refactor(voice): defer speech confirmation and add diagnostics
This commit is contained in:
@@ -6,7 +6,7 @@ import { useVoice, useVoiceTelemetry } from "@/contexts/voice-context";
|
||||
|
||||
export function VoiceCompactIndicator() {
|
||||
const { theme } = useUnistyles();
|
||||
const { volume, isDetecting, isSpeaking } = useVoiceTelemetry();
|
||||
const { volume, isSpeaking } = useVoiceTelemetry();
|
||||
const {
|
||||
isVoiceMode,
|
||||
isVoiceSwitching,
|
||||
@@ -24,7 +24,6 @@ export function VoiceCompactIndicator() {
|
||||
<VolumeMeter
|
||||
volume={volume}
|
||||
isMuted={isMuted}
|
||||
isDetecting={isDetecting}
|
||||
isSpeaking={isSpeaking}
|
||||
orientation="horizontal"
|
||||
variant="compact"
|
||||
|
||||
Reference in New Issue
Block a user