Phase 5: hands-free "Computer" wake word (esp-sr WakeNet + AFE VAD)
Test, Build and Push / test-gateway (push) Successful in 10s
Test, Build and Push / release (push) Skipped
Test, Build and Push / build-gateway (push) Skipped

Say "Computer" -> chime + listening -> speak -> AFE VAD detects you
stopped -> auto-sends. No taps. Touch still works as a manual override.

- esp-sr 2.4.6 added; wn9_computer_tts model packed into a new "model"
  flash partition (MODEL_IN_FLASH). App moved to 8M, model 4M.
- audio.c: replaced the on-demand capture_task with an AFE pipeline —
  feed_task is the SOLE mic reader (-> afe->feed); detect_task fetches,
  watches wakeup_state for the wake word and vad_state for end of
  speech, and forwards AFE-cleaned audio upstream during an utterance.
  One mic reader ever.
- short rising chime acknowledges the wake audibly.

Fixes from adversarial review before trusting it:
1. utterance framing (blocking WS sends) moved OFF the AFE fetch thread
   onto an app_task event queue (EV_TOUCH/EV_WAKE/EV_SPEECH_END) — a
   1.5s send could stall fetch and drop the first ~1.5s of speech.
2. app_task is now the single serializer of start/end -> no TOCTOU
   double-start (was: two utterance_start on a tap during wake).
3. VAD accounting resets on every streaming (re)start (wake OR tap),
   not just wake -> a tapped utterance can no longer end instantly on
   stale silence.
4. chime/reply set s_playing (+DMA tail hold) and detect_task skips the
   mic while s_playing -> our own audio no longer streams into STT or
   false-triggers the wake at a playback boundary (no AEC yet).
5. NULL-checked AFE create + feed buffer; tasks only start if AFE is up.

Verified on hardware: model loads, AFE inits with the Computer word,
boots and connects clean, no crash/wedge.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
2026-07-15 10:52:41 +02:00
co-authored by Claude Fable 5
parent cb5826e02b
commit b462dff082
7 changed files with 228 additions and 62 deletions
+60 -38
View File
@@ -12,6 +12,7 @@
#include "esp_log.h"
#include "esp_system.h"
#include "freertos/FreeRTOS.h"
#include "freertos/queue.h"
#include "freertos/semphr.h"
#include "freertos/task.h"
#include "bsp/esp-bsp.h"
@@ -23,59 +24,80 @@
static const char *TAG = "desklock";
static SemaphoreHandle_t s_touch_sem;
static bool s_talking;
/* All utterance framing (blocking WS sends, chime, face) runs on app_task, never
* on the LVGL touch callback nor the AFE detect/fetch thread. The wake word, VAD,
* and touch just POST an event; app_task is the single serializer of start/end,
* which removes both the fetch-thread stall and the double-start race. */
typedef enum { EV_TOUCH, EV_WAKE, EV_SPEECH_END } app_event_t;
static QueueHandle_t s_events;
/* Called from the LVGL touch callback — MUST be fast and non-blocking. */
void app_on_touch(void)
static void post_event(app_event_t ev)
{
if (s_touch_sem != NULL) {
xSemaphoreGive(s_touch_sem);
if (s_events != NULL) {
xQueueSend(s_events, &ev, 0);
}
}
/* Called from the WS event task when the link drops: forget any half-utterance
* so the next tap starts a fresh one instead of hitting the "stop" branch. */
void app_on_disconnect(void)
{
s_talking = false;
}
/* --- these three are the fast, non-blocking hooks called from other tasks --- */
void app_on_touch(void) { post_event(EV_TOUCH); } /* LVGL touch callback */
void app_on_wake(void) { post_event(EV_WAKE); } /* AFE detect_task: wake word */
void app_on_speech_end(void) { post_event(EV_SPEECH_END); } /* AFE detect_task: VAD silence */
/* Runs in app_task (normal context): blocking sends here don't stall rendering. */
static void handle_touch(void)
/* --- the actual work, all on app_task --- */
static void start_utterance(void)
{
face_activity();
if (!gw_connected()) {
s_talking = false; /* link gone; nothing to talk to */
if (!gw_connected() || audio_is_playing() || audio_capture_active()
|| face_get() == FACE_EFFORT) {
return;
}
if (!s_talking) {
if (audio_is_playing() || face_get() == FACE_EFFORT) {
return; /* barge-in is phase 4 */
}
s_talking = true;
gw_send_text("{\"type\":\"utterance_start\"}");
audio_capture_start();
face_set(FACE_LISTENING);
ESP_LOGI(TAG, "listening…");
} else {
s_talking = false;
audio_capture_stop();
gw_send_text("{\"type\":\"utterance_end\"}");
face_set(FACE_PENSIVE);
ESP_LOGI(TAG, "utterance sent");
face_activity();
audio_play_chime(); /* audible "I heard you" */
gw_send_text("{\"type\":\"utterance_start\"}"); /* must precede any audio frames */
audio_capture_start(); /* now detect_task forwards audio */
face_set(FACE_LISTENING);
ESP_LOGI(TAG, "listening…");
}
static void end_utterance(void)
{
if (!audio_capture_active()) {
return;
}
audio_capture_stop();
gw_send_text("{\"type\":\"utterance_end\"}");
face_set(FACE_PENSIVE);
ESP_LOGI(TAG, "utterance sent");
}
/* WS link dropped: abandon any half-utterance so the next wake starts fresh. */
void app_on_disconnect(void)
{
audio_capture_stop();
}
static void app_task(void *arg)
{
(void)arg;
app_event_t ev;
for (;;) {
if (xSemaphoreTake(s_touch_sem, portMAX_DELAY) == pdTRUE) {
vTaskDelay(pdMS_TO_TICKS(60)); /* debounce */
xSemaphoreTake(s_touch_sem, 0); /* coalesce repeats during debounce */
handle_touch();
if (xQueueReceive(s_events, &ev, portMAX_DELAY) != pdTRUE) {
continue;
}
if (ev == EV_TOUCH) {
vTaskDelay(pdMS_TO_TICKS(60)); /* debounce */
xQueueReset(s_events); /* coalesce a bouncy tap burst */
face_activity();
if (!gw_connected()) {
audio_capture_stop();
} else if (!audio_capture_active()) {
start_utterance(); /* tap = manual wake */
} else {
end_utterance(); /* tap = manual end */
}
} else if (ev == EV_WAKE) {
start_utterance();
} else if (ev == EV_SPEECH_END) {
end_utterance();
}
}
}
@@ -118,7 +140,7 @@ void app_main(void)
{
ESP_LOGI(TAG, "DeskLock starting");
s_touch_sem = xSemaphoreCreateBinary();
s_events = xQueueCreate(8, sizeof(app_event_t));
bsp_display_start();
bsp_display_backlight_on();