Skip to content

feat: add German (de) language support #316

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mitchellecm7 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: add German (de) language support #316

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
75 changes: 33 additions & 42 deletions kokoro.js/src/kokoro.js
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,10 @@
// @ts-nocheck
import { env as hf, StyleTextToSpeech2Model, AutoTokenizer, Tensor, RawAudio } from "@huggingface/transformers";
import { phonemize } from "./phonemize.js";
// ... rest of the file
import { env as hf, StyleTextToSpeech2Model, AutoTokenizer, Tensor, RawAudio } from "@huggingface/transformers";
import { phonemize } from "./phonemize.js";
import { phonemize_de } from "./phonemize_de.js";
import { TextSplitterStream } from "./splitter.js";
import { getVoiceData, VOICES } from "./voices.js";

Expand Down Expand Up @@ -41,7 +46,6 @@ export class KokoroTTS {
static async from_pretrained(model_id, { dtype = "fp32", device = null, progress_callback = null } = {}) {
const model = StyleTextToSpeech2Model.from_pretrained(model_id, { progress_callback, dtype, device });
const tokenizer = AutoTokenizer.from_pretrained(model_id, { progress_callback });

const info = await Promise.all([model, tokenizer]);
return new KokoroTTS(...info);
}
Expand All @@ -54,31 +58,41 @@ export class KokoroTTS {
console.table(VOICES);
}

/**
* Validate voice and return its language prefix.
* @param {string} voice
* @returns {"a"|"b"|"d"}
*/
_validate_voice(voice) {
if (!VOICES.hasOwnProperty(voice)) {
if (!Object.prototype.hasOwnProperty.call(VOICES, voice)) {
console.error(`Voice "${voice}" not found. Available voices:`);
console.table(VOICES);
throw new Error(`Voice "${voice}" not found. Should be one of: ${Object.keys(VOICES).join(", ")}.`);
}
const language = /** @type {"a"|"b"} */ (voice.at(0)); // "a" or "b"
return language;
return /** @type {"a"|"b"|"d"} */ (voice.at(0));
}

/**
* Phonemise text for the given language prefix.
* @param {string} text
* @param {"a"|"b"|"d"} language
* @returns {Promise<string>}
*/
async _phonemize(text, language) {
if (language === "d") return phonemize_de(text);
return phonemize(text, language);
}

/**
* Generate audio from text.
*
* @param {string} text The input text
* @param {GenerateOptions} options Additional options
* @returns {Promise<RawAudio>} The generated audio
*/
async generate(text, { voice = "af_heart", speed = 1 } = {}) {
const language = this._validate_voice(voice);

const phonemes = await phonemize(text, language);
const { input_ids } = this.tokenizer(phonemes, {
truncation: true,
});

const phonemes = await this._phonemize(text, language);
const { input_ids } = this.tokenizer(phonemes, { truncation: true });
return this.generate_from_ids(input_ids, { voice, speed });
}

Expand All @@ -89,22 +103,15 @@ export class KokoroTTS {
* @returns {Promise<RawAudio>} The generated audio
*/
async generate_from_ids(input_ids, { voice = "af_heart", speed = 1 } = {}) {
// Select voice style based on number of input tokens
const num_tokens = Math.min(Math.max(input_ids.dims.at(-1) - 2, 0), 509);

// Load voice style
const data = await getVoiceData(voice);
const offset = num_tokens * STYLE_DIM;
const voiceData = data.slice(offset, offset + STYLE_DIM);

// Prepare model inputs
const inputs = {
input_ids,
style: new Tensor("float32", voiceData, [1, STYLE_DIM]),
speed: new Tensor("float32", [speed], [1]),
};

// Generate audio
const { waveform } = await this.model(inputs);
return new RawAudio(waveform.data, SAMPLE_RATE);
}
Expand All @@ -125,43 +132,27 @@ export class KokoroTTS {
} else if (typeof text === "string") {
splitter = new TextSplitterStream();
const chunks = split_pattern
? text
.split(split_pattern)
.map((chunk) => chunk.trim())
.filter((chunk) => chunk.length > 0)
? text.split(split_pattern).map((c) => c.trim()).filter((c) => c.length > 0)
: [text];
splitter.push(...chunks);
} else {
throw new Error("Invalid input type. Expected string or TextSplitterStream.");
}

for await (const sentence of splitter) {
const phonemes = await phonemize(sentence, language);
const { input_ids } = this.tokenizer(phonemes, {
truncation: true,
});

// TODO: There may be some cases where - even with splitting - the text is too long.
// In that case, we should split the text into smaller chunks and process them separately.
// For now, we just truncate these exceptionally long chunks
const phonemes = await this._phonemize(sentence, language);
const { input_ids } = this.tokenizer(phonemes, { truncation: true });
const audio = await this.generate_from_ids(input_ids, { voice, speed });
yield { text: sentence, phonemes, audio };
}
}
}

export const env = {
set cacheDir(value) {
hf.cacheDir = value
},
get cacheDir() {
return hf.cacheDir
},
set wasmPaths(value) {
hf.backends.onnx.wasm.wasmPaths = value;
},
get wasmPaths() {
return hf.backends.onnx.wasm.wasmPaths;
},
set cacheDir(value) { hf.cacheDir = value; },
get cacheDir() { return hf.cacheDir; },
set wasmPaths(value) { hf.backends.onnx.wasm.wasmPaths = value; },
get wasmPaths() { return hf.backends.onnx.wasm.wasmPaths; },
};

export { TextSplitterStream };
241 changes: 241 additions & 0 deletions kokoro.js/src/phonemize_de.js
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,241 @@
/**
* German text normalisation and phonemisation for kokoro-js.
*
* normalize_text_de() — pure sync function, no network/IO
* phonemize_de() — async, calls eSpeak-NG via phonemizer package
*/

import { phonemize as espeakng } from "phonemizer";

// ─── helpers ──────────────────────────────────────────────────────────────────

/** Split text on regex, keeping delimiters as separate items. */
function splitKeepDelim(text, regex) {
const result = [];
let prev = 0;
for (const m of text.matchAll(regex)) {
if (prev < m.index) result.push({ match: false, text: text.slice(prev, m.index) });
if (m[0].length > 0) result.push({ match: true, text: m[0] });
prev = m.index + m[0].length;
}
if (prev < text.length) result.push({ match: false, text: text.slice(prev) });
return result;
}

// ─── cardinal numbers ─────────────────────────────────────────────────────────

const _ONES = [
"", "ein", "zwei", "drei", "vier", "fünf", "sechs", "sieben",
"acht", "neun", "zehn", "elf", "zwölf", "dreizehn", "vierzehn",
"fünfzehn", "sechzehn", "siebzehn", "achtzehn", "neunzehn",
];
const _TENS = [
"", "", "zwanzig", "dreißig", "vierzig", "fünfzig",
"sechzig", "siebzig", "achtzig", "neunzig",
];

function intToDE(n) {
if (n < 0) return "minus " + intToDE(-n);
if (n === 0) return "null";
if (n < 20) return _ONES[n];
if (n < 100) {
const ones = n % 10, tens = Math.floor(n / 10);
return ones ? _ONES[ones] + "und" + _TENS[tens] : _TENS[tens];
}
if (n < 1_000) {
const h = Math.floor(n / 100), r = n % 100;
return (h === 1 ? "ein" : _ONES[h]) + "hundert" + (r ? intToDE(r) : "");
}
if (n < 1_000_000) {
const t = Math.floor(n / 1_000), r = n % 1_000;
return (t === 1 ? "ein" : intToDE(t)) + "tausend" + (r ? intToDE(r) : "");
}
if (n < 1_000_000_000) {
const m = Math.floor(n / 1_000_000), r = n % 1_000_000;
return (m === 1 ? "eine Million" : intToDE(m) + " Millionen") + (r ? " " + intToDE(r) : "");
}
const b = Math.floor(n / 1_000_000_000), r = n % 1_000_000_000;
return (b === 1 ? "eine Milliarde" : intToDE(b) + " Milliarden") + (r ? " " + intToDE(r) : "");
}

// ─── ordinals ─────────────────────────────────────────────────────────────────

const _ORD_IRREG = { 1: "erst", 2: "zweit", 3: "dritt", 7: "siebt", 8: "acht" };

function ordinalStemDE(n) {
if (_ORD_IRREG[n]) return _ORD_IRREG[n];
return intToDE(n) + (n < 20 ? "t" : "st");
}

// ─── years ────────────────────────────────────────────────────────────────────

function yearDE(n) {
if (n >= 1100 && n <= 1999) {
const c = Math.floor(n / 100), r = n % 100;
return intToDE(c) + "hundert" + (r ? intToDE(r) : "");
}
return intToDE(n);
}

// ─── month names ──────────────────────────────────────────────────────────────

const DE_MONTHS = [
"", "Januar", "Februar", "März", "April", "Mai", "Juni",
"Juli", "August", "September", "Oktober", "November", "Dezember",
];

// ─── currency ─────────────────────────────────────────────────────────────────

const CURRENCY_WORDS = { "€": "Euro", "$": "Dollar", "£": "Pfund", "¥": "Yen" };

function currencyRepl(sym, num) {
const word = CURRENCY_WORDS[sym] || sym;
const cleaned = num.replace(/\./g, "").replace(",", ".");
const val = parseFloat(cleaned);
if (isNaN(val)) return sym + num;
const euros = Math.floor(val);
const cents = Math.round((val - euros) * 100);
if (cents === 0) return intToDE(euros) + " " + word;
return intToDE(euros) + " " + word + " und " + intToDE(cents) + " Cent";
}

// ─── main normalisation ───────────────────────────────────────────────────────

/**
* Normalise German text for TTS phonemisation.
* Handles: quotes, abbreviations, currency, times, dates, ordinals, years, numbers.
*
* @param {string} text
* @returns {string}
*/
export function normalize_text_de(text) {
if (!text) return text;

// 1. Quotes
text = text
.replace(/\u201e|\u201c/g, '"') // „ "
.replace(/\u2018|\u2019/g, "'") // ' '
.replace(/\u00ab|\u00bb/g, '"') // « »
.replace(/\u2039|\u203a/g, '"'); // ‹ ›

// 2. Non-breaking whitespace
text = text.replace(/[^\S \n]/g, " ");

// 3. Abbreviations
text = text
.replace(/\bDr\.(?=\s)/g, "Doktor")
.replace(/\bProf\.(?=\s)/g, "Professor")
.replace(/\bHrn?\.\s/g, "Herrn ")
.replace(/\bFr\.(?=\s[A-ZÄÖÜ])/g, "Frau")
.replace(/\bDipl\.\s*-?\s*Ing\./g, "Diplom-Ingenieur")
.replace(/[Ss]tr\.(?=\s)/g, "Straße") // "Str." and "Hauptstr."
.replace(/\bNr\.(?=\s*\d)/g, "Nummer")
.replace(/\bTel\.(?=\s)/g, "Telefon")
.replace(/\bAbt\.(?=\s)/g, "Abteilung")
.replace(/\bGmbH\b/g, "Gesellschaft mit beschränkter Haftung")
.replace(/\bAG\b(?=[\s,.]|$)/g, "Aktiengesellschaft")
.replace(/\bz\.\s*B\./gi, "zum Beispiel")
.replace(/\bd\.\s*h\./gi, "das heißt")
.replace(/\bu\.\s*a\./gi, "unter anderem")
.replace(/\bbzw\./gi, "beziehungsweise")
.replace(/\busw\./gi, "und so weiter")
.replace(/\betc\./gi, "et cetera")
.replace(/\bca\./gi, "circa")
.replace(/\bvgl\./gi, "vergleiche")
.replace(/\binkl\./gi, "inklusive")
.replace(/\bexkl\./gi, "exklusive")
.replace(/\bggf\./gi, "gegebenenfalls")
.replace(/\bi\.\s*d\.\s*R\./gi, "in der Regel")
.replace(/\bo\.\s*ä\./gi, "oder ähnliches")
.replace(/\bu\.\s*U\./gi, "unter Umständen")
.replace(/\bJan\.(?=\s)/g, "Januar")
.replace(/\bFeb\.(?=\s)/g, "Februar")
.replace(/\bMär\.(?=\s)/g, "März")
.replace(/\bApr\.(?=\s)/g, "April")
.replace(/\bJun\.(?=\s)/g, "Juni")
.replace(/\bJul\.(?=\s)/g, "Juli")
.replace(/\bAug\.(?=\s)/g, "August")
.replace(/\bSep\.(?=\s)/g, "September")
.replace(/\bOkt\.(?=\s)/g, "Oktober")
.replace(/\bNov\.(?=\s)/g, "November")
.replace(/\bDez\.(?=\s)/g, "Dezember");

// 4. Currency — symbol-first (€9,99) and amount-first (9,99€)
const cSym = "[€$£¥]";
text = text.replace(new RegExp(`(${cSym})\\s*(\\d[\\d.,]*)`, "g"), (_, s, n) => currencyRepl(s, n));
text = text.replace(new RegExp(`(\\d[\\d.,]*)\\s*(${cSym})`, "g"), (_, n, s) => currencyRepl(s, n));

// 5. Times hh:mm
text = text.replace(/\b(\d{1,2}):(\d{2})\b/g, (_, h, m) => {
const hi = parseInt(h, 10), mi = parseInt(m, 10);
return intToDE(hi) + " Uhr" + (mi ? " " + intToDE(mi) : "");
});

// 6. Full dates dd.mm.yyyy
text = text.replace(/\b(\d{1,2})\.(\d{1,2})\.(\d{4})\b/g, (_, d, mo, y) => {
const day = parseInt(d, 10), month = parseInt(mo, 10), year = parseInt(y, 10);
if (day < 1 || day > 31 || month < 1 || month > 12) return _;
return ordinalStemDE(day) + "e " + DE_MONTHS[month] + " " + yearDE(year);
});

// 7. Ordinals mid-sentence "am 3. Mai"
text = text.replace(/(?<!\n)(?<!\d)(\d{1,4})\.\s/g, (_, n) => ordinalStemDE(parseInt(n, 10)) + "e ");

// 8. Standalone years (1100-2099)
text = text.replace(/\b(\d{4})\b/g, (_, n) => {
const ni = parseInt(n, 10);
return ni >= 1100 && ni <= 2099 ? yearDE(ni) : intToDE(ni);
});

// 9. German-format numbers: 1.234.567 or 3,14 or plain integers
text = text.replace(/\b\d{1,3}(?:\.\d{3})+(?:,\d+)?\b/g, (m) => {
const cleaned = m.replace(/\./g, "").replace(",", ".");
const val = parseFloat(cleaned);
if (isNaN(val)) return m;
if (Number.isInteger(val)) return intToDE(Math.round(val));
const [ip, fp] = cleaned.split(".");
return intToDE(parseInt(ip, 10)) + " Komma " + fp.split("").map(d => intToDE(parseInt(d, 10))).join(" ");
});
text = text.replace(/\b(\d+),(\d+)\b/g, (_, a, b) =>
intToDE(parseInt(a, 10)) + " Komma " + b.split("").map(d => intToDE(parseInt(d, 10))).join(" ")
);
text = text.replace(/\b(\d+)\b/g, (_, n) => intToDE(parseInt(n, 10)));

// 10. Whitespace
text = text.replace(/[ \t]{2,}/g, " ").replace(/\n{3,}/g, "\n\n").trim();

return text;
}

// ─── phonemisation ────────────────────────────────────────────────────────────

const PUNCT = ';:,.!?¡¿—..."«»""(){}[]';
const PUNCT_RE = new RegExp(`(\\s*[${PUNCT.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}]+\\s*)+`, "g");

/**
* Phonemise German text using eSpeak-NG.
*
* @param {string} text
* @param {boolean} [norm=true] Run normalize_text_de first
* @returns {Promise<string>} IPA phoneme string
*/
export async function phonemize_de(text, norm = true) {
if (norm) text = normalize_text_de(text);

const sections = splitKeepDelim(text, PUNCT_RE);
const parts = await Promise.all(
sections.map(async ({ match, text: chunk }) => {
if (match) return chunk;
const res = await espeakng(chunk, "de");
return res.join(" ");
})
);

return parts
.join("")
// Normalise eSpeak script-g → plain g
.replace(/ɡ/g, "g")
// Collapse double spaces
.replace(/\s{2,}/g, " ")
.trim();
}
Loading