diff --git a/src/lib/components/admin/Settings/Audio.svelte b/src/lib/components/admin/Settings/Audio.svelte
index c8934c36d2..9d25e43018 100644
--- a/src/lib/components/admin/Settings/Audio.svelte
+++ b/src/lib/components/admin/Settings/Audio.svelte
@@ -267,7 +267,7 @@
 						</div>
 					</div>
 					<div class="w-full">
-						<div class=" mb-2.5 text-sm font-medium">{$i18n.t('TTS Model')}</div>
+						<div class=" mb-1.5 text-sm font-medium">{$i18n.t('TTS Model')}</div>
 						<div class="flex w-full">
 							<div class="flex-1">
 								<input
diff --git a/src/lib/components/chat/MessageInput/VoiceRecording.svelte b/src/lib/components/chat/MessageInput/VoiceRecording.svelte
index 6e0d834723..b2e8596493 100644
--- a/src/lib/components/chat/MessageInput/VoiceRecording.svelte
+++ b/src/lib/components/chat/MessageInput/VoiceRecording.svelte
@@ -1,7 +1,7 @@
 <script lang="ts">
 	import { toast } from 'svelte-sonner';
 	import { createEventDispatcher, tick, getContext } from 'svelte';
-	import { settings } from '$lib/stores';
+	import { config, settings } from '$lib/stores';
 	import { blobToFile, calculateSHA256, findWordIndices } from '$lib/utils';
 
 	import { transcribeAudio } from '$lib/apis/audio';
@@ -185,8 +185,7 @@
 			}
 		};
 		mediaRecorder.start();
-
-		if (($settings?.audio?.stt?.engine ?? '') === 'web') {
+		if ($config.audio.stt.engine === 'web' || ($settings?.audio?.stt?.engine ?? '') === 'web') {
 			if ('SpeechRecognition' in window || 'webkitSpeechRecognition' in window) {
 				// Create a SpeechRecognition object
 				speechRecognition = new (window.SpeechRecognition || window.webkitSpeechRecognition)();
@@ -195,7 +194,7 @@
 				speechRecognition.continuous = true;
 
 				// Set the timeout for turning off the recognition after inactivity (in milliseconds)
-				const inactivityTimeout = 3000; // 3 seconds
+				const inactivityTimeout = 2000; // 3 seconds
 
 				let timeoutId;
 				// Start recognition