Download python/frontend_entry.mjs from KaleComputer/kale-tts-1-tiny: direct link, hf CLI and curl.
- Browser
- Download file 4 kB
-
https://huggingface.co/KaleComputer/kale-tts-1-tiny/resolve/main/python/frontend_entry.mjs
- Command line
-
hf download hf://KaleComputer/kale-tts-1-tiny/python/frontend_entry.mjs
-
curl -L -o frontend_entry.mjs https://huggingface.co/KaleComputer/kale-tts-1-tiny/resolve/main/python/frontend_entry.mjs
4 kB
| // the release text front end as one plain script for embedded JS engines (QuickJS from Python, JavaScriptCore on iOS, ...): | |
| // globalThis.kaleFrontEnd.init(lexiconTsv, overrideTsv, termsTsv, g2pBase64); kaleFrontEnd.run(text) -> JSON [{text, ipa}] | |
| // Same code path as js/kale-tts.mjs (dizione rules -> terms -> CMUdict lexicon + overrides + g2p model). Build: npm run build-frontend | |
| import { G2p, G2pModel, normalizer, readTable, sentences } from "../frontend/words/kaleTts.js"; | |
| import { BritishG2p, britishNormalizer } from "../frontend/words/enGb.js"; | |
| import { EttinTokenizer, ettinInputs, wordIndex } from "../frontend/ettin.js"; | |
| function termsReplacer(table) { // = js/kale-tts.mjs termsReplacer | |
| 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) : `(?<![\\p{L}\\p{N}_])${esc(f)}`)); | |
| const pattern = new RegExp(`(?:${forms.join("|")})(?![\\p{L}\\p{N}_])`, "gu"); | |
| const lookUp = (t) => 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]) ? ` ${spoken}` : spoken; // "Node.js" -> "Node dot J S" | |
| }); | |
| const CAMEL = /(?<![\p{L}\p{N}_])[a-z]+(?:[A-Z][a-z0-9]*)+(?![\p{L}\p{N}_])/gu; | |
| const split = (t) => t.replace(CAMEL, (w) => w.replace(/(?<=[a-z0-9])(?=[A-Z])/g, " ")); | |
| return (t) => lookUp(split(lookUp(t))); | |
| } | |
| function b64(s) { | |
| const A = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/", m = new Uint8Array(256); for (let i = 0; i < 64; i++) m[A.charCodeAt(i)] = i; | |
| s = s.replace(/=+$/, ""); const out = new Uint8Array((s.length * 3) >> 2); let o = 0; | |
| for (let i = 0; i < s.length; i += 4) { const n = (m[s.charCodeAt(i)] << 18) | (m[s.charCodeAt(i + 1)] << 12) | (m[s.charCodeAt(i + 2)] << 6) | m[s.charCodeAt(i + 3)]; | |
| out[o++] = n >> 16; if (o < out.length) out[o++] = (n >> 8) & 255; if (o < out.length) out[o++] = n & 255; } | |
| return out.buffer; | |
| } | |
| const fes = {}; | |
| function make(lexicon, override, terms, g2pB64, gb) { | |
| const lex = readTable(lexicon); readTable(override, lex); | |
| const g2p = gb ? new BritishG2p(lex, new G2pModel(b64(g2pB64))) : new G2p(lex, new G2pModel(b64(g2pB64))); | |
| const norm = gb ? britishNormalizer(lex) : normalizer(lex), applyTerms = termsReplacer(terms); | |
| return { g2p, normalize: (t) => norm(applyTerms(t)) }; | |
| } | |
| globalThis.kaleFrontEnd = { | |
| init(lexicon, override, terms, g2pB64) { fes["en-us"] = make(lexicon, override, terms, g2pB64, false); return true; }, | |
| initGb(lexicon, override, terms, g2pB64) { fes["en-gb"] = make(lexicon, override, terms, g2pB64, true); return true; }, // = frontend/release-loader.mjs | |
| initEttin(tokenizerJson) { fes.ettin = new EttinTokenizer(JSON.parse(tokenizerJson)); return true; }, | |
| // Ettin inputs for one sentence (kale-tts-1-tiny's Ettin text side): sub-tokens with the previous sentence as context, word averages, | |
| // and the word index for the sentence's sound ids (sentIds, per word: the same front end's ids for that word) | |
| ettin(text, prev, sentIdsJson, wordIdsJson) { | |
| const e = ettinInputs(fes.ettin, text, prev), words = text.split(/\s+/).filter(Boolean); | |
| return JSON.stringify({ ids: e.ettin_ids, W: e.W, S: e.S, wavg: Array.from(e.wavg), widx: wordIndex(JSON.parse(sentIdsJson), JSON.parse(wordIdsJson), words.length) }); | |
| }, | |
| ipaWords(text, accent = "en-us") { // the sentence's words through the same g2p, one IPA string per word | |
| const fe = fes[accent] ?? fes["en-us"]; | |
| return JSON.stringify(text.split(/\s+/).filter(Boolean).map((w) => fe.g2p.phonemes(w))); | |
| }, | |
| run(text, accent = "en-us") { | |
| const fe = fes[accent] ?? fes["en-us"]; | |
| return JSON.stringify(sentences(text, fe.normalize).map((s) => ({ text: s, ipa: fe.g2p.phonemes(s) }))); | |
| }, | |
| }; | |