feat(voice): add local VAD speech endpoint detection

2026-10-23 10:59:26 +08:00
parent ee32b4d63f
commit f17ce857ff
17 changed files with 1183 additions and 141 deletions
@@ -619,6 +619,28 @@ menu "TQ Cat Configuration"
                Maximum microphone capture window started by one PTT short press.
                When it expires, uplink capture is stopped and StopSpeech is sent.

+        config TQ_SPEECH_ENDPOINT_ENABLE
+            bool "End click-to-talk input using local VAD"
+            default y
+            depends on TQ_KEY_INPUT_ENABLE && TQ_PTT_MODE_CLICK_TO_TALK
+            help
+                Run WebRTC VAD on AEC-clean PCM while streaming audio. After
+                confirmed speech and continuous silence, finish the current
+                input through the existing PTT teardown. Keep the capture
+                timeout as fallback. No additional audio worker is created.
+
+        config TQ_SPEECH_ENDPOINT_MIN_SPEECH_MS
+            int "Minimum continuous speech before endpoint detection (ms)"
+            default 200
+            range 100 1000
+            depends on TQ_SPEECH_ENDPOINT_ENABLE
+
+        config TQ_SPEECH_ENDPOINT_SILENCE_MS
+            int "Continuous silence to end confirmed speech (ms)"
+            default 900
+            range 400 2000
+            depends on TQ_SPEECH_ENDPOINT_ENABLE
+
    endmenu

    menu "Display & Screen"