diff --git a/public/src/chat.js b/public/src/chat.js
index bfce1a8..0dd0174 100644
--- a/public/src/chat.js
+++ b/public/src/chat.js
@@ -240,7 +240,7 @@ export async function handleUserInput() {
 				playPauseButton.textContent = "Processing...";
 				playPauseButton.disabled = true; // Disable button while processing
 				const mainVoiceSelect = document.getElementById("read-page-voice");
-				const selectedVoice = mainVoiceSelect ? mainVoiceSelect.value : "aria";
+				const selectedVoice = mainVoiceSelect ? mainVoiceSelect.value : "clara";
 				const result = await speakText(accumulatedContent, selectedVoice, 0.9);
 				aiAudio = result.audio;
 				aiBlob = result.blob;
diff --git a/public/src/page-reader.js b/public/src/page-reader.js
index 77cd4ad..b1802c4 100644
--- a/public/src/page-reader.js
+++ b/public/src/page-reader.js
@@ -30,8 +30,8 @@ export async function readPageWithHermes(button = null) {
 
 	const processedContent = await processContentWithHermes(content);
 
-	// Get voice preference from settings (format: "model:voice", e.g., "tts-1-f5:aria")
-	let voice = "aria";
+	// Get voice preference from settings (format: "model:voice", e.g., "tts-1-f5:clara")
+	let voice = "clara";
 	let model = "tts-1-f5";
 	try {
 		const savedVoice = localStorage.getItem("uncloseai_selected_voice");
diff --git a/public/src/tts-modal.js b/public/src/tts-modal.js
index e73da00..621451a 100644
--- a/public/src/tts-modal.js
+++ b/public/src/tts-modal.js
@@ -116,10 +116,10 @@ function renderVoiceOptions(voiceSelection, voices) {
 	voiceSelection.innerHTML = ''; // Clear existing options
 
 	// Get saved voice preference (vault first, fallback to default)
-	let savedVoice = 'tts-1-f5:aria'; // Default
+	let savedVoice = 'tts-1-f5:clara'; // Default
 	try {
 		if (typeof window !== 'undefined' && window.UncloseVault && window.UncloseVault.isUnlocked()) {
-			savedVoice = window.UncloseVault.get('uncloseai_selected_voice', 'tts-1-f5:aria');
+			savedVoice = window.UncloseVault.get('uncloseai_selected_voice', 'tts-1-f5:clara');
 		}
 	} catch (error) {
 		console.warn('Failed to read voice preference:', error);
diff --git a/public/src/tts.js b/public/src/tts.js
index f6217ca..076d292 100644
--- a/public/src/tts.js
+++ b/public/src/tts.js
@@ -350,7 +350,7 @@ function cleanTextForTTS(text) {
 
 // Function to read text using TTS - for Read Page button (uses Hermes).
 // Multi-chunk uses pipelined playlist: playback starts as soon as chunk 1 lands.
-export async function speakText(text, voice = "aria", rate = 0.9, model = "tts-1-f5") {
+export async function speakText(text, voice = "clara", rate = 0.9, model = "tts-1-f5") {
 	try {
 		const spokenText = await extractSpokenTokens(text);
 		const chunks = splitTextIntoChunks(spokenText);
@@ -415,11 +415,13 @@ const LEGACY_OPENAI_VOICES = new Set([
 ]);
 
 // Normalize a model+voice pair: remap legacy / unknown models to F5, and
-// remap legacy OpenAI voice names to F5's "aria" default.
+// remap legacy OpenAI voice names + aria (which has loud F5 phantom bleed) to
+// "clara" which produces clean audio per multi-voice transcription test.
+const PHANTOM_VOICES = new Set(["aria", "luna", "sage", "cole", "jude"]);
 export function normalizeModelVoice(model, voice) {
 	const isQwen = !model || model === "tts-1" || model.includes("qwen");
 	const normModel = isQwen ? "tts-1-f5" : model;
-	const normVoice = LEGACY_OPENAI_VOICES.has(voice) ? "aria" : voice;
+	const normVoice = LEGACY_OPENAI_VOICES.has(voice) || PHANTOM_VOICES.has(voice) ? "clara" : voice;
 	return { model: normModel, voice: normVoice };
 }
 
@@ -448,7 +450,7 @@ export function primeAudioPlayback() {
 
 // Direct TTS function for modal - no Hermes preprocessing.
 // Multi-chunk uses pipelined playlist: playback starts as soon as chunk 1 lands.
-export async function speakTextDirect(text, voice = "aria", rate = 0.9, model = "tts-1-f5") {
+export async function speakTextDirect(text, voice = "clara", rate = 0.9, model = "tts-1-f5") {
 	try {
 		const cleanedText = cleanTextForTTS(text);
 		const chunks = splitTextIntoChunks(cleanedText);
@@ -476,7 +478,7 @@ export async function speakTextDirect(text, voice = "aria", rate = 0.9, model =
 // Streaming TTS function - plays audio as chunks arrive using MediaSource API
 // Returns immediately with audio element; done promise resolves when stream completes
 // Automatically chunks long text and streams each chunk sequentially
-export async function speakTextStreaming(text, voice = "aria", rate = 0.9, model = "tts-1-f5") {
+export async function speakTextStreaming(text, voice = "clara", rate = 0.9, model = "tts-1-f5") {
 	const cleanedText = cleanTextForTTS(text);
 	const textChunks = splitTextIntoChunks(cleanedText);
 	console.log("TTS streaming input preview:", cleanedText.substring(0, 100) + "...");
@@ -793,7 +795,7 @@ export async function generateTitleForTTS(text) {
 // Handle TTS - respects streaming mode setting
 // Caches audio blobs in memory so replay doesn't regenerate
 // Cache is keyed by text+voice+model and clears when the tab closes
-export async function handleTTS(text, voice = "aria", rate = 0.9, model = "tts-1-f5") {
+export async function handleTTS(text, voice = "clara", rate = 0.9, model = "tts-1-f5") {
 	try {
 		const cacheKey = `${model}:${voice}:${text}`;
 
diff --git a/public/src/vault.js b/public/src/vault.js
index 115cb56..7181de3 100644
--- a/public/src/vault.js
+++ b/public/src/vault.js
@@ -409,7 +409,7 @@ const UncloseVault = {
       // Model selection
       selectedModel: null,
       selectedEndpoint: null,
-      selectedVoice: 'tts-1-f5:aria',
+      selectedVoice: 'tts-1-f5:clara',
 
       // Custom API
       useCustomAPI: false,