engineering/uncloseai.com
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

tts: switch default voice aria -> clara (aria has loud F5 phantom bleed)

Browse source
This commit is contained in:
russell@unturf.com 2026-05-24 11:14:53 -04:00
parent ad3568ff3c
commit 378662b62f
No known key found for this signature in database
5 changed files with 14 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

2
public/src/chat.js
View file

@ -240,7 +240,7 @@ export async function handleUserInput() {
playPauseButton.textContent = "Processing..."; playPauseButton.textContent = "Processing...";
playPauseButton.disabled = true; // Disable button while processing playPauseButton.disabled = true; // Disable button while processing
const mainVoiceSelect = document.getElementById("read-page-voice"); 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); const result = await speakText(accumulatedContent, selectedVoice, 0.9);
aiAudio = result.audio; aiAudio = result.audio;
aiBlob = result.blob; aiBlob = result.blob;

4
public/src/page-reader.js
View file

@ -30,8 +30,8 @@ export async function readPageWithHermes(button = null) {
const processedContent = await processContentWithHermes(content); const processedContent = await processContentWithHermes(content);
// Get voice preference from settings (format: "model:voice", e.g., "tts-1-f5:aria") // Get voice preference from settings (format: "model:voice", e.g., "tts-1-f5:clara")
let voice = "aria"; let voice = "clara";
let model = "tts-1-f5"; let model = "tts-1-f5";
try { try {
const savedVoice = localStorage.getItem("uncloseai_selected_voice"); const savedVoice = localStorage.getItem("uncloseai_selected_voice");

4
public/src/tts-modal.js
View file

@ -116,10 +116,10 @@ function renderVoiceOptions(voiceSelection, voices) {
voiceSelection.innerHTML = ''; // Clear existing options voiceSelection.innerHTML = ''; // Clear existing options
// Get saved voice preference (vault first, fallback to default) // Get saved voice preference (vault first, fallback to default)
let savedVoice = 'tts-1-f5:aria'; // Default let savedVoice = 'tts-1-f5:clara'; // Default
try { try {
if (typeof window !== 'undefined' && window.UncloseVault && window.UncloseVault.isUnlocked()) { 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) { } catch (error) {
console.warn('Failed to read voice preference:', error); console.warn('Failed to read voice preference:', error);

14
public/src/tts.js
View file

@ -350,7 +350,7 @@ function cleanTextForTTS(text) {
// Function to read text using TTS - for Read Page button (uses Hermes). // 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. // 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 { try {
const spokenText = await extractSpokenTokens(text); const spokenText = await extractSpokenTokens(text);
const chunks = splitTextIntoChunks(spokenText); 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 // 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) { export function normalizeModelVoice(model, voice) {
const isQwen = !model || model === "tts-1" || model.includes("qwen"); const isQwen = !model || model === "tts-1" || model.includes("qwen");
const normModel = isQwen ? "tts-1-f5" : model; 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 }; return { model: normModel, voice: normVoice };
} }
@ -448,7 +450,7 @@ export function primeAudioPlayback() {
// Direct TTS function for modal - no Hermes preprocessing. // Direct TTS function for modal - no Hermes preprocessing.
// Multi-chunk uses pipelined playlist: playback starts as soon as chunk 1 lands. // 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 { try {
const cleanedText = cleanTextForTTS(text); const cleanedText = cleanTextForTTS(text);
const chunks = splitTextIntoChunks(cleanedText); 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 // Streaming TTS function - plays audio as chunks arrive using MediaSource API
// Returns immediately with audio element; done promise resolves when stream completes // Returns immediately with audio element; done promise resolves when stream completes
// Automatically chunks long text and streams each chunk sequentially // 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 cleanedText = cleanTextForTTS(text);
const textChunks = splitTextIntoChunks(cleanedText); const textChunks = splitTextIntoChunks(cleanedText);
console.log("TTS streaming input preview:", cleanedText.substring(0, 100) + "..."); 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 // Handle TTS - respects streaming mode setting
// Caches audio blobs in memory so replay doesn't regenerate // Caches audio blobs in memory so replay doesn't regenerate
// Cache is keyed by text+voice+model and clears when the tab closes // 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 { try {
const cacheKey = `${model}:${voice}:${text}`; const cacheKey = `${model}:${voice}:${text}`;

2
public/src/vault.js
View file

@ -409,7 +409,7 @@ const UncloseVault = {
// Model selection // Model selection
selectedModel: null, selectedModel: null,
selectedEndpoint: null, selectedEndpoint: null,
selectedVoice: 'tts-1-f5:aria', selectedVoice: 'tts-1-f5:clara',
// Custom API // Custom API
useCustomAPI: false, useCustomAPI: false,