feat(audio): add intelligent multi-lingual romanization tool (Japanese & Korean -> Latin) to voice asset uploader

2026-09-22 21:27:45 +08:00
parent 28bf192b6b
commit 2f70ef71b6
7 changed files with 225 additions and 9 deletions
@@ -84,6 +84,7 @@ DOM or live LiteGraph state.
  `api/folder_types.py` own the formerly mixed utility route families.
 - `model_policies.py` owns shared backend rename and protected-category policy.
 - `api/audio_catalog.py` owns audio preset scanning, voice character grouping, emotion slice indexing, audio streaming, audio gallery output history listing/deletion, and multipart voice asset uploading (`api_upload_audio_voice`) with companion transcript `.txt` pairing into F5-TTS directories.
+- `api/romanizer.py` owns Japanese and Korean automatic language detection and phonetic romanization (`api_romanize_text`), converting Kanji/Kana and Hangul to clean Latin text with speech-alignment punctuation normalization.

 ### Frontend

@@ -95,7 +96,7 @@ DOM or live LiteGraph state.
  (`isValidSavedTriggerPosition`); `api/__init__.py` injects an aiohttp no-cache middleware (`Cache-Control: no-cache, no-store, must-revalidate`) for extension static files to eliminate browser memory/disk cache desynchronization across regular page refreshes; and `interface_settings.js` owns language and theme preferences.
 - `ui_domain_switcher.js` owns the dual-domain mode toggle (Visual Studio 🎨 vs Audio & Voice Studio 🎙️) embedded in the brand header.
 - `ui_audio_studio.js` owns the Audio & Voice Studio workspace, presenting character voice cards, emotion slice playback, prompt syntax copying, real-time search, animated equalizers, 1-click Arona workflow loading (`app.loadGraphData`), direct canvas node drop injection, and voice asset creation button triggers.
- `ui_audio_uploader.js` owns the Voice Asset Ingestion modal dialog, featuring audio file drag-and-drop, instant HTML5 preview playback, character/emotion tag selection with quick chips, companion reference transcript typing, and seamless hot reload into the studio view.
+- `ui_audio_uploader.js` owns the Voice Asset Ingestion modal dialog, featuring audio file drag-and-drop, instant HTML5 preview playback, character/emotion tag selection with quick chips, companion reference transcript typing, 1-click Japanese/Korean intelligent romanization (`[✨ 智能转罗马音]`), and seamless hot reload into the studio view.
 - `ui_audio_sidebar.js` owns the Audio Sidebar navigation, rendering trees for All Voices, Character Presets (e.g. Arona) with count badges, and Output History Vault.
 - `ui_audio_gallery.js` owns the Generated Audio Vault, supporting track playback, scrubber progress, duration timecodes, audio download, deletion, and canvas drag-and-drop.
 - `ui_sidebar.js` creates the browser shell, domain-aware header tabs (`[音色预设]`, `[音频库]`, `[工作流]`), domain-aware bottom-left tool suppression, and folder navigation.
@@ -1,7 +1,7 @@
 from . import (
    audio_catalog, folder_types, gallery_routes, materials, media_routes, model_catalog,
    model_media, model_metadata, model_resolution, recipe_packages, recipes,
-    translation_routes,
+    romanizer, translation_routes,
 )
 from .scanner import *
 from .config import *
@@ -105,3 +105,4 @@ def setup_routes(app):
    app.router.add_post('/anomalous/delete_audio_gallery', audio_catalog.api_delete_audio_gallery)
    app.router.add_get('/anomalous/audio_template_workflow', audio_catalog.api_get_audio_template_workflow)
    app.router.add_post('/anomalous/upload_audio_voice', audio_catalog.api_upload_audio_voice)
+    app.router.add_post('/anomalous/romanize_text', romanizer.api_romanize_text)
@@ -0,0 +1,113 @@
+"""Intelligent Romanization Engine for Japanese and Korean text alignment in F5-TTS."""
+
+import re
+import json
+from aiohttp import web
+
+_kakasi_instance = None
+_hangul_transliter = None
+
+
+def _get_kakasi():
+    global _kakasi_instance
+    if _kakasi_instance is None:
+        try:
+            import pykakasi
+            _kakasi_instance = pykakasi.kakasi()
+        except Exception:
+            _kakasi_instance = False
+    return _kakasi_instance if _kakasi_instance else None
+
+
+def _get_hangul_transliter():
+    global _hangul_transliter
+    if _hangul_transliter is None:
+        try:
+            from hangul_romanize import Transliter
+            from hangul_romanize.rule import academic
+            _hangul_transliter = Transliter(academic)
+        except Exception:
+            _hangul_transliter = False
+    return _hangul_transliter if _hangul_transliter else None
+
+
+def _clean_romanized_spacing(text: str) -> str:
+    """Normalize whitespace and punctuation spacing for speech synthesis alignment."""
+    # Remove hyphens often generated in academic Korean romanization between syllables
+    text = re.sub(r'(\w)-(\w)', r'\1\2', text)
+    # Fix spaces preceding punctuation: 'word ,' -> 'word,'
+    text = re.sub(r'\s+([,.:;!?~])', r'\1', text)
+    # Ensure a space follows punctuation if directly adjacent to a letter: 'word,next' -> 'word, next'
+    text = re.sub(r'([,.:;!?])([A-Za-z0-9])', r'\1 \2', text)
+    # Collapse multiple whitespaces
+    text = re.sub(r'[ \t]+', ' ', text)
+    return text.strip()
+
+
+def romanize_text(text: str) -> dict:
+    """
+    Detect language and convert Japanese or Korean into standard Latin/Romaji phonetics.
+    Returns dict with original, romanized, detected language, and success status.
+    """
+    raw = (text or "").strip()
+    if not raw:
+        return {"success": True, "original": "", "romanized": "", "lang": "empty"}
+
+    has_hangul = bool(re.search(r'[\uac00-\ud7a3]', raw))
+    has_kana = bool(re.search(r'[\u3040-\u30ff]', raw))
+
+    # 1. Process Korean text
+    if has_hangul:
+        ht = _get_hangul_transliter()
+        if ht:
+            try:
+                romanized = ht.translit(raw)
+                return {
+                    "success": True,
+                    "original": raw,
+                    "romanized": _clean_romanized_spacing(romanized),
+                    "lang": "ko"
+                }
+            except Exception as e:
+                return {"success": False, "error": f"Korean transliteration failed: {e}", "original": raw}
+
+    # 2. Process Japanese text (Kana or Japanese context)
+    if has_kana or (not has_hangul and bool(re.search(r'[\u4e00-\u9fff]', raw))):
+        kakasi = _get_kakasi()
+        if kakasi:
+            try:
+                result = kakasi.convert(raw)
+                parts = []
+                for item in result:
+                    hep = item.get("hepburn", "")
+                    parts.append(hep if hep else item.get("orig", ""))
+                romanized = " ".join(parts)
+                return {
+                    "success": True,
+                    "original": raw,
+                    "romanized": _clean_romanized_spacing(romanized),
+                    "lang": "ja"
+                }
+            except Exception as e:
+                return {"success": False, "error": f"Japanese romanization failed: {e}", "original": raw}
+
+    # Default: Already Latin or other language, keep as is
+    return {
+        "success": True,
+        "original": raw,
+        "romanized": raw,
+        "lang": "unchanged"
+    }
+
+
+async def api_romanize_text(request):
+    """POST /anomalous/romanize_text - Convert input text to Romaji / Latin transliteration."""
+    try:
+        data = await request.json()
+    except Exception:
+        return web.json_response({"success": False, "error": "Invalid JSON body"}, status=400)
+
+    text = data.get("text", "")
+    res = romanize_text(text)
+    status_code = 200 if res.get("success") else 500
+    return web.json_response(res, status=status_code)
@@ -1291,7 +1291,12 @@ export const i18n = {
        audioSaveAndIngest: "保存并入库",
        audioSaving: "正在入库...",
        audioSaveSuccess: "音色资产已成功入库！",
-        audioUploadRequiredError: "请上传音频文件并填写角色名称与情绪标签"
+        audioUploadRequiredError: "请上传音频文件并填写角色名称与情绪标签",
+        audioRomanizeBtn: "✨ 智能转罗马音",
+        audioRomanizeConverting: "转换中...",
+        audioRomanizeSuccess: "已转为罗马音！",
+        audioRomanizeEmptyHint: "请先在文本框中输入或粘贴日文/韩文台词",
+        audioRomanizeFailed: "转换失败，请重试"
    },
    en: {
        actionTools: "Tools",
@@ -2585,7 +2590,12 @@ export const i18n = {
        audioSaveAndIngest: "Save & Ingest",
        audioSaving: "Saving...",
        audioSaveSuccess: "Voice asset saved successfully!",
-        audioUploadRequiredError: "Please provide an audio file, character name, and emotion tag"
+        audioUploadRequiredError: "Please provide an audio file, character name, and emotion tag",
+        audioRomanizeBtn: "✨ Romanize (Romaji/RR)",
+        audioRomanizeConverting: "Converting...",
+        audioRomanizeSuccess: "Converted to Romaji!",
+        audioRomanizeEmptyHint: "Please enter Japanese or Korean dialogue first",
+        audioRomanizeFailed: "Conversion failed, please retry"
    }
 };

@@ -162,20 +162,76 @@ function createEmotionSection() {
    return { wrap, input };
 }

-function createTranscriptSection() {
-    const wrap = document.createElement('div');
-    wrap.style.cssText = 'display:flex;flex-direction:column;gap:6px;';
+function createTranscriptHeader(textarea) {
+    const header = document.createElement('div');
+    header.style.cssText = 'display:flex;align-items:center;justify-content:space-between;';

    const label = document.createElement('label');
    label.style.cssText = 'font-size:12px;font-weight:600;color:#cbd5e1;';
    label.textContent = t('audioRefTextLabel');

+    const romanizeBtn = document.createElement('button');
+    romanizeBtn.type = 'button';
+    romanizeBtn.className = 'anomalous-romanize-btn';
+    romanizeBtn.innerHTML = `<span>✨</span> <span>${t('audioRomanizeBtn')}</span>`;
+    romanizeBtn.title = '将输入的日文（假名/汉字）或韩文（谚文）一键转换为标准罗马音';
+
+    romanizeBtn.onclick = async () => {
+        const originalText = textarea.value.trim();
+        if (!originalText) {
+            alert(t('audioRomanizeEmptyHint'));
+            textarea.focus();
+            return;
+        }
+
+        const prevHtml = romanizeBtn.innerHTML;
+        romanizeBtn.disabled = true;
+        romanizeBtn.innerHTML = `<span>⏳</span> <span>${t('audioRomanizeConverting')}</span>`;
+
+        try {
+            const resp = await fetch('/anomalous/romanize_text', {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({ text: originalText })
+            });
+            const data = await resp.json();
+            if (data.success && data.romanized) {
+                textarea.value = data.romanized;
+                romanizeBtn.innerHTML = `<span>✅</span> <span>${t('audioRomanizeSuccess')}</span>`;
+                setTimeout(() => {
+                    romanizeBtn.innerHTML = prevHtml;
+                    romanizeBtn.disabled = false;
+                }, 1800);
+            } else {
+                alert(t('audioRomanizeFailed') + ': ' + (data.error || ''));
+                romanizeBtn.innerHTML = prevHtml;
+                romanizeBtn.disabled = false;
+            }
+        } catch (err) {
+            console.error('Failed to romanize text:', err);
+            alert(t('audioRomanizeFailed'));
+            romanizeBtn.innerHTML = prevHtml;
+            romanizeBtn.disabled = false;
+        }
+    };
+
+    header.appendChild(label);
+    header.appendChild(romanizeBtn);
+    return header;
+}
+
+function createTranscriptSection() {
+    const wrap = document.createElement('div');
+    wrap.style.cssText = 'display:flex;flex-direction:column;gap:6px;';
+
    const textarea = document.createElement('textarea');
    textarea.className = 'anomalous-uploader-textarea';
    textarea.rows = 3;
    textarea.placeholder = t('audioRefTextPlaceholder');

-    wrap.appendChild(label);
+    const header = createTranscriptHeader(textarea);
+
+    wrap.appendChild(header);
    wrap.appendChild(textarea);
    return { wrap, textarea };
 }
@@ -9,4 +9,4 @@
@import url("./styles/08-theme-gallery-overrides.css?v=20260918-model-sources-1");
@import url("./styles/09-prompt-drawers.css?v=20260918-model-sources-1");
@import url("./styles/10-model-sources.css?v=20260918-model-sources-1");
-@import url("./styles/11-audio-studio.css?v=20260922-audio-uploader-1");
+@import url("./styles/11-audio-studio.css?v=20260922-romanizer-1");
@@ -331,3 +331,38 @@
    color: #ffffff;
    font-weight: 600;
 }
+
+/* Romanize Action Button */
+.anomalous-romanize-btn {
+    display: inline-flex;
+    align-items: center;
+    gap: 4px;
+    padding: 2px 8px;
+    border-radius: 6px;
+    background: linear-gradient(135deg, rgba(236, 72, 153, 0.15), rgba(168, 85, 247, 0.15));
+    border: 1px solid rgba(236, 72, 153, 0.35);
+    color: #f472b6;
+    font-size: 11px;
+    font-weight: 500;
+    cursor: pointer;
+    transition: all 0.2s cubic-bezier(0.16, 1, 0.3, 1);
+    box-shadow: 0 1px 4px rgba(236, 72, 153, 0.1);
+}
+
+.anomalous-romanize-btn:hover {
+    background: linear-gradient(135deg, rgba(236, 72, 153, 0.3), rgba(168, 85, 247, 0.25));
+    border-color: rgba(236, 72, 153, 0.6);
+    color: #fbcfe8;
+    transform: translateY(-1px);
+    box-shadow: 0 2px 8px rgba(236, 72, 153, 0.25);
+}
+
+.anomalous-romanize-btn:active {
+    transform: translateY(0);
+}
+
+.anomalous-romanize-btn:disabled {
+    opacity: 0.5;
+    cursor: not-allowed;
+    transform: none;
+}