feat: per model tts voice

src/lib/components/chat/MessageInput/CallOverlay.svelte

@@ -363,6 +363,19 @@
 	let currentMessageId = null;
 	let currentUtterance = null;
 
+	// Get voice: model-specific > user settings > config default
+	const getVoiceId = () => {
+		// Check for model-specific TTS voice first
+		if (model?.info?.meta?.tts?.voice) {
+			return model.info.meta.tts.voice;
+		}
+		// Fall back to user settings or config default
+		if ($settings?.audio?.tts?.defaultVoice === $config.audio.tts.voice) {
+			return $settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice;
+		}
+		return $config?.audio?.tts?.voice;
+	};
+
 	const speakSpeechSynthesisHandler = (content) => {
 		if ($showCallOverlay) {
 			return new Promise((resolve) => {
@@ -372,11 +385,10 @@
 					if (voices.length > 0) {
 						clearInterval(getVoicesLoop);
 
+						const voiceId = getVoiceId();
 						const voice =
 							voices
-								?.filter(
-									(v) => v.voiceURI === ($settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice)
-								)
+								?.filter((v) => v.voiceURI === voiceId)
 								?.at(0) ?? undefined;
 
 						currentUtterance = new SpeechSynthesisUtterance(content);
@@ -471,7 +483,7 @@
 					const url = await $TTSWorker
 						.generate({
 							text: content,
-							voice: $settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice
+							voice: getVoiceId()
 						})
 						.catch((error) => {
 							console.error(error);
@@ -484,9 +496,7 @@
 				} else if ($config.audio.tts.engine !== '') {
 					const res = await synthesizeOpenAISpeech(
 						localStorage.token,
-						$settings?.audio?.tts?.defaultVoice === $config.audio.tts.voice
-							? ($settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice)
-							: $config?.audio?.tts?.voice,
+						getVoiceId(),
 						content
 					).catch((error) => {
 						console.error(error);

src/lib/components/chat/Messages/ResponseMessage.svelte

@@ -207,6 +207,19 @@
 		speaking = true;
 		const content = removeAllDetails(message.content);
 
+		// Get voice: model-specific > user settings > config default
+		const getVoiceId = () => {
+			// Check for model-specific TTS voice first
+			if (model?.info?.meta?.tts?.voice) {
+				return model.info.meta.tts.voice;
+			}
+			// Fall back to user settings or config default
+			if ($settings?.audio?.tts?.defaultVoice === $config.audio.tts.voice) {
+				return $settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice;
+			}
+			return $config?.audio?.tts?.voice;
+		};
+
 		if ($config.audio.tts.engine === '') {
 			let voices = [];
 			const getVoicesLoop = setInterval(() => {
@@ -214,11 +227,10 @@
 				if (voices.length > 0) {
 					clearInterval(getVoicesLoop);
 
+					const voiceId = getVoiceId();
 					const voice =
 						voices
-							?.filter(
-								(v) => v.voiceURI === ($settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice)
-							)
+							?.filter((v) => v.voiceURI === voiceId)
 							?.at(0) ?? undefined;
 
 					console.log(voice);
@@ -265,7 +277,9 @@
 				return;
 			}
 
-			console.debug('Prepared message content for TTS', messageContentParts);
+			const voiceId = getVoiceId();
+			console.debug('Prepared message content for TTS', messageContentParts, 'voice:', voiceId);
+
 			if ($settings.audio?.tts?.engine === 'browser-kokoro') {
 				if (!$TTSWorker) {
 					await TTSWorker.set(
@@ -281,7 +295,7 @@
 					const url = await $TTSWorker
 						.generate({
 							text: sentence,
-							voice: $settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice
+							voice: voiceId
 						})
 						.catch((error) => {
 							console.error(error);
@@ -300,9 +314,7 @@
 				for (const [idx, sentence] of messageContentParts.entries()) {
 					const res = await synthesizeOpenAISpeech(
 						localStorage.token,
-						$settings?.audio?.tts?.defaultVoice === $config.audio.tts.voice
-							? ($settings?.audio?.tts?.voice ?? $config?.audio?.tts?.voice)
-							: $config?.audio?.tts?.voice,
+						voiceId,
 						sentence
 					).catch((error) => {
 						console.error(error);
@@ -324,6 +336,7 @@
 		}
 	};
 
+
 	let preprocessedDetailsCache = [];
 
 	function preprocessForEditing(content: string): string {

src/lib/components/workspace/Models/ModelEditor.svelte

@@ -107,6 +107,7 @@
 
 	let actionIds = [];
 	let accessControl = {};
+	let tts = { voice: '' };
 
 	const submitHandler = async () => {
 		loading = true;
@@ -194,6 +195,18 @@
 			}
 		}
 
+		if (tts.voice !== '') {
+			if (!info.meta.tts) info.meta.tts = {};
+			info.meta.tts.voice = tts.voice;
+		} else {
+			if (info.meta.tts?.voice) {
+				delete info.meta.tts.voice;
+				if (Object.keys(info.meta.tts).length === 0) {
+					delete info.meta.tts;
+				}
+			}
+		}
+
 		info.params.system = system.trim() === '' ? null : system;
 		info.params.stop = params.stop ? params.stop.split(',').filter((s) => s.trim()) : null;
 		Object.keys(info.params).forEach((key) => {
@@ -275,6 +288,7 @@
 
 			capabilities = { ...capabilities, ...(model?.meta?.capabilities ?? {}) };
 			defaultFeatureIds = model?.meta?.defaultFeatureIds ?? [];
+			tts = { voice: model?.meta?.tts?.voice ?? '' };
 
 			if ('access_control' in model) {
 				accessControl = model.access_control;
@@ -766,6 +780,20 @@
 						{/if}
 					{/if}
 
+					<div class="my-2">
+						<div class="flex w-full justify-between mb-1">
+							<div class="self-center text-xs font-medium text-gray-500">
+								{$i18n.t('TTS Voice')}
+							</div>
+						</div>
+						<input
+							class="w-full text-sm bg-transparent outline-hidden"
+							type="text"
+							bind:value={tts.voice}
+							placeholder={$i18n.t('e.g. alloy, echo, shimmer')}
+						/>
+					</div>
+
 					<hr class=" border-gray-100/30 dark:border-gray-850/30 my-2" />
 
 					<div class="my-2 flex justify-end">