// The en-gb front-end loader (16 Nov). For the release: it reads en-gb/override.public.tsv (the en-gb // override minus the same 51 entries dropped from the public en-us list, OVERRIDE-REVIEW.md). // The en-gb loader for the release's JS API (release-draft/js/kale-tts.mjs), the same shape as its loadEnUS: // import { loadEnGB } from "../frontend/release-loader.mjs"; KaleTTS.registerFrontEnd("en-gb", loadEnGB); // or put loadEnGB into FRONT_ENDS["en-gb"] directly. voices.json "accent": "en-gb" then picks it by itself (d03, d13). // Files: /words/en-gb/{lexicon.tsv, override.public.tsv, terms.tsv, g2p.bin} + /words/enGb.js. // The model's tokenizer.json needs "ɒ": 71 (vocab + normaliser regex); every other en-gb symbol is in it already. import { G2pModel, readTable, sentences } from "./words/kaleTts.js"; import { BritishG2p, britishNormalizer } from "./words/enGb.js"; const isNode = typeof process !== "undefined" && !!process.versions?.node && typeof window === "undefined"; async function readBytes(url) { if (isNode && !/^https?:/.test(url)) { const { readFile } = await import("node:fs/promises"); const b = await readFile(url); return b.buffer.slice(b.byteOffset, b.byteOffset + b.byteLength); } const r = await fetch(url); if (!r.ok) throw new Error(`${url}: ${r.status}`); return r.arrayBuffer(); } const readText = async (url) => new TextDecoder().decode(await readBytes(url)); const join = (base, p) => (base.endsWith("/") ? base : base + "/") + p; /** terms table: whole tokens, longest first; camelCase split; again (same as the release's en-us) */ function termsReplacer(table) { const terms = readTable(table); if (terms.size === 0) return (t) => t; const esc = (x) => x.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"); // a written form starting with "." is a suffix too: ".js" also matches in "Node.js" const forms = [...terms.keys()].sort((a, b) => b.length - a.length).map((f) => (f.startsWith(".") ? esc(f) : `(? t.replace(pattern, (f, at, whole) => { const spoken = terms.get(f) ?? f; return f.startsWith(".") && at > 0 && /[\p{L}\p{N}]/u.test(whole[at - 1]) ? ` ${spoken}` : spoken; // "Node.js" -> "Node dot J S" }); const CAMEL = /(? t.replace(CAMEL, (w) => w.replace(/(?<=[a-z0-9])(?=[A-Z])/g, " ")); return (t) => lookUp(split(lookUp(t))); } /** British English: files under /words/en-gb/ */ export async function loadEnGB(frontendBase) { const d = join(frontendBase, "words/en-gb/"); const [lexicon, override, terms, g2pBin] = await Promise.all([readText(d + "lexicon.tsv"), readText(d + "override.public.tsv"), readText(d + "terms.tsv"), readBytes(d + "g2p.bin")]); const lex = readTable(lexicon); readTable(override, lex); const g2p = new BritishG2p(lex, new G2pModel(g2pBin)); const norm = britishNormalizer(lex); const terms_ = termsReplacer(terms); const normalize = (t) => norm(terms_(t)); return { sentences: (text) => sentences(text, normalize), ipa: (s) => g2p.phonemes(s) }; }