy(task_arg->prompt, prompt, sizeof(task_arg->prompt));
-    if (dialog_id != NULL) {
-        strlcpy(task_arg->dialog_id, dialog_id, sizeof(task_arg->dialog_id));
-    }
-
-    BaseType_t ok = xTaskCreate(voice_image_generation_task,
-                                "voice_img_gen",
-                                VOICE_IMAGE_GEN_TASK_STACK,
-                                task_arg,
-                                VOICE_IMAGE_GEN_TASK_PRIO,
-                                NULL);
-    if (ok != pdPASS) {
-        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: create task failed");
-        voice_marker_image_generation_release();
-        free(task_arg);
-        return;
-    }
-
-    ESP_LOGI(TAG,
-             "[stage] marker_image_gen_queued: dialog_id=%s prompt=%s",
-             task_arg->dialog_id[0] != '\0' ? task_arg->dialog_id : "-",
-             task_arg->prompt);
-}
-
-static void voice_try_trigger_marker_image_generation(const char *dialog_id, const char *final_text) {
-    if (final_text == NULL || final_text[0] == '\0') {
-        return;
-    }
-
-    char first_sentence[VOICE_IMAGE_MARKER_FIRST_SENTENCE_MAX_LEN] = {0};
-    if (!voice_extract_first_sentence_text(final_text, first_sentence, sizeof(first_sentence))) {
-        return;
-    }
-
-    char marker_subject[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1] = {0};
-    if (!voice_extract_marker_prompt(first_sentence, marker_subject, sizeof(marker_subject))) {
-        return;
-    }
-
-    char templated_prompt[VOICE_IMAGE_MARKER_PROMPT_MAX_LEN + 1] = {0};
-    voice_build_marker_image_prompt(marker_subject,
-                                    templated_prompt,
-                                    sizeof(templated_prompt));
-    if (templated_prompt[0] == '\0') {
-        ESP_LOGW(TAG, "[stage] marker_image_gen_skip: build prompt failed");
-        return;
-    }
-
-    ESP_LOGI(TAG,
-             "[stage] marker_image_gen_detected: first_sentence=%s",
-             first_sentence);
-    ESP_LOGI(TAG,
-             "[stage] marker_image_gen_subject: keyword=%s",
-             marker_subject);
-    voice_log_final_text("marker_image_gen_prompt_templated", templated_prompt);
-    voice_schedule_marker_image_generation(templated_prompt, dialog_id);
-}
-
-static void voice_handle_output_event(cJSON *output) {
-    const char *event_name = voice_get_json_str(output, "event");
-    if (event_name == NULL) {
-        return;
-    }
-
-    const char *dialog_id = voice_get_json_str(output, "dialog_id");
-    const char *state = NULL;
-    if (strcmp(event_name, "DialogStateChanged") == 0) {
-        state = voice_get_json_str(output, "state");
-    }
-    ESP_LOGI(TAG,
-             "[stage] ws_event_output: event=%s dialog_id=%s state=%s",
-             event_name,
-             dialog_id != NULL ? dialog_id : "-",
-             state != NULL ? state : "-");
-
-    bool finished = false;
-    bool has_finished = voice_get_json_bool(output, "finished", &finished);
-    const char *responding_final_text = NULL;
-    bool should_check_marker_in_first_sentence = false;
-    if (has_finished && finished) {
-        if (strcmp(event_name, "SpeechContent") == 0) {
-            voice_log_final_text("asr_final_text", voice_get_json_str(output, "text"));
-        } else if (strcmp(event_name, "RespondingContent") == 0) {
-            const char *final_text = voice_get_json_str(output, "text");
-            if (final_text == NULL || final_text[0] == '\0') {
-                final_text = voice_get_json_str(output, "spoken");
-            }
-            voice_log_final_text("response_final_text", final_text);
-            responding_final_text = final_text;
-            if (responding_final_text != NULL && responding_final_text[0] != '\0' &&
-                voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-                if (!s_voice.response_first_sentence_checked) {
-                    s_voice.response_first_sentence_checked = true;
-                    should_check_marker_in_first_sentence = true;
-                }
-                voice_unlock();
-            }
-        }
-    }
-
-    bool should_close_audio = false;
-    if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-        if (dialog_id != NULL) {
-            strlcpy(s_voice.dialog_id, dialog_id, sizeof(s_voice.dialog_id));
-        }
-        s_voice.last_event_ms = voice_now_ms();
-
-        if (strcmp(event_name, "Started") == 0) {
-            s_voice.started = true;
-            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
-            s_voice.response_first_sentence_checked = false;
-            ESP_LOGI(TAG, "[stage] session_started: dialog_id=%s", s_voice.dialog_id);
-        } else if (strcmp(event_name, "DialogStateChanged") == 0) {
-            if (state != NULL) {
-                voice_dialog_state_t prev_state = s_voice.dialog_state;
-                if (strcmp(state, "Listening") == 0) {
-                    s_voice.dialog_state = VOICE_DIALOG_STATE_LISTENING;
-                    if (prev_state != VOICE_DIALOG_STATE_LISTENING) {
-                        s_voice.response_first_sentence_checked = false;
-                    }
-                } else if (strcmp(state, "Thinking") == 0) {
-                    s_voice.dialog_state = VOICE_DIALOG_STATE_THINKING;
-                    if (prev_state != VOICE_DIALOG_STATE_THINKING) {
-                        s_voice.response_first_sentence_checked = false;
-                    }
-                } else if (strcmp(state, "Responding") == 0) {
-                    s_voice.dialog_state = VOICE_DIALOG_STATE_RESPONDING;
-                    if (prev_state != VOICE_DIALOG_STATE_RESPONDING) {
-                        s_voice.response_first_sentence_checked = false;
-                    }
-                }
-            }
-            ESP_LOGI(TAG,
-                     "[stage] dialog_state_changed: state=%s current=%s",
-                     state != NULL ? state : "-",
-                     voice_interaction_dialog_state_str(s_voice.dialog_state));
-        } else if (strcmp(event_name, "SpeechEnded") == 0) {
-            ESP_LOGI(TAG, "[stage] speech_ended: keep tap_active=%d for continuous rounds", s_voice.tap_active);
-        } else if (strcmp(event_name, "Stopped") == 0) {
-            s_voice.started = false;
-            s_voice.tap_active = false;
-            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
-            s_voice.response_first_sentence_checked = false;
-            should_close_audio = true;
-            ESP_LOGI(TAG, "[stage] session_stopped_by_server");
-        } else if (strcmp(event_name, "Error") == 0) {
-            const char *error_msg = voice_get_json_str(output, "error_message");
-            if (error_msg != NULL) {
-                voice_set_last_error_locked(error_msg);
-                ESP_LOGW(TAG, "[stage] server_error: %s", error_msg);
-            }
-            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
-            s_voice.tap_active = false;
-            s_voice.response_first_sentence_checked = false;
-            should_close_audio = true;
-        }
-        voice_unlock();
-    }
-
-    if (should_close_audio) {
-        voice_close_audio_with_drain(VOICE_AUDIO_DRAIN_WAIT_MS, "server_event_close");
-    }
-
-    if (should_check_marker_in_first_sentence) {
-        voice_try_trigger_marker_image_generation(dialog_id, responding_final_text);
-    }
-
-    if (strcmp(event_name, "RespondingStarted") == 0) {
-        (void)voice_send_directive("continue-task", "LocalRespondingStarted", true);
-    } else if (strcmp(event_name, "RespondingEnded") == 0) {
-        (void)voice_send_directive("continue-task", "LocalRespondingEnded", true);
-    }
-}
-
-static void voice_handle_text_message(const char *text, size_t len) {
-    voice_cjson_init_hooks_once();
-
-    cJSON *root = cJSON_ParseWithLength(text, len);
-    if (root == NULL) {
-        ESP_LOGW(TAG, "Invalid ws text payload");
-        return;
-    }
-
-    cJSON *header = cJSON_GetObjectItemCaseSensitive(root, "header");
-    const char *header_event = voice_get_json_str(header, "event");
-
-    cJSON *payload = cJSON_GetObjectItemCaseSensitive(root, "payload");
-    cJSON *output = NULL;
-    if (payload != NULL) {
-        output = cJSON_GetObjectItemCaseSensitive(payload, "output");
-    }
-
-    if (header != NULL) {
-        cJSON *status_code = cJSON_GetObjectItemCaseSensitive(header, "status_code");
-        cJSON *status_msg = cJSON_GetObjectItemCaseSensitive(header, "status_message");
-        if (cJSON_IsNumber(status_code) && status_code->valueint >= 400) {
-            ESP_LOGW(TAG,
-                     "[stage] ws_header_error: status_code=%d status_message=%s",
-                     status_code->valueint,
-                     (cJSON_IsString(status_msg) && status_msg->valuestring != NULL) ? status_msg->valuestring : "-");
-            if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-                if (cJSON_IsString(status_msg) && status_msg->valuestring != NULL) {
-                    voice_set_last_error_locked(status_msg->valuestring);
-                } else {
-                    voice_set_last_error_locked("websocket status error");
-                }
-                voice_unlock();
-            }
-        }
-    }
-
-    if (output != NULL) {
-        voice_handle_output_event(output);
-    } else if (header_event != NULL && strcmp(header_event, "task-failed") == 0) {
-        const char *error_msg = voice_get_json_str(header, "error_message");
-        ESP_LOGW(TAG, "[stage] ws_task_failed: %s", error_msg != NULL ? error_msg : "task failed");
-        if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
-            voice_set_last_error_locked(error_msg != NULL ? error_msg : "task failed");
-            voice_unlock();
-        }
-    }
-
-    cJSON_Delete(root);
-}
-
 static void voice_handle_text_chunk(esp_websocket_event_data_t *data) {
     if (data->payload_offset == 0) {
         free(s_voice.text_agg);
@@ -963,7 +205,7 @@ static void voice_handle_text_chunk(esp_websocket_event_data_t *data) {
         return;
     }
 
-    voice_handle_text_message((const char *)s_voice.text_agg, s_voice.text_agg_size);
+    voice_ws_protocol_handle_text_message((const char *)s_voice.text_agg, s_voice.text_agg_size);
     free(s_voice.text_agg);
     s_voice.text_agg = NULL;
     s_voice.text_agg_size = 0;
@@ -1006,7 +248,7 @@ static void voice_handle_binary_chunk(esp_websocket_event_data_t *data) {
         return;
     }
 
-    voice_handle_downstream_packet(s_voice.bin_agg, s_voice.bin_agg_size);
+    voice_ws_audio_handle_downstream_packet(s_voice.bin_agg, s_voice.bin_agg_size);
     free(s_voice.bin_agg);
     s_voice.bin_agg = NULL;
     s_voice.bin_agg_size = 0;

main/domain/src/voice_interaction_ws_audio.c

@@ -0,0 +1,100 @@
+#include "voice_interaction_internal.h"
+
+#include "esp_audio_dec.h"
+#include "esp_log.h"
+#include "voice_audio.h"
+
+static const char *TAG = "voice_interaction";
+
+void voice_ws_audio_handle_downstream_packet(const uint8_t *data, size_t len) {
+    if (data == NULL || len == 0) {
+        return;
+    }
+
+    if (!voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+        return;
+    }
+
+    void *opus_dec = s_voice.opus_dec;
+    uint8_t *pcm_buf = s_voice.pcm_rx_buf;
+    size_t pcm_buf_size = s_voice.pcm_rx_buf_size;
+    voice_unlock();
+
+    if (opus_dec == NULL || pcm_buf == NULL || pcm_buf_size == 0) {
+        return;
+    }
+
+    esp_audio_dec_in_raw_t raw = {
+        .buffer = (uint8_t *)data,
+        .len = (uint32_t)len,
+        .consumed = 0,
+    };
+
+    while (raw.len > 0) {
+        esp_audio_dec_out_frame_t out = {
+            .buffer = pcm_buf,
+            .len = (uint32_t)pcm_buf_size,
+            .needed_size = 0,
+            .decoded_size = 0,
+        };
+        esp_audio_dec_info_t dec_info = {0};
+
+        esp_audio_err_t ret = esp_opus_dec_decode(opus_dec, &raw, &out, &dec_info);
+        if (ret == ESP_AUDIO_ERR_BUFF_NOT_ENOUGH) {
+            if (out.needed_size > pcm_buf_size) {
+                uint8_t *new_buf = (uint8_t *)voice_realloc_prefer_psram(pcm_buf, out.needed_size);
+                if (new_buf == NULL) {
+                    ESP_LOGE(TAG, "No memory to extend pcm rx buffer to %u", out.needed_size);
+                    break;
+                }
+                if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+                    s_voice.pcm_rx_buf = new_buf;
+                    s_voice.pcm_rx_buf_size = out.needed_size;
+                    voice_unlock();
+                }
+                pcm_buf = new_buf;
+                pcm_buf_size = out.needed_size;
+                continue;
+            }
+            break;
+        }
+
+        if (ret != ESP_AUDIO_ERR_OK) {
+            ESP_LOGW(TAG, "opus decode failed: %d", ret);
+            break;
+        }
+
+        if (out.decoded_size > 0) {
+            char audio_err[64] = {0};
+            size_t samples = (size_t)out.decoded_size / VOICE_SAMPLE_BYTES;
+            esp_err_t write_err = voice_audio_write_pcm((const int16_t *)out.buffer,
+                                                        samples,
+                                                        VOICE_AUDIO_IO_TIMEOUT_MS,
+                                                        audio_err,
+                                                        sizeof(audio_err));
+            if (write_err == ESP_OK && voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+                int64_t now_ms = voice_now_ms();
+                int64_t pcm_ms = ((int64_t)samples * 1000 +
+                                  (int64_t)CONFIG_TQ_VOICE_SAMPLE_RATE - 1) /
+                                 (int64_t)CONFIG_TQ_VOICE_SAMPLE_RATE;
+                int64_t base_ms = s_voice.playback_deadline_ms > now_ms ?
+                                  s_voice.playback_deadline_ms : now_ms;
+                s_voice.playback_deadline_ms = base_ms + pcm_ms + VOICE_AUDIO_DRAIN_MARGIN_MS;
+                s_voice.last_downstream_ms = now_ms;
+                voice_unlock();
+            }
+        }
+
+        if (raw.consumed == 0 || raw.consumed > raw.len) {
+            break;
+        }
+        raw.buffer += raw.consumed;
+        raw.len -= raw.consumed;
+    }
+
+    if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+        s_voice.downstream_packets++;
+        s_voice.last_event_ms = voice_now_ms();
+        voice_unlock();
+    }
+}

main/domain/src/voice_interaction_ws_protocol.c

@@ -0,0 +1,224 @@
+#include "voice_interaction_internal.h"
+
+#include <string.h>
+
+#define cJSON_malloc voice_malloc_prefer_psram
+#define cJSON_realloc voice_realloc_prefer_psram
+#include "cJSON.h"
+#undef cJSON_realloc
+#undef cJSON_malloc
+#include "esp_log.h"
+
+static const char *TAG = "voice_interaction";
+
+static const char *voice_get_json_str(cJSON *obj, const char *name) {
+    if (obj == NULL || name == NULL) {
+        return NULL;
+    }
+
+    cJSON *item = cJSON_GetObjectItemCaseSensitive(obj, name);
+    if (cJSON_IsString(item) && item->valuestring != NULL) {
+        return item->valuestring;
+    }
+    return NULL;
+}
+
+static bool voice_get_json_bool(cJSON *obj, const char *name, bool *out_value) {
+    if (obj == NULL || name == NULL || out_value == NULL) {
+        return false;
+    }
+
+    cJSON *item = cJSON_GetObjectItemCaseSensitive(obj, name);
+    if (!cJSON_IsBool(item)) {
+        return false;
+    }
+    *out_value = cJSON_IsTrue(item);
+    return true;
+}
+
+static void voice_log_final_text(const char *stage, const char *text) {
+    if (stage == NULL || text == NULL || text[0] == '\0') {
+        return;
+    }
+
+    size_t len = strlen(text);
+    const int preview = 240;
+    ESP_LOGI(TAG,
+             "[stage] %s: len=%u text=%.*s%s",
+             stage,
+             (unsigned)len,
+             preview,
+             text,
+             (len > (size_t)preview) ? "..." : "");
+}
+
+static void voice_handle_output_event(cJSON *output) {
+    const char *event_name = voice_get_json_str(output, "event");
+    if (event_name == NULL) {
+        return;
+    }
+
+    const char *dialog_id = voice_get_json_str(output, "dialog_id");
+    const char *state = NULL;
+    if (strcmp(event_name, "DialogStateChanged") == 0) {
+        state = voice_get_json_str(output, "state");
+    }
+    ESP_LOGI(TAG,
+             "[stage] ws_event_output: event=%s dialog_id=%s state=%s",
+             event_name,
+             dialog_id != NULL ? dialog_id : "-",
+             state != NULL ? state : "-");
+
+    bool finished = false;
+    bool has_finished = voice_get_json_bool(output, "finished", &finished);
+    const char *responding_final_text = NULL;
+    bool should_check_marker_in_first_sentence = false;
+    if (has_finished && finished) {
+        if (strcmp(event_name, "SpeechContent") == 0) {
+            voice_log_final_text("asr_final_text", voice_get_json_str(output, "text"));
+        } else if (strcmp(event_name, "RespondingContent") == 0) {
+            const char *final_text = voice_get_json_str(output, "text");
+            if (final_text == NULL || final_text[0] == '\0') {
+                final_text = voice_get_json_str(output, "spoken");
+            }
+            voice_log_final_text("response_final_text", final_text);
+            responding_final_text = final_text;
+            if (responding_final_text != NULL && responding_final_text[0] != '\0' &&
+                voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+                if (!s_voice.response_first_sentence_checked) {
+                    s_voice.response_first_sentence_checked = true;
+                    should_check_marker_in_first_sentence = true;
+                }
+                voice_unlock();
+            }
+        }
+    }
+
+    bool should_close_audio = false;
+    if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+        if (dialog_id != NULL) {
+            strlcpy(s_voice.dialog_id, dialog_id, sizeof(s_voice.dialog_id));
+        }
+        s_voice.last_event_ms = voice_now_ms();
+
+        if (strcmp(event_name, "Started") == 0) {
+            s_voice.started = true;
+            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
+            s_voice.response_first_sentence_checked = false;
+            ESP_LOGI(TAG, "[stage] session_started: dialog_id=%s", s_voice.dialog_id);
+        } else if (strcmp(event_name, "DialogStateChanged") == 0) {
+            if (state != NULL) {
+                voice_dialog_state_t prev_state = s_voice.dialog_state;
+                if (strcmp(state, "Listening") == 0) {
+                    s_voice.dialog_state = VOICE_DIALOG_STATE_LISTENING;
+                    if (prev_state != VOICE_DIALOG_STATE_LISTENING) {
+                        s_voice.response_first_sentence_checked = false;
+                    }
+                } else if (strcmp(state, "Thinking") == 0) {
+                    s_voice.dialog_state = VOICE_DIALOG_STATE_THINKING;
+                    if (prev_state != VOICE_DIALOG_STATE_THINKING) {
+                        s_voice.response_first_sentence_checked = false;
+                    }
+                } else if (strcmp(state, "Responding") == 0) {
+                    s_voice.dialog_state = VOICE_DIALOG_STATE_RESPONDING;
+                    if (prev_state != VOICE_DIALOG_STATE_RESPONDING) {
+                        s_voice.response_first_sentence_checked = false;
+                    }
+                }
+            }
+            ESP_LOGI(TAG,
+                     "[stage] dialog_state_changed: state=%s current=%s",
+                     state != NULL ? state : "-",
+                     voice_interaction_dialog_state_str(s_voice.dialog_state));
+        } else if (strcmp(event_name, "SpeechEnded") == 0) {
+            ESP_LOGI(TAG, "[stage] speech_ended: keep tap_active=%d for continuous rounds", s_voice.tap_active);
+        } else if (strcmp(event_name, "Stopped") == 0) {
+            s_voice.started = false;
+            s_voice.tap_active = false;
+            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
+            s_voice.response_first_sentence_checked = false;
+            should_close_audio = true;
+            ESP_LOGI(TAG, "[stage] session_stopped_by_server");
+        } else if (strcmp(event_name, "Error") == 0) {
+            const char *error_msg = voice_get_json_str(output, "error_message");
+            if (error_msg != NULL) {
+                voice_set_last_error_locked(error_msg);
+                ESP_LOGW(TAG, "[stage] server_error: %s", error_msg);
+            }
+            s_voice.dialog_state = VOICE_DIALOG_STATE_IDLE;
+            s_voice.tap_active = false;
+            s_voice.response_first_sentence_checked = false;
+            should_close_audio = true;
+        }
+        voice_unlock();
+    }
+
+    if (should_close_audio) {
+        voice_close_audio_with_drain(VOICE_AUDIO_DRAIN_WAIT_MS, "server_event_close");
+    }
+
+    if (should_check_marker_in_first_sentence) {
+        voice_marker_try_trigger_image_generation(dialog_id, responding_final_text);
+    }
+
+    if (strcmp(event_name, "RespondingStarted") == 0) {
+        (void)voice_send_directive("continue-task", "LocalRespondingStarted", true);
+    } else if (strcmp(event_name, "RespondingEnded") == 0) {
+        (void)voice_send_directive("continue-task", "LocalRespondingEnded", true);
+    }
+}
+
+void voice_ws_protocol_handle_text_message(const char *text, size_t len) {
+    if (text == NULL || len == 0) {
+        return;
+    }
+
+    voice_interaction_ws_init_cjson_hooks();
+
+    cJSON *root = cJSON_ParseWithLength(text, len);
+    if (root == NULL) {
+        ESP_LOGW(TAG, "Invalid ws text payload");
+        return;
+    }
+
+    cJSON *header = cJSON_GetObjectItemCaseSensitive(root, "header");
+    const char *header_event = voice_get_json_str(header, "event");
+
+    cJSON *payload = cJSON_GetObjectItemCaseSensitive(root, "payload");
+    cJSON *output = NULL;
+    if (payload != NULL) {
+        output = cJSON_GetObjectItemCaseSensitive(payload, "output");
+    }
+
+    if (header != NULL) {
+        cJSON *status_code = cJSON_GetObjectItemCaseSensitive(header, "status_code");
+        cJSON *status_msg = cJSON_GetObjectItemCaseSensitive(header, "status_message");
+        if (cJSON_IsNumber(status_code) && status_code->valueint >= 400) {
+            ESP_LOGW(TAG,
+                     "[stage] ws_header_error: status_code=%d status_message=%s",
+                     status_code->valueint,
+                     (cJSON_IsString(status_msg) && status_msg->valuestring != NULL) ? status_msg->valuestring : "-");
+            if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+                if (cJSON_IsString(status_msg) && status_msg->valuestring != NULL) {
+                    voice_set_last_error_locked(status_msg->valuestring);
+                } else {
+                    voice_set_last_error_locked("websocket status error");
+                }
+                voice_unlock();
+            }
+        }
+    }
+
+    if (output != NULL) {
+        voice_handle_output_event(output);
+    } else if (header_event != NULL && strcmp(header_event, "task-failed") == 0) {
+        const char *error_msg = voice_get_json_str(header, "error_message");
+        ESP_LOGW(TAG, "[stage] ws_task_failed: %s", error_msg != NULL ? error_msg : "task failed");
+        if (voice_lock(VOICE_STATUS_LOCK_TIMEOUT_MS)) {
+            voice_set_last_error_locked(error_msg != NULL ? error_msg : "task failed");
+            voice_unlock();
+        }
+    }
+
+    cJSON_Delete(root);
+}