fix: OOV english words now transcribed via phonemize G2P + IPA
- Add phonemize (MIT) rule-based G2P as second fallback tier - IPA phoneme -> katakana CV-syllable transcription table - proposal -> プラポーザル, bilibili -> バイリーバイリー (no more letter-spelling) - Letter spelling kept as last resort only - Bridge logs SYNTH IN/OUT to stderr; tts.py drains stderr to Python logs so conversion results are visible in the server log - Rebuild portable zip (83.8 MB)
This commit is contained in:
Generated
+35
@@ -0,0 +1,35 @@
|
||||
{
|
||||
"name": "bililive-touhou-tts",
|
||||
"lockfileVersion": 3,
|
||||
"requires": true,
|
||||
"packages": {
|
||||
"": {
|
||||
"name": "bililive-touhou-tts",
|
||||
"dependencies": {
|
||||
"phonemize": "^1.2.0"
|
||||
}
|
||||
},
|
||||
"node_modules/number-to-words": {
|
||||
"version": "1.2.4",
|
||||
"resolved": "https://registry.npmmirror.com/number-to-words/-/number-to-words-1.2.4.tgz",
|
||||
"integrity": "sha512-/fYevVkXRcyBiZDg6yzZbm0RuaD6i0qRfn8yr+6D0KgBMOndFPxuW10qCHpzs50nN8qKuv78k8MuotZhcVX6Pw==",
|
||||
"license": "MIT"
|
||||
},
|
||||
"node_modules/phonemize": {
|
||||
"version": "1.2.0",
|
||||
"resolved": "https://registry.npmmirror.com/phonemize/-/phonemize-1.2.0.tgz",
|
||||
"integrity": "sha512-+zEpOPXrEaylYCIXMSDVhiAwFbVzIXRf+7vheuNxozg4hLKbQVDXCOpI0GAJw41xEgE9LcwMjLUN30aWrpwSkw==",
|
||||
"license": "MIT",
|
||||
"dependencies": {
|
||||
"number-to-words": "^1.2.4",
|
||||
"pinyin-pro": "^3.26.0"
|
||||
}
|
||||
},
|
||||
"node_modules/pinyin-pro": {
|
||||
"version": "3.28.2",
|
||||
"resolved": "https://registry.npmmirror.com/pinyin-pro/-/pinyin-pro-3.28.2.tgz",
|
||||
"integrity": "sha512-jV38yxXHLfidirMC4hrXasLDozLCSq/4DfX88GnHcSEJ2+GpSedG6I9VOiEXJu6iQ5dbJC/RjmzyMuS5h/wH5A==",
|
||||
"license": "MIT"
|
||||
}
|
||||
}
|
||||
}
|
||||
+3
-1
@@ -1,5 +1,7 @@
|
||||
{
|
||||
"name": "bililive-touhou-tts",
|
||||
"type": "module",
|
||||
"dependencies": {}
|
||||
"dependencies": {
|
||||
"phonemize": "^1.2.0"
|
||||
}
|
||||
}
|
||||
|
||||
@@ -45,6 +45,7 @@ class TTSBridge:
|
||||
self._process: asyncio.subprocess.Process | None = None
|
||||
self._bridge_script = bridge_script or _resolve_path("tts_bridge.js")
|
||||
self._node_exe = _get_node_exe()
|
||||
self._stderr_task: asyncio.Task | None = None
|
||||
|
||||
async def start(self) -> None:
|
||||
if not pathlib.Path(self._bridge_script).exists():
|
||||
@@ -61,6 +62,7 @@ class TTSBridge:
|
||||
stdout=asyncio.subprocess.PIPE,
|
||||
stderr=asyncio.subprocess.PIPE,
|
||||
)
|
||||
self._stderr_task = asyncio.create_task(self._drain_stderr())
|
||||
|
||||
line = await asyncio.wait_for(self._process.stdout.readline(), timeout=60)
|
||||
line_str = line.decode("utf-8").strip()
|
||||
@@ -73,6 +75,23 @@ class TTSBridge:
|
||||
|
||||
logger.info("Node.js TTS bridge ready.")
|
||||
|
||||
async def _drain_stderr(self) -> None:
|
||||
"""Continuously read bridge stderr and forward to Python logs."""
|
||||
if self._process is None or self._process.stderr is None:
|
||||
return
|
||||
try:
|
||||
while True:
|
||||
raw = await self._process.stderr.readline()
|
||||
if not raw:
|
||||
break
|
||||
line = raw.decode("utf-8", errors="replace").rstrip()
|
||||
if line:
|
||||
logger.info("[bridge] %s", line)
|
||||
except asyncio.CancelledError:
|
||||
pass
|
||||
except Exception:
|
||||
logger.debug("Bridge stderr drain stopped", exc_info=True)
|
||||
|
||||
async def synthesize(self, kana_text: str) -> bytes:
|
||||
if self._process is None or self._process.stdin is None:
|
||||
raise RuntimeError("Bridge not started. Call start() first.")
|
||||
@@ -114,6 +133,13 @@ class TTSBridge:
|
||||
async def stop(self) -> None:
|
||||
if self._process is not None:
|
||||
logger.info("Stopping TTS bridge...")
|
||||
if self._stderr_task is not None:
|
||||
self._stderr_task.cancel()
|
||||
try:
|
||||
await self._stderr_task
|
||||
except (asyncio.CancelledError, Exception):
|
||||
pass
|
||||
self._stderr_task = None
|
||||
try:
|
||||
if self._process.stdin is not None:
|
||||
self._process.stdin.close()
|
||||
|
||||
+153
-15
@@ -1,8 +1,10 @@
|
||||
/** tts_bridge.js — Persistent Node.js bridge for aquestalk.js TTS synthesis.
|
||||
|
||||
Converts English words to katakana via vendored english-to-kana dictionary
|
||||
(english-kana-matcher.js, 49K words, MIT). Out-of-dictionary words fall back
|
||||
to letter-by-letter romanized spelling so no word is ever skipped.
|
||||
English word → katakana pipeline (never skips):
|
||||
1. english-to-kana dictionary (49K words, MIT, vendored english-kana-matcher.js)
|
||||
2. phonemize (MIT) G2P → IPA → rule-based katakana transcription
|
||||
3. letter-by-letter romanized spelling (last resort)
|
||||
|
||||
English punctuation (,.!?) is converted to Japanese pauses (、。).
|
||||
*/
|
||||
|
||||
@@ -10,6 +12,7 @@ import { readFileSync, writeFileSync } from "fs";
|
||||
import { createInterface } from "readline";
|
||||
import { fileURLToPath, pathToFileURL } from "url";
|
||||
import { dirname, join } from "path";
|
||||
import { createRequire } from "module";
|
||||
|
||||
const __filename = fileURLToPath(import.meta.url);
|
||||
const __dirname = dirname(__filename);
|
||||
@@ -18,6 +21,16 @@ const __dirname = dirname(__filename);
|
||||
const matcherMod = pathToFileURL(join(__dirname, "english-kana-matcher.js")).href;
|
||||
const { lookupKana } = await import(matcherMod);
|
||||
|
||||
// ── phonemize (G2P, MIT) via CJS entry (avoids JSON import issue in Node ESM) ──
|
||||
const require = createRequire(import.meta.url);
|
||||
let phonemize = null;
|
||||
try {
|
||||
phonemize = require("phonemize").phonemize;
|
||||
console.error("[bridge] phonemize G2P loaded.");
|
||||
} catch (err) {
|
||||
console.error(`[bridge] phonemize unavailable (${err.message}), using letter fallback only.`);
|
||||
}
|
||||
|
||||
// ── aquestalk.js ──────────────────────────────────────────────────────
|
||||
const aquestalkMod = pathToFileURL(join(__dirname, "aquestalk.js", "dist", "index.js")).href;
|
||||
const { load } = await import(aquestalkMod);
|
||||
@@ -46,7 +59,109 @@ try {
|
||||
process.exit(1);
|
||||
}
|
||||
|
||||
// Letter-by-letter fallback readings (AquesTalk1-safe, no ヴ)
|
||||
// ── IPA → Katakana transcription (rule-based) ─────────────────────────
|
||||
const KANA = {
|
||||
k: { a: "カ", i: "キ", u: "ク", e: "ケ", o: "コ" },
|
||||
g: { a: "ガ", i: "ギ", u: "グ", e: "ゲ", o: "ゴ" },
|
||||
s: { a: "サ", i: "シ", u: "ス", e: "セ", o: "ソ" },
|
||||
z: { a: "ザ", i: "ジ", u: "ズ", e: "ゼ", o: "ゾ" },
|
||||
t: { a: "タ", i: "チ", u: "トゥ", e: "テ", o: "ト" },
|
||||
d: { a: "ダ", i: "ジ", u: "ドゥ", e: "デ", o: "ド" },
|
||||
n: { a: "ナ", i: "ニ", u: "ヌ", e: "ネ", o: "ノ" },
|
||||
h: { a: "ハ", i: "ヒ", u: "フ", e: "ヘ", o: "ホ" },
|
||||
b: { a: "バ", i: "ビ", u: "ブ", e: "ベ", o: "ボ" },
|
||||
p: { a: "パ", i: "ピ", u: "プ", e: "ペ", o: "ポ" },
|
||||
m: { a: "マ", i: "ミ", u: "ム", e: "メ", o: "モ" },
|
||||
r: { a: "ラ", i: "リ", u: "ル", e: "レ", o: "ロ" },
|
||||
f: { a: "ファ", i: "フィ", u: "フ", e: "フェ", o: "フォ" },
|
||||
v: { a: "バ", i: "ビ", u: "ブ", e: "ベ", o: "ボ" },
|
||||
tʃ: { a: "チャ", i: "チ", u: "チュ", e: "チェ", o: "チョ" },
|
||||
dʒ: { a: "ジャ", i: "ジ", u: "ジュ", e: "ジェ", o: "ジョ" },
|
||||
ʃ: { a: "シャ", i: "シ", u: "シュ", e: "シェ", o: "ショ" },
|
||||
ʒ: { a: "ジャ", i: "ジ", u: "ジュ", e: "ジェ", o: "ジョ" },
|
||||
θ: { a: "サ", i: "シ", u: "ス", e: "セ", o: "ソ" },
|
||||
ð: { a: "ザ", i: "ジ", u: "ズ", e: "ゼ", o: "ゾ" },
|
||||
w: { a: "ワ", i: "ウィ", u: "ウ", e: "ウェ", o: "ウォ" },
|
||||
j: { a: "ヤ", i: "イ", u: "ユ", e: "イェ", o: "ヨ" },
|
||||
ŋ: { a: "ンガ", i: "ンギ", u: "ング", e: "ンゲ", o: "ンゴ" },
|
||||
};
|
||||
|
||||
const CONS_ROW = {
|
||||
"p": "p", "b": "b", "t": "t", "d": "d", "k": "k", "ɡ": "g", "g": "g",
|
||||
"f": "f", "v": "v", "s": "s", "z": "z", "ʃ": "ʃ", "ʒ": "ʒ",
|
||||
"h": "h", "tʃ": "tʃ", "dʒ": "dʒ", "θ": "θ", "ð": "ð",
|
||||
"m": "m", "n": "n", "ŋ": "ŋ", "l": "r", "ɫ": "r", "ɹ": "r",
|
||||
"r": "r", "j": "j", "w": "w", "ɾ": "r",
|
||||
};
|
||||
|
||||
const VOWEL_KANA = { a: "ア", i: "イ", u: "ウ", e: "エ", o: "オ" };
|
||||
const DIPH_KANA = { ai: "アイ", au: "アウ", oi: "オイ" };
|
||||
|
||||
function vClass(v) {
|
||||
switch (v) {
|
||||
case "ə": case "ɚ": case "ɝ": case "ɑ": case "ʌ": case "ɒ": case "æ": case "ɜ":
|
||||
return { vowel: "a", long: false, diph: null };
|
||||
case "ɪ": return { vowel: "i", long: false, diph: null };
|
||||
case "i": return { vowel: "i", long: true, diph: null };
|
||||
case "ʊ": return { vowel: "u", long: false, diph: null };
|
||||
case "u": return { vowel: "u", long: true, diph: null };
|
||||
case "ɛ": case "e": return { vowel: "e", long: false, diph: null };
|
||||
case "eɪ": return { vowel: "e", long: true, diph: null };
|
||||
case "ɔ": return { vowel: "o", long: false, diph: null };
|
||||
case "o": case "oʊ": return { vowel: "o", long: true, diph: null };
|
||||
case "aɪ": return { vowel: "a", long: false, diph: "ai" };
|
||||
case "aʊ": return { vowel: "a", long: false, diph: "au" };
|
||||
case "ɔɪ": return { vowel: "o", long: false, diph: "oi" };
|
||||
default: return null;
|
||||
}
|
||||
}
|
||||
|
||||
const SONORANT = new Set(["m", "n", "ŋ", "l", "ɫ", "ɹ", "r", "j", "w", "ɾ"]);
|
||||
|
||||
function ipaToKana(ipaStr) {
|
||||
const tokens = [];
|
||||
let i = 0;
|
||||
while (i < ipaStr.length) {
|
||||
const two = ipaStr.slice(i, i + 2);
|
||||
if (CONS_ROW[two] || vClass(two)) { tokens.push(two); i += 2; continue; }
|
||||
const one = ipaStr[i];
|
||||
if (CONS_ROW[one] || vClass(one)) { tokens.push(one); i += 1; continue; }
|
||||
if (one === " " || one === "ː") { tokens.push(one); i += 1; continue; }
|
||||
i += 1;
|
||||
}
|
||||
|
||||
let out = "";
|
||||
for (let i = 0; i < tokens.length; i++) {
|
||||
const t = tokens[i];
|
||||
if (t === " ") { out += " "; continue; }
|
||||
if (t === "ː") { out += "ー"; continue; }
|
||||
const vc = vClass(t);
|
||||
if (vc) {
|
||||
out += vc.diph ? DIPH_KANA[vc.diph] : (VOWEL_KANA[vc.vowel] + (vc.long ? "ー" : ""));
|
||||
continue;
|
||||
}
|
||||
const row = CONS_ROW[t];
|
||||
if (!row) continue;
|
||||
|
||||
const next = tokens[i + 1];
|
||||
const nextVc = next !== undefined ? vClass(next) : null;
|
||||
if (nextVc) {
|
||||
if (nextVc.diph) {
|
||||
out += KANA[row][nextVc.vowel] + DIPH_KANA[nextVc.diph].slice(1);
|
||||
} else {
|
||||
out += KANA[row][nextVc.vowel] + (nextVc.long ? "ー" : "");
|
||||
}
|
||||
i++;
|
||||
} else if (next !== undefined) {
|
||||
out += KANA[row]["u"];
|
||||
} else {
|
||||
out += (t === "n" || t === "ŋ") ? "ン" : KANA[row]["u"];
|
||||
}
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
// ── Letter-by-letter fallback (last resort, AquesTalk1-safe) ──────────
|
||||
const LETTER_KANA = {
|
||||
a: "エー", b: "ビー", c: "シー", d: "ディー", e: "イー",
|
||||
f: "エフ", g: "ジー", h: "エイチ", i: "アイ", j: "ジェー",
|
||||
@@ -60,28 +175,48 @@ const DIGIT_KANA = {
|
||||
"5": "ファイブ", "6": "シックス", "7": "セブン", "8": "エイト", "9": "ナイン",
|
||||
};
|
||||
|
||||
// English punctuation → Japanese pause equivalents
|
||||
function kanaForChar(ch) {
|
||||
return LETTER_KANA[ch.toLowerCase()] || DIGIT_KANA[ch] || null;
|
||||
}
|
||||
|
||||
function spellWord(word) {
|
||||
let out = "";
|
||||
for (const ch of word) {
|
||||
const k = kanaForChar(ch);
|
||||
out += k ? k : "";
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
// ── English punctuation → Japanese pause equivalents ──────────────────
|
||||
const PUNCT_MAP = {
|
||||
",": "、", ".": "。", "!": "！", "?": "？",
|
||||
"，": "、", "。": "。", "！": "！", "？": "？",
|
||||
};
|
||||
|
||||
function kanaForChar(ch) {
|
||||
return LETTER_KANA[ch.toLowerCase()] || DIGIT_KANA[ch] || null;
|
||||
}
|
||||
|
||||
function wordToKana(word) {
|
||||
const clean = word.replace(/'/g, "").toLowerCase();
|
||||
if (!clean) return "";
|
||||
|
||||
// 1. dictionary
|
||||
const hit = lookupKana(clean);
|
||||
if (hit) return hit;
|
||||
|
||||
// OOV fallback: spell out letter by letter (never skip)
|
||||
let out = "";
|
||||
for (const ch of clean) {
|
||||
const k = kanaForChar(ch);
|
||||
out += k ? k : "";
|
||||
// 2. phonemize G2P → IPA → katakana
|
||||
if (phonemize && /^[a-z]+$/.test(clean)) {
|
||||
try {
|
||||
const ipa = phonemize(clean, { stripStress: true });
|
||||
if (ipa && /[^a-z]/.test(ipa)) {
|
||||
const kana = ipaToKana(ipa);
|
||||
if (kana) return kana;
|
||||
}
|
||||
} catch (err) {
|
||||
// fall through to letter spelling
|
||||
}
|
||||
}
|
||||
return out;
|
||||
|
||||
// 3. letter-by-letter spelling (never skip)
|
||||
return spellWord(clean);
|
||||
}
|
||||
|
||||
function convertEnglishSegment(text) {
|
||||
@@ -130,6 +265,9 @@ function synthesize(kanaText) {
|
||||
result += convertEnglishSegment(englishBuf);
|
||||
}
|
||||
|
||||
console.error(`[bridge] SYNTH IN: ${trimmed}`);
|
||||
console.error(`[bridge] SYNTH OUT: ${result}`);
|
||||
|
||||
const wav = aq.run(result, speed);
|
||||
return Buffer.from(wav);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user