dsh-plugin-lookatstudy 0.12.5 → 0.13.0

package/lib/index.mjs

@@ -4,11 +4,11 @@ import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from "
 import { basename, dirname, join, relative, sep } from "node:path";
 import z from "@deepseek-ai/schemastery";
 import { fileURLToPath } from "node:url";
+import { createHash, randomBytes } from "node:crypto";
+import https from "node:https";
 import { homedir } from "node:os";
-import { randomBytes } from "node:crypto";
 import { readFile, readdir } from "node:fs/promises";
 import { defineTool } from "@deepseek-ai/dsh-tools";
-import https from "node:https";
 //#region \0rolldown/runtime.js
 var __defProp = Object.defineProperty;
 var __exportAll = (all, no_symbols) => {
@@ -233,6 +233,228 @@ function levelFromTotalXp(totalXp) {
 		levelSpan: span
 	};
 }
+"])".slice(0, 1);
+function normalizeSpeechText(md) {
+	let s = md;
+	s = s.replace(/\r\n?/g, "\n");
+	s = s.replace(/(?:^|\n)[ \t]*(?:```|~~~)[^\n]*\n[\s\S]*?(?:\n[ \t]*(?:```|~~~)[^\n]*|\n?$)/g, "\n");
+	s = s.replace(/`[^`\n]*`/g, "");
+	s = s.replace(/!\[[^\]]*\]\([^)\n]*\)/g, "");
+	s = s.replace(/\[([^\]]+)\]\([^)\n]*\)/g, "$1");
+	s = s.replace(/^[ \t]{0,3}#{1,6}[ \t]+/gm, "");
+	s = s.replace(/^[ \t]*(?:[-*+]|\d+\.)[ \t]+/gm, "");
+	s = s.replace(/^[ \t]*>[ \t]?/gm, "");
+	s = s.replace(/\|/g, " ");
+	s = s.replace(/^[ \t]*[-: ]{3,}[ \t]*$/gm, "");
+	s = s.replace(/(\*\*|__)(.*?)\1/g, "$2");
+	s = s.replace(/(?<![*\w])(\*|_)(?!\s)(.+?)(?<!\s)\1(?![*\w])/g, "$2");
+	s = s.replace(/~~(.+?)~~/g, "$1");
+	s = s.replace(/\n{3,}/g, "\n\n");
+	return s.trim();
+}
+//#endregion
+//#region src/tts.ts
+/**
+* Edge read-aloud synthesis: Microsoft's read-aloud endpoint (the one behind
+* Edge's own 朗读 feature), spoken to over a hand-rolled WebSocket on
+* node:https/node:crypto — zero dependencies. Synthesis runs HOST-side (the
+* browser cannot open that socket), so the dashboard streams the cached MP3
+* down and the client falls back to speechSynthesis when this path fails.
+*
+* Three live-verified protocol details this module pins (2026-09-06 probes):
+* 1. Sec-MS-GEC ticks are ~1.36e17 — past Number's 2^53. BigInt or the hash
+*    silently mismatches (403).
+* 2. `Sec-MS-GEC-Version` must carry a CURRENT Edge full version; stale
+*    strings are rejected with 403 before any upgrade.
+* 3. The service sends WebSocket PINGs; unanswered PONGs get the socket reset
+*    mid-turn (read the turn.end path name with the dot — `\w+` eats it).
+* @module dsh-plugin-lookatstudy/tts
+*/
+/** Public read-aloud client token (same constant every edge-tts client ships). */
+const TRUSTED_CLIENT_TOKEN = "6A5AA1D4EAFF4E9FB37E23D68491D6F4";
+/** Must track a shipped Edge release; a stale string is rejected outright. */
+const CHROMIUM_FULL_VERSION = "143.0.3650.75";
+/** The plugin's default tutor voice (晓晓, zh-CN female neural). */
+const DEFAULT_TTS_VOICE = "zh-CN-XiaoxiaoNeural";
+/** Voices the dashboard route accepts (others fall back to the default). */
+const VOICES = /* @__PURE__ */ new Set([
+	DEFAULT_TTS_VOICE,
+	"zh-CN-YunxiNeural",
+	"zh-CN-YunyangNeural",
+	"zh-CN-XiaoyiNeural",
+	"en-US-AriaNeural",
+	"en-US-GuyNeural"
+]);
+function normalizeVoice(voice) {
+	return voice !== void 0 && VOICES.has(voice) ? voice : DEFAULT_TTS_VOICE;
+}
+/**
+* The DRM token (Sec-MS-GEC): SHA-256 over the Windows-filetime ticks of the
+* current 5-minute window plus the trusted token. BigInt is mandatory — the
+* ticks exceed Number.MAX_SAFE_INTEGER and the hash consumes exact digits.
+* @param nowMs - caller clock in unix milliseconds.
+*/
+function secMsGec(nowMs = Date.now()) {
+	let ticks = (BigInt(Math.floor(nowMs / 1e3)) + 11644473600n) * 10000000n;
+	ticks -= ticks % 3000000000n;
+	return createHash("sha256").update(`${ticks}${TRUSTED_CLIENT_TOKEN}`).digest("hex").toUpperCase();
+}
+function xmlEscape(s) {
+	return s.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll("\"", "&quot;").replaceAll("'", "&apos;");
+}
+/** Build one speak utterance. */
+function buildSsml(text, voice) {
+	return `<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='${voice.split("-").slice(0, 2).join("-")}'><voice name='${voice}'><prosody rate='+0%' pitch='+0Hz'>${xmlEscape(text)}</prosody></voice></speak>`;
+}
+/** Client→server frames are masked; server→client frames are not. */
+function maskFrame(opcode, payload) {
+	const mask = randomBytes(4);
+	const head = payload.length < 126 ? Buffer.from([128 | opcode, 128 | payload.length]) : Buffer.from([
+		128 | opcode,
+		254,
+		payload.length >> 8,
+		payload.length & 255
+	]);
+	const masked = Buffer.from(payload.map((b, i) => b ^ mask[i % 4]));
+	return Buffer.concat([
+		head,
+		mask,
+		masked
+	]);
+}
+const timestamp = () => (/* @__PURE__ */ new Date()).toISOString().replace(/\.\d{3}Z$/, "Z");
+/**
+* Synthesize one utterance to MP3 (audio-24khz-48kbit-mono-mp3).
+* @param text - plain speakable text (short: one sentence or a small group).
+* @param voice - neural voice name.
+* @param timeoutMs - hard ceiling on the whole turn.
+* @returns the MP3 bytes.
+*/
+function synthesizeSpeech(text, voice = DEFAULT_TTS_VOICE, timeoutMs = 2e4) {
+	return new Promise((resolve, reject) => {
+		const gec = secMsGec();
+		const host = "speech.platform.bing.com";
+		const path = `/consumer/speech/synthesize/readaloud/edge/v1?TrustedClientToken=${TRUSTED_CLIENT_TOKEN}&Sec-MS-GEC=${gec}&Sec-MS-GEC-Version=1-${CHROMIUM_FULL_VERSION}`;
+		const key = randomBytes(16).toString("base64");
+		const req = https.request(`https://${host}${path}`, { headers: {
+			Connection: "Upgrade",
+			Upgrade: "websocket",
+			"Sec-WebSocket-Key": key,
+			"Sec-WebSocket-Version": "13",
+			Pragma: "no-cache",
+			"Cache-Control": "no-cache",
+			Origin: "chrome-extension://jdiccldimpdaibmpdkjnbmckianbfold",
+			"User-Agent": `Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/${CHROMIUM_FULL_VERSION.split(".")[0]}.0.0.0 Safari/537.36 Edg/${CHROMIUM_FULL_VERSION}`
+		} });
+		const timer = setTimeout(() => {
+			socket?.destroy();
+			reject(/* @__PURE__ */ new Error("edge tts: turn timed out"));
+		}, timeoutMs);
+		const debug = process.env.LKS_TTS_DEBUG === "1";
+		const log = (line) => {
+			if (debug) console.error(`[lks-tts] ${line}`);
+		};
+		let socket;
+		const finish = (err, mp3) => {
+			clearTimeout(timer);
+			socket?.destroy();
+			if (err !== null) reject(err);
+			else resolve(mp3);
+		};
+		req.on("upgrade", (res, sock) => {
+			log(`upgrade ok (${String(res.statusCode)})`);
+			socket = sock;
+			const accept = createHash("sha1").update(`${key}258EAFA5-E914-47DA-95CA-C5AB0DC85B11`).digest("base64");
+			if (res.headers["sec-websocket-accept"] !== accept) return finish(/* @__PURE__ */ new Error("edge tts: bad upgrade accept"));
+			if (res.statusCode !== 101) return finish(/* @__PURE__ */ new Error(`edge tts: upgrade refused (${String(res.statusCode)})`));
+			sock.setNoDelay(true);
+			sock.write(maskFrame(1, Buffer.from(`X-Timestamp:${timestamp()}\r\nContent-Type:application/json; charset=utf-8\r\nPath:speech.config\r\n\r\n` + JSON.stringify({ context: { synthesis: { audio: {
+				metadataoptions: {
+					sentenceBoundaryEnabled: "false",
+					wordBoundaryEnabled: "false"
+				},
+				outputFormat: "audio-24khz-48kbitrate-mono-mp3"
+			} } } }))));
+			sock.write(maskFrame(1, Buffer.from(`X-RequestId:${randomBytes(16).toString("hex")}\r\nContent-Type:application/ssml+xml\r\nX-Timestamp:${timestamp()}\r\nPath:ssml\r\n\r\n${buildSsml(text, voice)}`)));
+			let acc = Buffer.alloc(0);
+			const audio = [];
+			sock.on("data", (chunk) => {
+				if (debug) log(`recv ${chunk.length}B head=${chunk.subarray(0, 8).toString("hex")}`);
+				acc = Buffer.concat([acc, chunk]);
+				for (;;) {
+					if (acc.length < 2) return;
+					const opcode = acc[0] & 15;
+					let len = acc[1] & 127;
+					let off = 2;
+					if (len === 126) {
+						if (acc.length < 4) return;
+						len = acc.readUInt16BE(2);
+						off = 4;
+					} else if (len === 127) {
+						if (acc.length < 10) return;
+						len = Number(acc.readBigUInt64BE(2));
+						off = 10;
+					}
+					if (acc.length < off + len) return;
+					const payload = acc.subarray(off, off + len);
+					acc = acc.subarray(off + len);
+					if (opcode === 9) {
+						log("ping→pong");
+						sock.write(maskFrame(10, payload));
+						continue;
+					}
+					if (opcode === 8) {
+						log("close frame");
+						return finish(audio.length > 0 ? null : /* @__PURE__ */ new Error("edge tts: closed before audio"));
+					}
+					if (opcode !== 1 && opcode !== 2) continue;
+					const headerLen = payload.readUInt16BE(0);
+					const head = payload.subarray(2, 2 + headerLen).toString("utf8");
+					if (debug) log(`frame op=${opcode} len=${len} path=${/Path:([\w.]+)/.exec(head)?.[1] ?? "?"}`);
+					if (opcode === 1 && /Path:turn\.end/.test(head)) {
+						log(`turn.end with ${audio.length} audio chunk(s)`);
+						return finish(audio.length > 0 ? null : /* @__PURE__ */ new Error("edge tts: turn ended without audio"), Buffer.concat(audio));
+					}
+					if (opcode === 2 && /Path:audio/.test(head)) {
+						const chunk = payload.subarray(2 + headerLen);
+						audio.push(chunk);
+						if (debug) log(`audio chunk ${chunk.length}B (total ${audio.reduce((n, c) => n + c.length, 0)}B)`);
+					}
+				}
+			});
+			sock.on("error", (err) => finish(err));
+			sock.on("close", () => {
+				if (timer.hasRef()) finish(/* @__PURE__ */ new Error("edge tts: socket closed mid-turn"));
+			});
+		});
+		req.on("response", (res) => {
+			res.resume();
+			finish(/* @__PURE__ */ new Error(`edge tts: endpoint refused the upgrade (HTTP ${String(res.statusCode)})`));
+		});
+		req.on("error", (err) => finish(err));
+		req.end();
+	});
+}
+/** Cache key: voice-scoped (different voices never share files). */
+function ttsCachePath(cacheDir, text, voice) {
+	const hash = createHash("sha256").update(`${voice}\n${text}`).digest("hex");
+	return join(cacheDir, `${hash}.mp3`);
+}
+/**
+* Cache-first synthesis: study-area/tts-cache/{sha256(voice+text)}.mp3.
+* Misses synthesize once and persist; later listens (and other learners on
+* the same install) replay from disk without touching the endpoint.
+* @param synth - the synthesizer (injectable for tests; real Edge TTS by default).
+*/
+async function cachedTtsMp3(cacheDir, text, voice = DEFAULT_TTS_VOICE, synth = synthesizeSpeech) {
+	const file = ttsCachePath(cacheDir, text, voice);
+	if (existsSync(file)) return readFileSync(file);
+	mkdirSync(cacheDir, { recursive: true });
+	const mp3 = await synth(text, voice);
+	if (mp3.length === 0) throw new Error("edge tts: synthesis returned empty audio (not cached)");
+	writeFileSync(file, mp3);
+	return mp3;
+}
 //#endregion
 //#region src/vendor/sm2.ts
 function computeSm2(prev, quality, now = /* @__PURE__ */ new Date()) {
@@ -1412,7 +1634,8 @@ function workbenchState(state, now) {
 				quote: n.quote
 			})),
 			html: renderMarkdown(normalizeMathNotation(ref.lesson.translation === void 0 ? ref.lesson.body : renderBilingual(ref.lesson.body, ref.lesson.translation))),
-			markdown: ref.lesson.body
+			markdown: ref.lesson.body,
+			speechText: normalizeSpeechText(normalizeMathNotation(ref.lesson.body))
 		};
 	} catch {
 		lesson = null;
@@ -1650,6 +1873,40 @@ function registerDashboard(webServer, deps) {
 				sendJson(res, 200, { ok: true });
 				return;
 			}
+			if (req.method === "POST" && pathname === "/lookatstudy/api/tts") {
+				const body = await readJsonBodySafe(req, res);
+				if (body === void 0) return;
+				const text = typeof body.text === "string" ? body.text.trim() : "";
+				if (text === "") {
+					sendJson(res, 400, {
+						ok: false,
+						error: "text (non-empty string) required"
+					});
+					return;
+				}
+				if (text.length > 4e3) {
+					sendJson(res, 400, {
+						ok: false,
+						error: "text exceeds 4000 chars (speak sentence groups, not whole lessons)"
+					});
+					return;
+				}
+				const voice = normalizeVoice(typeof body.voice === "string" ? body.voice : void 0);
+				const cacheDir = join(deps.studyAreaPath, "tts-cache");
+				try {
+					sendJson(res, 200, {
+						ok: true,
+						mime: "audio/mpeg",
+						dataBase64: (await cachedTtsMp3(cacheDir, text, voice, deps.tts?.synthesize)).toString("base64")
+					});
+				} catch (error) {
+					sendJson(res, 502, {
+						ok: false,
+						error: error instanceof Error ? error.message : String(error)
+					});
+				}
+				return;
+			}
 			if (req.method === "POST" && pathname === "/lookatstudy/api/mode") {
 				const body = await readJsonBodySafe(req, res);
 				if (body === void 0) return;
@@ -1727,6 +1984,12 @@ const ZH = {
 	"rail.delete.title.confirm": "再点一次确认删除(含全部进度与笔记)。反悔前可先在设置页备份状态文件",
 	"note.delete": "删除本条笔记",
 	"note.delete.confirm": "确认删除?",
+	"read.play": "朗读本课",
+	"read.stop": "停止朗读",
+	"read.pause": "暂停朗读",
+	"read.resume": "继续朗读",
+	"read.engine.system": "网络合成不可用，已切换为系统语音",
+	"read.unavailable": "朗读不可用",
 	"rail.section.collapse": "折叠本章节",
 	"rail.section.expand": "展开本章节({count} 课时)",
 	"rail.section.count": "{count} 课",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-plugin-lookatstudy",
-  "version": "0.12.5",
+  "version": "0.13.0",
   "packageManager": "pnpm@11.7.0",
   "description": "Turn any markdown, local folder, or GitHub learning repo into a guided course inside DeepSeek Harness: gated skill-tree progression, BKT mastery tracking, SM-2 spaced repetition.",
   "type": "module",