Antigravity/notes-qpid-pro
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(tts): multi-voice switcher for EN & RU, Irina default voice, punctuation pause normalization and speed calibration

Browse Source
This commit is contained in:
Antigravity committed 2026-10-07 12:07:01 +03:00
1 parent 7893b353aa
commit 31e105b2c2
11 files changed
+486 -61
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

app/listening.py
+9 -8
View File
@@ -167,6 +167,7 @@ async def get_listening_exercise(
topic: Optional[str] = None,
level: Optional[str] = None,
user_cards: Optional[List[Dict[str, Any]]] = None,
voice: Optional[str] = None,
) -> Dict[str, Any]:
"""
Selects or generates a listening exercise, ensuring audio is pre-synthesized.
@@ -197,13 +198,13 @@ async def get_listening_exercise(
item_id = f"card-{card.get('id', 0)}"
# Pre-synthesize audio
story_hash = get_audio_hash(ctx, "en", 1.0)
story_hash = get_audio_hash(ctx, "en", 1.0, voice=voice)
question_text = f"Which key word or concept is highlighted in this context?"
question_hash = get_audio_hash(question_text, "en", 1.0)
question_hash = get_audio_hash(question_text, "en", 1.0, voice=voice)
# Fire synthesis
asyncio.create_task(synthesize_speech(ctx, lang="en", speed=1.0))
asyncio.create_task(synthesize_speech(question_text, lang="en", speed=1.0))
asyncio.create_task(synthesize_speech(ctx, lang="en", speed=1.0, voice=voice))
asyncio.create_task(synthesize_speech(question_text, lang="en", speed=1.0, voice=voice))
return {
"id": item_id,
@@ -225,11 +226,11 @@ async def get_listening_exercise(
scenario = random.choice(candidates)
# Ensure audio is pre-synthesized
story_hash = get_audio_hash(scenario["story"], "en", 1.0)
question_hash = get_audio_hash(scenario["question"], "en", 1.0)
story_hash = get_audio_hash(scenario["story"], "en", 1.0, voice=voice)
question_hash = get_audio_hash(scenario["question"], "en", 1.0, voice=voice)
asyncio.create_task(synthesize_speech(scenario["story"], lang="en", speed=1.0))
asyncio.create_task(synthesize_speech(scenario["question"], lang="en", speed=1.0))
asyncio.create_task(synthesize_speech(scenario["story"], lang="en", speed=1.0, voice=voice))
asyncio.create_task(synthesize_speech(scenario["question"], lang="en", speed=1.0, voice=voice))
return {
"id": scenario["id"],
app/main.py
+9 -11
View File
@@ -66,6 +66,7 @@ from app.tts import (
prewarm_card_audio,
AUDIO_CACHE_DIR,
is_piper_available,
get_available_voices,
)
from app.listening import (
get_listening_exercise,
@@ -523,16 +524,17 @@ async def api_tts(
text: str = Query(..., min_length=1, max_length=1000),
lang: str = Query("en"),
speed: float = Query(1.0, ge=0.5, le=2.0),
voice: Optional[str] = Query(None),
user: dict = Depends(require_user),
):
"""
Synthesizes speech on-demand using local Piper TTS (en_US-lessac / ru_RU-dmitri).
Instantly returns cached audio if previously synthesized.
Synthesizes speech on-demand using local Piper TTS.
Instantly returns cached audio if previously synthesized for this voice, speed, and text.
"""
clean_text = text.strip()
if not clean_text:
raise HTTPException(status_code=400, detail="Text cannot be empty")
audio_path = await synthesize_speech(clean_text, lang=lang, speed=speed)
audio_path = await synthesize_speech(clean_text, lang=lang, speed=speed, voice=voice)
if not audio_path or not audio_path.is_file():
raise HTTPException(status_code=500, detail="TTS synthesis failed or unavailable")
return FileResponse(
@@ -544,13 +546,8 @@ async def api_tts(
@app.get("/api/tts/status")
async def api_tts_status(user: dict = Depends(require_user)):
"""Returns status of local Piper TTS and available voice models."""
return {
"available": is_piper_available(),
"engine": "piper-vits",
"voices": ["en_US-lessac", "ru_RU-dmitri"],
"cache_dir": str(AUDIO_CACHE_DIR),
}
"""Returns status of local Piper TTS and all available voice models."""
return get_available_voices()
# ---------------------------------------------------------------------------
@@ -561,6 +558,7 @@ async def api_tts_status(user: dict = Depends(require_user)):
async def api_listening_exercise(
topic: Optional[str] = Query("all"),
level: Optional[str] = Query("B1"),
voice: Optional[str] = Query(None),
user: dict = Depends(require_user),
):
"""
@@ -581,7 +579,7 @@ async def api_listening_exercise(
).fetchall()
user_cards = [dict(r) for r in rows]
ex = await get_listening_exercise(topic=topic, level=level, user_cards=user_cards)
ex = await get_listening_exercise(topic=topic, level=level, user_cards=user_cards, voice=voice)
# Strip keywords and ideal_answer from client response to prevent inspection
safe_ex = {k: v for k, v in ex.items() if k not in ("keywords", "ideal_answer")}
return safe_ex
app/static/css/style.css
+115
View File
@@ -2431,6 +2431,121 @@ body.drawer-open {
border-top: 1px solid var(--border-color);
}
/* Voice Settings Modal */
.voice-modal-window {
max-width: 680px;
width: 95%;
}
.voice-modal-body {
padding: 20px 24px;
display: flex;
flex-direction: column;
gap: 16px;
max-height: 80vh;
overflow-y: auto;
}
.modal-desc-sub {
font-size: 13px;
color: var(--text-muted);
margin: 0 0 10px 0;
}
.voice-group-title {
font-size: 14px;
font-weight: 700;
color: var(--text-main);
display: block;
margin-bottom: 10px;
}
.voice-options-grid {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 12px;
}
@media (max-width: 680px) {
.voice-options-grid {
grid-template-columns: 1fr;
}
}
.voice-card-label {
cursor: pointer;
display: block;
position: relative;
user-select: none;
}
.voice-card-label input[type="radio"] {
position: absolute;
opacity: 0;
pointer-events: none;
}
.voice-card {
border: 1px solid var(--border-color);
background: var(--bg-card);
border-radius: var(--radius-md);
padding: 14px 14px;
display: flex;
flex-direction: column;
gap: 8px;
height: 100%;
transition: var(--transition);
}
.voice-card:hover {
border-color: var(--primary-dim);
}
.voice-card-label input[type="radio"]:checked + .voice-card {
border-color: var(--primary);
background: var(--bg-main);
box-shadow: 0 0 0 2px var(--primary-dim);
}
.voice-card-head {
display: flex;
justify-content: space-between;
align-items: center;
gap: 6px;
}
.voice-card-desc {
font-size: 12px;
line-height: 1.4;
color: var(--text-muted);
flex: 1;
}
.badge-recommended {
background: rgba(16, 185, 129, 0.15);
color: #10b981;
font-size: 10px;
font-weight: 700;
padding: 2px 6px;
border-radius: 4px;
white-space: nowrap;
}
.badge-uk {
background: rgba(59, 130, 246, 0.15);
color: #3b82f6;
font-size: 10px;
font-weight: 700;
padding: 2px 6px;
border-radius: 4px;
white-space: nowrap;
}
.voice-test-btn {
align-self: flex-start;
margin-top: 4px;
}
/* Summary Card */
.srs-summary-card {
background-color: var(--bg-card);
app/static/index.html
+102
View File
@@ -512,6 +512,11 @@
<option value="business">💼 Работа & Бизнес</option>
<option value="cards">📚 Мои слова из словаря</option>
</select>
<select id="listening-voice-select" class="select-styled select-xs" title="Голос английского диктора">
<option value="en_US-lessac">👩 Paula (US)</option>
<option value="en_US-ryan">👨 Ryan (US)</option>
<option value="en_GB-alan">👨 Alan (UK)</option>
</select>
<select id="listening-speed-select" class="select-styled select-xs" title="Скорость речи">
<option value="1.0">⚡ 1.0x</option>
<option value="0.75">🐢 0.75x</option>
@@ -1184,6 +1189,12 @@
<span class="drawer-item-label">Переключить тему оформления</span>
</button>
</li>
<li>
<button class="drawer-nav-item" id="drawer-nav-voice-settings">
<span class="drawer-item-icon">🎙️</span>
<span class="drawer-item-label">Голоса озвучки (TTS)</span>
</button>
</li>
<li>
<button class="drawer-nav-item text-danger" id="drawer-nav-logout">
<span class="drawer-item-icon">🚪</span>
@@ -1392,6 +1403,97 @@
</div>
</div>
<!-- Voice Settings Modal -->
<div id="voice-settings-modal" class="modal-backdrop hidden">
<div class="modal-window voice-modal-window">
<div class="modal-header">
<h3>🎙️ Настройка дикторов (TTS)</h3>
<button id="voice-settings-close-btn" class="icon-btn" title="Закрыть">✕</button>
</div>
<div class="modal-body voice-modal-body">
<p class="modal-desc-sub">Выберите голос для озвучки английских и русских текстов. Нейросеть работает полностью автономно.</p>
<div class="voice-config-group">
<label class="voice-group-title">🇷🇺 Русский диктор (перевод и контекст):</label>
<div class="voice-options-grid">
<label class="voice-card-label">
<input type="radio" name="tts_voice_ru" value="ru_RU-irina" checked>
<div class="voice-card">
<div class="voice-card-head">
<strong>👩 Ирина</strong>
<span class="badge-recommended">Рекомендуется</span>
</div>
<small class="voice-card-desc">Плавный студийный голос, естественные паузы на тире и запятых, спокойный темп.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="ru" data-voice="ru_RU-irina" data-text="Это важный проект — он требует внимания, концентрации и времени.">▶ Тест звука</button>
</div>
</label>
<label class="voice-card-label">
<input type="radio" name="tts_voice_ru" value="ru_RU-denis">
<div class="voice-card">
<div class="voice-card-head">
<strong>👨 Денис</strong>
</div>
<small class="voice-card-desc">Спокойный мужской тембр, ровная дикция.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="ru" data-voice="ru_RU-denis" data-text="Это важный проект — он требует внимания, концентрации и времени.">▶ Тест звука</button>
</div>
</label>
<label class="voice-card-label">
<input type="radio" name="tts_voice_ru" value="ru_RU-dmitri">
<div class="voice-card">
<div class="voice-card-head">
<strong>👨 Дмитрий</strong>
</div>
<small class="voice-card-desc">Энергичный мужской голос с оптимизированным размеренным темпом.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="ru" data-voice="ru_RU-dmitri" data-text="Это важный проект — он требует внимания, концентрации и времени.">▶ Тест звука</button>
</div>
</label>
</div>
</div>
<div class="voice-config-group" style="margin-top: 20px;">
<label class="voice-group-title">🇬🇧 / 🇺🇸 Английский диктор (слова, аудирование, книги):</label>
<div class="voice-options-grid">
<label class="voice-card-label">
<input type="radio" name="tts_voice_en" value="en_US-lessac" checked>
<div class="voice-card">
<div class="voice-card-head">
<strong>👩 Paula (US)</strong>
<span class="badge-recommended">По умолчанию</span>
</div>
<small class="voice-card-desc">Американский английский. Четкий академический женский голос.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="en" data-voice="en_US-lessac" data-text="The quick brown fox jumps over the lazy dog.">▶ Test Voice</button>
</div>
</label>
<label class="voice-card-label">
<input type="radio" name="tts_voice_en" value="en_US-ryan">
<div class="voice-card">
<div class="voice-card-head">
<strong>👨 Ryan (US)</strong>
</div>
<small class="voice-card-desc">Американский английский. Естественный, выразительный мужской голос.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="en" data-voice="en_US-ryan" data-text="The quick brown fox jumps over the lazy dog.">▶ Test Voice</button>
</div>
</label>
<label class="voice-card-label">
<input type="radio" name="tts_voice_en" value="en_GB-alan">
<div class="voice-card">
<div class="voice-card-head">
<strong>👨 Alan (UK)</strong>
<span class="badge-uk">British</span>
</div>
<small class="voice-card-desc">Британский английский (Received Pronunciation), классический мужской тембр.</small>
<button type="button" class="btn btn-xs btn-outline voice-test-btn" data-lang="en" data-voice="en_GB-alan" data-text="Good afternoon! The train to Oxford departs from platform four.">▶ Test Voice</button>
</div>
</label>
</div>
</div>
</div>
<div class="modal-footer">
<button type="button" id="voice-settings-save-btn" class="btn btn-primary">Сохранить выбор</button>
</div>
</div>
</div>
<!-- Wikipedia Settings Modal -->
<div id="wiki-settings-modal" class="modal-backdrop hidden">
<div class="modal-window wiki-modal-window">
app/static/js/app.js
+69 -1
View File
@@ -4,7 +4,7 @@
// =======================================================================
import { api } from './modules/api.js';
import { initTheme, openModal, closeModal, showToast } from './modules/ui.js';
import { initTheme, openModal, closeModal, showToast, playTextSpeech } from './modules/ui.js';
import { initDictionary, loadInitialData } from './modules/dictionary.js';
import { initSrs, startSrsTrainingSession, openStatsDashboard } from './modules/srs.js';
import { initReader, loadBookshelf, openBook } from './modules/reader.js';
@@ -546,6 +546,14 @@ export function initNavigationDrawer() {
themeToggle.click();
});
}
const drawerNavVoiceSettings = document.getElementById('drawer-nav-voice-settings');
const voiceSettingsModal = document.getElementById('voice-settings-modal');
if (drawerNavVoiceSettings && voiceSettingsModal) {
drawerNavVoiceSettings.addEventListener('click', () => {
closeDrawer();
openVoiceSettingsModal();
});
}
if (drawerNavLogout && logoutBtn) {
drawerNavLogout.addEventListener('click', () => {
closeDrawer();
@@ -554,6 +562,63 @@ export function initNavigationDrawer() {
}
}
// -----------------------------------------------------------------------
// Voice Settings Modal Controller (TTS)
// -----------------------------------------------------------------------
function openVoiceSettingsModal() {
const voiceSettingsModal = document.getElementById('voice-settings-modal');
if (!voiceSettingsModal) return;
const currentRu = localStorage.getItem('notes_tts_voice_ru') || 'ru_RU-irina';
const currentEn = localStorage.getItem('notes_tts_voice_en') || 'en_US-lessac';
const ruInput = voiceSettingsModal.querySelector(`input[name="tts_voice_ru"][value="${currentRu}"]`);
if (ruInput) ruInput.checked = true;
const enInput = voiceSettingsModal.querySelector(`input[name="tts_voice_en"][value="${currentEn}"]`);
if (enInput) enInput.checked = true;
openModal(voiceSettingsModal);
}
function initVoiceSettingsModal() {
const voiceSettingsModal = document.getElementById('voice-settings-modal');
const voiceSettingsCloseBtn = document.getElementById('voice-settings-close-btn');
const voiceSettingsSaveBtn = document.getElementById('voice-settings-save-btn');
if (voiceSettingsCloseBtn && voiceSettingsModal) {
voiceSettingsCloseBtn.addEventListener('click', () => closeModal(voiceSettingsModal));
}
if (voiceSettingsSaveBtn && voiceSettingsModal) {
voiceSettingsSaveBtn.addEventListener('click', () => {
const ruVal = voiceSettingsModal.querySelector('input[name="tts_voice_ru"]:checked')?.value || 'ru_RU-irina';
const enVal = voiceSettingsModal.querySelector('input[name="tts_voice_en"]:checked')?.value || 'en_US-lessac';
localStorage.setItem('notes_tts_voice_ru', ruVal);
localStorage.setItem('notes_tts_voice_en', enVal);
// Sync with listening selector if on screen
const listeningVoiceSelect = document.getElementById('listening-voice-select');
if (listeningVoiceSelect) {
listeningVoiceSelect.value = enVal;
}
closeModal(voiceSettingsModal);
showToast('✅ Голоса дикторов сохранены!', 'success');
});
}
// Voice test buttons
if (voiceSettingsModal) {
voiceSettingsModal.querySelectorAll('.voice-test-btn').forEach(btn => {
btn.addEventListener('click', (e) => {
e.preventDefault();
e.stopPropagation();
const lang = btn.dataset.lang || 'en';
const voice = btn.dataset.voice;
const text = btn.dataset.text || 'Test voice';
playTextSpeech(text, lang, 1.0, voice);
});
});
}
}
// -----------------------------------------------------------------------
// Password Change Modal
// -----------------------------------------------------------------------
@@ -695,6 +760,9 @@ document.addEventListener('DOMContentLoaded', () => {
// 7. Change Password Modal
initChangePasswordModal();
// 7b. Voice Settings Modal (TTS)
initVoiceSettingsModal();
// 8. Hash Change Listener
window.addEventListener('hashchange', handleRouting);
app/static/js/modules/api.js
+14 -2
View File
@@ -185,8 +185,12 @@ export const api = {
// Listening Comprehension Trainer
listening: {
async getExercise(topic = 'all', level = 'B1') {
const resp = await fetch(`/api/listening/exercise?topic=${encodeURIComponent(topic)}&level=${encodeURIComponent(level)}`);
async getExercise(topic = 'all', level = 'B1', voice = null) {
let url = `/api/listening/exercise?topic=${encodeURIComponent(topic)}&level=${encodeURIComponent(level)}`;
if (voice) {
url += `&voice=${encodeURIComponent(voice)}`;
}
const resp = await fetch(url);
if (resp.ok) return await resp.json();
throw new Error(`Failed to load listening exercise: ${resp.status}`);
},
@@ -206,6 +210,14 @@ export const api = {
}
},
// Neural Speech Synthesis (TTS)
tts: {
async getStatus() {
const resp = await fetch('/api/tts/status');
return resp.ok ? await resp.json() : null;
}
},
// Bookshelf & Reader
books: {
async getAll() {
app/static/js/modules/srs.js
+11 -1
View File
@@ -111,6 +111,7 @@ export function initSrs() {
const listeningLevelBadge = document.getElementById('listening-level-badge');
const listeningTopicBadge = document.getElementById('listening-topic-badge');
const listeningTopicSelect = document.getElementById('listening-topic-select');
const listeningVoiceSelect = document.getElementById('listening-voice-select');
const listeningSpeedSelect = document.getElementById('listening-speed-select');
const listeningNewScenarioBtn = document.getElementById('listening-new-scenario-btn');
const listeningTitle = document.getElementById('listening-title');
@@ -715,8 +716,9 @@ export function initSrs() {
if (listeningTitle) listeningTitle.textContent = 'Загрузка аудио-ситуации...';
const topic = listeningTopicSelect ? listeningTopicSelect.value : 'all';
const voice = listeningVoiceSelect ? listeningVoiceSelect.value : (localStorage.getItem('notes_tts_voice_en') || 'en_US-lessac');
try {
const data = await api.listening.getExercise(topic, 'B1');
const data = await api.listening.getExercise(topic, 'B1', voice);
currentListeningExercise = data;
if (listeningTitle) listeningTitle.textContent = data.title || 'Ситуация для аудирования';
@@ -788,6 +790,14 @@ export function initSrs() {
});
}
if (listeningVoiceSelect) {
listeningVoiceSelect.value = localStorage.getItem('notes_tts_voice_en') || 'en_US-lessac';
listeningVoiceSelect.addEventListener('change', () => {
localStorage.setItem('notes_tts_voice_en', listeningVoiceSelect.value);
loadListeningExercise(true);
});
}
if (listeningForm) {
listeningForm.addEventListener('submit', async (e) => {
e.preventDefault();
app/static/js/modules/ui.js
+8 -2
View File
@@ -194,8 +194,14 @@ export function playAudio(url, speed = 1.0) {
}
}
export function playTextSpeech(text, lang = 'en', speed = 1.0) {
export function playTextSpeech(text, lang = 'en', speed = 1.0, voice = null) {
if (!text || !text.trim()) return null;
const url = `/api/tts?text=${encodeURIComponent(text.trim())}&lang=${lang}&speed=${speed}`;
const isRu = lang && lang.toLowerCase().startsWith('ru');
const savedVoice = isRu ? localStorage.getItem('notes_tts_voice_ru') : localStorage.getItem('notes_tts_voice_en');
const activeVoice = voice || savedVoice;
let url = `/api/tts?text=${encodeURIComponent(text.trim())}&lang=${lang}&speed=${speed}`;
if (activeVoice) {
url += `&voice=${encodeURIComponent(activeVoice)}`;
}
return playAudio(url, speed);
}
app/static/sw.js
+1 -1
View File
@@ -1,7 +1,7 @@
// Service Worker for Notes English PWA (Stage 2)
// Strategy: Cache-first for static assets, Network-first for API calls.
const CACHE_NAME = 'notes-v6.5';
const CACHE_NAME = 'notes-v6.6';
const OFFLINE_URL = '/';
const PRECACHE_URLS = [
app/tts.py
+137 -34
View File
@@ -4,7 +4,7 @@ import hashlib
import asyncio
import logging
from pathlib import Path
from typing import Optional, Dict, Any
from typing import Optional, Dict, Any, List
from app import config
@@ -18,45 +18,127 @@ AUDIO_CACHE_DIR = config.BASE_DIR / "data" / "audio_cache"
# Ensure audio cache directory exists
AUDIO_CACHE_DIR.mkdir(parents=True, exist_ok=True)
VOICE_MODELS = {
"en": {
# Curated high-fidelity Piper ONNX models for English and Russian
VOICE_MODELS: Dict[str, Dict[str, Any]] = {
# --- English Voices ---
"en_US-lessac": {
"id": "en_US-lessac",
"lang": "en",
"gender": "female",
"title": "Paula (US • Женский, спокойный)",
"model": MODELS_DIR / "en_US-lessac-medium.onnx",
"config": MODELS_DIR / "en_US-lessac-medium.onnx.json",
"name": "en_US-lessac",
"base_length_scale": 1.0,
},
"ru": {
"en_US-ryan": {
"id": "en_US-ryan",
"lang": "en",
"gender": "male",
"title": "Ryan (US • Мужской, естественный)",
"model": MODELS_DIR / "en_US-ryan-medium.onnx",
"config": MODELS_DIR / "en_US-ryan-medium.onnx.json",
"base_length_scale": 1.0,
},
"en_GB-alan": {
"id": "en_GB-alan",
"lang": "en",
"gender": "male",
"title": "Alan (UK • Британский акцент)",
"model": MODELS_DIR / "en_GB-alan-medium.onnx",
"config": MODELS_DIR / "en_GB-alan-medium.onnx.json",
"base_length_scale": 1.0,
},
# --- Russian Voices ---
"ru_RU-irina": {
"id": "ru_RU-irina",
"lang": "ru",
"gender": "female",
"title": "Ирина (Женский, студийный, плавный)",
"model": MODELS_DIR / "ru_RU-irina-medium.onnx",
"config": MODELS_DIR / "ru_RU-irina-medium.onnx.json",
"base_length_scale": 1.0,
},
"ru_RU-denis": {
"id": "ru_RU-denis",
"lang": "ru",
"gender": "male",
"title": "Денис (Мужской, умеренный)",
"model": MODELS_DIR / "ru_RU-denis-medium.onnx",
"config": MODELS_DIR / "ru_RU-denis-medium.onnx.json",
"base_length_scale": 1.08,
},
"ru_RU-dmitri": {
"id": "ru_RU-dmitri",
"lang": "ru",
"gender": "male",
"title": "Дмитрий (Мужской, энергичный)",
"model": MODELS_DIR / "ru_RU-dmitri-medium.onnx",
"config": MODELS_DIR / "ru_RU-dmitri-medium.onnx.json",
"name": "ru_RU-dmitri",
"base_length_scale": 1.18, # Calibrated to eliminate rushing
},
}
DEFAULT_VOICES = {
"en": "en_US-lessac",
"ru": "ru_RU-irina", # Irina provides natural pacing and clear punctuation pauses
}
def clean_tts_text(text: str) -> str:
"""Cleans input text from markdown, HTML tags, and abnormal whitespace."""
def clean_tts_text(text: str, lang: str = "en") -> str:
"""
Cleans input text from markdown, HTML tags, abnormal whitespace and
normalizes punctuation pauses so the engine doesn't swallow commas or dashes.
"""
if not text:
return ""
# Remove HTML tags if any
t = re.sub(r"<[^>]+>", " ", text)
# Remove markdown link formatting [text](url) -> text
t = re.sub(r"\[([^\]]+)\]\([^)]+\)", r"\1", t)
# Normalize quotes
t = t.replace("«", '"').replace("»", '"').replace("“", '"').replace("”", '"')
# Normalize em-dashes and en-dashes for natural acoustic pauses:
# Em-dash (—) and en-dash (–) are not in Piper's phoneme map, so eSpeak discards them without pause.
# Replacing surrounded dashes with a comma-pause ensures an intonational breath pause.
t = re.sub(r"\s*[—–]\s*", ", — ", t)
t = re.sub(r"\s*--\s*", ", — ", t)
# Fix punctuation without trailing space: "слово,слово" -> "слово, слово"
t = re.sub(r"([,;:!?])([^\s0-9])", r"\1 \2", t)
# Collapse multiple whitespaces
t = " ".join(t.strip().split())
return t
def get_audio_hash(text: str, lang: str = "en", speed: float = 1.0) -> str:
"""Computes a deterministic MD5 hash for audio caching."""
clean = clean_tts_text(text).lower()
def resolve_voice(lang: str = "en", voice: Optional[str] = None) -> Dict[str, Any]:
"""Resolves requested voice ID or falls back to default voice for language."""
norm_lang = "ru" if lang.lower().startswith("ru") else "en"
if voice and voice in VOICE_MODELS:
candidate = VOICE_MODELS[voice]
if candidate["lang"] == norm_lang:
return candidate
def_id = DEFAULT_VOICES.get(norm_lang, "en_US-lessac")
return VOICE_MODELS.get(def_id, list(VOICE_MODELS.values())[0])
def get_audio_hash(text: str, lang: str = "en", speed: float = 1.0, voice: Optional[str] = None) -> str:
"""Computes a deterministic MD5 hash for audio caching, accounting for voice, speed and text."""
clean = clean_tts_text(text, lang).lower()
norm_speed = round(float(speed), 2)
norm_lang = "ru" if lang.lower().startswith("ru") else "en"
key = f"{norm_lang}:{norm_speed:.2f}:{clean}".encode("utf-8")
v_info = resolve_voice(norm_lang, voice)
v_id = v_info["id"]
key = f"{norm_lang}:{v_id}:{norm_speed:.2f}:{clean}".encode("utf-8")
return hashlib.md5(key).hexdigest()
def get_cached_audio_path(text: str, lang: str = "en", speed: float = 1.0) -> Optional[Path]:
def get_cached_audio_path(text: str, lang: str = "en", speed: float = 1.0, voice: Optional[str] = None) -> Optional[Path]:
"""Returns Path if already cached and valid, else None."""
h = get_audio_hash(text, lang, speed)
h = get_audio_hash(text, lang, speed, voice)
file_path = AUDIO_CACHE_DIR / f"{h}.wav"
if file_path.is_file() and file_path.stat().st_size > 100:
return file_path
@@ -64,28 +146,52 @@ def get_cached_audio_path(text: str, lang: str = "en", speed: float = 1.0) -> Op
def is_piper_available() -> bool:
"""Checks if piper binary and at least en model exist and are executable."""
"""Checks if piper binary and at least one model exist and are executable."""
if not PIPER_BIN.is_file() or not os.access(PIPER_BIN, os.X_OK):
return False
en_model = VOICE_MODELS["en"]["model"]
return en_model.is_file()
return any(v["model"].is_file() for v in VOICE_MODELS.values())
def get_available_voices() -> Dict[str, Any]:
"""Returns metadata for all available models and active default voices."""
available = []
for k, v in VOICE_MODELS.items():
if v["model"].is_file():
available.append({
"id": v["id"],
"lang": v["lang"],
"gender": v["gender"],
"title": v["title"],
"is_default": (v["id"] == DEFAULT_VOICES.get(v["lang"])),
})
return {
"available": bool(available),
"engine": "piper-vits",
"default_voices": DEFAULT_VOICES,
"voices": available,
}
async def synthesize_speech(
text: str,
lang: str = "en",
speed: float = 1.0,
voice: Optional[str] = None,
) -> Optional[Path]:
"""
Synthesizes speech using local Piper TTS and returns Path to the generated WAV file.
Fast disk-cached: if already generated, returns immediately without running synthesis.
"""
clean = clean_tts_text(text)
clean = clean_tts_text(text, lang)
if not clean:
return None
norm_lang = "ru" if lang.lower().startswith("ru") else "en"
voice_info = resolve_voice(norm_lang, voice)
v_id = voice_info["id"]
# Check cache first
cached_path = get_cached_audio_path(clean, lang, speed)
cached_path = get_cached_audio_path(clean, norm_lang, speed, v_id)
if cached_path:
return cached_path
@@ -93,25 +199,22 @@ async def synthesize_speech(
logger.warning("Piper TTS binary or models not available")
return None
norm_lang = "ru" if lang.lower().startswith("ru") else "en"
voice_info = VOICE_MODELS.get(norm_lang, VOICE_MODELS["en"])
model_path = voice_info["model"]
config_path = voice_info["config"]
base_length_scale = voice_info.get("base_length_scale", 1.0)
if not model_path.is_file() or not config_path.is_file():
logger.warning(f"Voice model files missing for {norm_lang}: {model_path}")
logger.warning(f"Voice model files missing for {v_id}: {model_path}")
return None
h = get_audio_hash(clean, norm_lang, speed)
h = get_audio_hash(clean, norm_lang, speed, v_id)
out_wav = AUDIO_CACHE_DIR / f"{h}.wav"
temp_wav = AUDIO_CACHE_DIR / f"{h}.tmp.wav"
# Piper length_scale: > 1.0 is slower, < 1.0 is faster
# 0.75x speed -> length_scale ~ 1.33
# 1.0x speed -> length_scale = 1.0
# 1.25x speed -> length_scale ~ 0.8
clamped_speed = max(0.5, min(2.0, float(speed)))
length_scale = str(round(1.0 / clamped_speed, 2))
# Apply voice-specific base length scale calibration
eff_length_scale = base_length_scale * (1.0 / clamped_speed)
length_scale_str = str(round(eff_length_scale, 2))
env = os.environ.copy()
ld_path = str(PIPER_DIR)
@@ -125,7 +228,8 @@ async def synthesize_speech(
"--model", str(model_path),
"--config", str(config_path),
"--espeak_data", str(PIPER_DIR / "espeak-ng-data"),
"--length_scale", length_scale,
"--length_scale", length_scale_str,
"--sentence_silence", "0.35",
"--output_file", str(temp_wav),
]
@@ -141,7 +245,7 @@ async def synthesize_speech(
if proc.returncode == 0 and temp_wav.is_file() and temp_wav.stat().st_size > 100:
temp_wav.replace(out_wav)
logger.info(f"Synthesized speech [{norm_lang}, speed={speed}]: {clean[:40]}... -> {out_wav.name}")
logger.info(f"Synthesized speech [{v_id}, speed={speed}]: {clean[:40]}... -> {out_wav.name}")
return out_wav
else:
err_text = stderr.decode("utf-8", errors="replace").strip()
@@ -156,19 +260,18 @@ async def synthesize_speech(
return None
async def prewarm_card_audio(term: str, context: Optional[str] = None):
async def prewarm_card_audio(term: str, context: Optional[str] = None, voice: Optional[str] = None):
"""
Background pre-synthesis of word and context sentence.
Invoked when a user saves a card to ensure zero-latency audio playback in SRS.
"""
try:
if term and term.strip():
await synthesize_speech(term.strip(), lang="en", speed=1.0)
await synthesize_speech(term.strip(), lang="en", speed=1.0, voice=voice)
if context and context.strip():
clean_ctx = context.strip().replace('"', '').strip()
# If context is very long, cap to first sentence or 300 chars
if len(clean_ctx) > 300:
clean_ctx = clean_ctx[:300]
await synthesize_speech(clean_ctx, lang="en", speed=1.0)
await synthesize_speech(clean_ctx, lang="en", speed=1.0, voice=voice)
except Exception as e:
logger.warning(f"Background prewarm error: {e}")
tests/test_api.py
+11 -1
View File
@@ -2242,7 +2242,11 @@ def test_piper_tts_endpoints_and_caching():
data = resp.json()
assert "available" in data
assert data["engine"] == "piper-vits"
assert "en_US-lessac" in data["voices"]
voice_ids = [v["id"] for v in data["voices"]]
assert "en_US-lessac" in voice_ids
assert "ru_RU-irina" in voice_ids
assert "en_US-ryan" in voice_ids
assert "en_GB-alan" in voice_ids
# 2. Test TTS synthesis of English word
tts_resp = client.get("/api/tts?text=Astronomy&lang=en&speed=1.0", headers=auth_headers)
@@ -2250,6 +2254,12 @@ def test_piper_tts_endpoints_and_caching():
assert "audio/wav" in tts_resp.headers.get("content-type", "")
assert len(tts_resp.content) > 1000
# 2b. Test TTS synthesis of Russian phrase with dash and Irina voice
ru_resp = client.get("/api/tts?text=Важный%20проект%20—%20он%20работает&lang=ru&voice=ru_RU-irina", headers=auth_headers)
assert ru_resp.status_code == 200
assert "audio/wav" in ru_resp.headers.get("content-type", "")
assert len(ru_resp.content) > 1000
# 3. Test repeat request returns cached audio
cached_resp = client.get("/api/tts?text=Astronomy&lang=en&speed=1.0", headers=auth_headers)
assert cached_resp.status_code == 200