diff --git a/scripts/check.mjs b/scripts/check.mjs index d352257..87b07c0 100644 --- a/scripts/check.mjs +++ b/scripts/check.mjs @@ -112,7 +112,7 @@ if (/https:\/\/esm\.sh/i.test(cloud)) { const piper = await readFile("src/piper-tts.js", "utf8"); for (const marker of [ - 'export const VOICE = "/piper/en_US-lessac-high"', + 'export const VOICE = "https://voice.shadow.wang/piper/en_US-ljspeech-medium"', 'export const VOICE_FILES', ]) { if (!piper.includes(marker)) throw new Error(`piper-tts.js is missing ${marker}`); diff --git a/src/piper-tts.js b/src/piper-tts.js index 6a24237..10160b4 100644 --- a/src/piper-tts.js +++ b/src/piper-tts.js @@ -1,27 +1,19 @@ /* 本地 Piper 英语语音兜底（无 GMS 的国产 Android） * * 系统语音（speechSynthesis）在无 GMS 的国产 Android 上通常不可用， - * 影伴提供浏览器本地 Piper 合成作为兜底：模型与运行时全部托管在本应用内， - * 不上传录音，可离线使用。 + * 影伴提供浏览器本地 Piper 合成作为兜底：运行时托管在本应用内，模型首次从 CDN 下载， + * 下载完成后缓存到浏览器，不上传录音，可离线使用。 * * 致谢（开源项目详见 README 致谢一节）： * - piper-tts-web（MIT） - * - rhasspy/piper 语音模型 en_US-lessac-high + * - rhasspy/piper 语音模型 en_US-ljspeech-medium（模型卡标注训练数据为 public domain） * - ONNX Runtime Web（MIT） */ -export const VOICE = "/piper/en_US-lessac-high"; +export const VOICE = "https://voice.shadow.wang/piper/en_US-ljspeech-medium"; const VOICE_CACHE = "shadow-mate-voice"; const ENGINE_URL = "/piper-tts-web.js"; -export const VOICE_MODEL_PARTS = [VOICE + ".onnx.part-00", VOICE + ".onnx.part-01"]; -export const VOICE_FILES = [...VOICE_MODEL_PARTS, VOICE + ".onnx.json"]; -// Android 某些网络环境会隐藏响应的 Content-Length；这里用当前随包文件大小兜底计算百分比。 -// 替换语音模型文件时，需要同步更新这些数值。 -export const VOICE_FILE_SIZES = { - [VOICE_MODEL_PARTS[0]]: 62_914_560, - [VOICE_MODEL_PARTS[1]]: 50_980_641, - [VOICE + ".onnx.json"]: 4_883, -}; +export const VOICE_FILES = [VOICE + ".onnx", VOICE + ".onnx.json"]; export const ENGINE_LOAD_TIMEOUT_MS = 60_000; export const SYNTHESIS_TIMEOUT_MS = 30_000; @@ -62,9 +54,7 @@ async function getContentLength(url, signal) { export async function isVoiceCached() { if (!("caches" in window)) return false; try { - const cache = await openVoiceCache(); - const cachedFiles = await Promise.all(VOICE_FILES.map((url) => cache.match(url))); - return cachedFiles.every(Boolean); + return !!(await (await openVoiceCache()).match(VOICE + ".onnx")); } catch (_) { return false; } @@ -78,10 +68,7 @@ export async function downloadVoice(onProgress, signal) { } const lengths = await Promise.all(pendingFiles.map((url) => getContentLength(url, signal))); - const progressLengths = lengths.map((length, index) => length || VOICE_FILE_SIZES[pendingFiles[index]] || 0); - const total = progressLengths.every((length) => length > 0) - ? progressLengths.reduce((sum, length) => sum + length, 0) - : 0; + const total = lengths.every((length) => length > 0) ? lengths.reduce((sum, length) => sum + length, 0) : 0; let receivedTotal = 0; for (let fileIndex = 0; fileIndex < pendingFiles.length; fileIndex += 1) { @@ -89,7 +76,7 @@ export async function downloadVoice(onProgress, signal) { if (signal?.aborted) throw new DOMException("The operation was aborted.", "AbortError"); const res = await fetch(url, { signal }); if (!res.ok) throw new Error("语音包下载失败"); - const fileTotal = Number(res.headers.get("content-length")) || progressLengths[fileIndex] || 0; + const fileTotal = Number(res.headers.get("content-length")) || lengths[fileIndex] || 0; const reader = res.body.getReader(); const chunks = []; let receivedFile = 0; @@ -133,7 +120,7 @@ async function loadEngine() { const mod = await import(/* @vite-ignore */ ENGINE_URL); const voiceProvider = { async fetch(voice) { - const readResponse = async (url) => { + const read = async (url) => { let res = null; if ("caches" in window) { try { @@ -144,24 +131,14 @@ async function loadEngine() { } if (!res) res = await fetch(url); if (!res.ok) throw new Error("语音模型读取失败"); - return res; + return url.endsWith(".json") ? res.json() : URL.createObjectURL(await res.blob()); }; - const config = await (await readResponse(voice + ".onnx.json")).json(); - const modelParts = await Promise.all( - VOICE_MODEL_PARTS.map(async (url) => (await readResponse(url)).blob()) - ); - const modelUrl = URL.createObjectURL(new Blob(modelParts, { type: "application/octet-stream" })); - return [config, modelUrl]; + return Promise.all([read(voice + ".onnx.json"), read(voice + ".onnx")]); }, }; - const useWorkerRuntime = typeof Worker === "function" - && typeof mod.OnnxWebWorkerRuntime === "function" - && typeof mod.PhonemizeWebWorkerRuntime === "function"; - const OnnxRuntime = useWorkerRuntime ? mod.OnnxWebWorkerRuntime : mod.OnnxWebRuntime; - const PhonemizeRuntime = useWorkerRuntime ? mod.PhonemizeWebWorkerRuntime : mod.PhonemizeWebRuntime; return new mod.PiperWebEngine({ - onnxRuntime: new OnnxRuntime({ basePath: "/onnx/", numThreads: 1 }), - phonemizeRuntime: new PhonemizeRuntime({ basePath: "/piper/" }), + onnxRuntime: new mod.OnnxWebRuntime({ basePath: "/onnx/", numThreads: 1 }), + phonemizeRuntime: new mod.PhonemizeWebRuntime({ basePath: "/piper/" }), voiceProvider, }); })().catch((err) => { @@ -190,7 +167,7 @@ function buildDialog() { dlg.className = "voice-dialog"; dlg.innerHTML = '
离线英语语音
' + - '
当前设备没有可用的英语发音。影伴内置了约 115MB 的高质量离线英语语音包（一次性下载，之后可离线使用，不上传录音）。是否现在下载？
' + + '
当前设备没有可用的英语发音。影伴内置了约 90MB 的离线英语语音包（一次性下载，之后可离线使用，不上传录音）。是否现在下载？
' + '' + '
' + '' + diff --git a/tests/e2e/offline-tts-error.spec.js b/tests/e2e/offline-tts-error.spec.js index bf4b677..e251cd6 100644 --- a/tests/e2e/offline-tts-error.spec.js +++ b/tests/e2e/offline-tts-error.spec.js @@ -16,7 +16,7 @@ test.describe("Offline voice download errors", () => { `, }); }); - await page.route("**/piper/en_US-lessac-high.onnx.part-00", async (route) => { + await page.route("https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", async (route) => { await route.fulfill({ status: 503, contentType: "text/plain", body: "unavailable" }); }); await page.goto("/"); diff --git a/tests/e2e/offline-tts-warmup.spec.js b/tests/e2e/offline-tts-warmup.spec.js index 2b5f5cc..e216b1a 100644 --- a/tests/e2e/offline-tts-warmup.spec.js +++ b/tests/e2e/offline-tts-warmup.spec.js @@ -50,7 +50,7 @@ test.describe("Offline voice warmup", () => { `, }); }); - await page.route("**/piper/en_US-lessac-high.onnx*", async (route) => { + await page.route("https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx*", async (route) => { const request = route.request(); const isConfig = request.url().endsWith(".json"); if (request.method() === "HEAD") { @@ -130,9 +130,8 @@ test.describe("Offline voice warmup", () => { value: function SpeechSynthesisUtterance() {}, }); const cacheStore = new Map([ - ["/piper/en_US-lessac-high.onnx.part-00", new Response(new Blob(["cached model 00"]))], - ["/piper/en_US-lessac-high.onnx.part-01", new Response(new Blob(["cached model 01"]))], - ["/piper/en_US-lessac-high.onnx.json", new Response("{}")], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", new Response(new Blob(["cached model"]))], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx.json", new Response("{}")], ]); Object.defineProperty(window, "caches", { configurable: true, @@ -193,9 +192,8 @@ test.describe("Offline voice warmup", () => { value: function SpeechSynthesisUtterance() {}, }); const cacheStore = new Map([ - ["/piper/en_US-lessac-high.onnx.part-00", new Response(new Blob(["cached model 00"]))], - ["/piper/en_US-lessac-high.onnx.part-01", new Response(new Blob(["cached model 01"]))], - ["/piper/en_US-lessac-high.onnx.json", new Response("{}")], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", new Response(new Blob(["cached model"]))], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx.json", new Response("{}")], ]); Object.defineProperty(window, "caches", { configurable: true, @@ -229,7 +227,7 @@ test.describe("Offline voice warmup", () => { await expect(page.locator("audio")).toHaveCount(0); }); - test("runs local speech inference in worker runtimes when available", async ({ page }) => { + test("runs local speech inference on the main thread", async ({ page }) => { await disableWebAudio(page); await page.route("**/piper-tts-web.js*", async (route) => { await route.fulfill({ @@ -276,9 +274,8 @@ test.describe("Offline voice warmup", () => { value: function SpeechSynthesisUtterance() {}, }); const cacheStore = new Map([ - ["/piper/en_US-lessac-high.onnx.part-00", new Response(new Blob(["cached model 00"]))], - ["/piper/en_US-lessac-high.onnx.part-01", new Response(new Blob(["cached model 01"]))], - ["/piper/en_US-lessac-high.onnx.json", new Response("{}")], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", new Response(new Blob(["cached model"]))], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx.json", new Response("{}")], ]); Object.defineProperty(window, "caches", { configurable: true, @@ -295,7 +292,7 @@ test.describe("Offline voice warmup", () => { await page.locator("[data-speak]").first().click(); await expect.poll(() => page.evaluate(() => window.__ttsRuntimeKinds), { timeout: 3000 }) - .toEqual(["worker", "worker"]); + .toEqual(["main", "main"]); }); test("plays generated speech through a decoded Web Audio buffer when available", async ({ page }) => { @@ -338,9 +335,8 @@ test.describe("Offline voice warmup", () => { value: function SpeechSynthesisUtterance() {}, }); const cacheStore = new Map([ - ["/piper/en_US-lessac-high.onnx.part-00", new Response(new Blob(["cached model 00"]))], - ["/piper/en_US-lessac-high.onnx.part-01", new Response(new Blob(["cached model 01"]))], - ["/piper/en_US-lessac-high.onnx.json", new Response("{}")], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", new Response(new Blob(["cached model"]))], + ["https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx.json", new Response("{}")], ]); Object.defineProperty(window, "caches", { configurable: true, diff --git a/tests/unit/piper-tts.test.js b/tests/unit/piper-tts.test.js index 934ce9e..2fa2f0d 100644 --- a/tests/unit/piper-tts.test.js +++ b/tests/unit/piper-tts.test.js @@ -1,5 +1,5 @@ import { afterEach, describe, expect, test, vi } from "vitest"; -import { askDownloadVoice, downloadVoice, isVoiceCached, VOICE, VOICE_FILES, withTimeout } from "../../src/piper-tts.js"; +import { askDownloadVoice, downloadVoice, VOICE, VOICE_FILES, withTimeout } from "../../src/piper-tts.js"; const originalDialogDescriptors = new Map( ["showModal", "close"].map((name) => [name, Object.getOwnPropertyDescriptor(HTMLDialogElement.prototype, name)]) @@ -34,27 +34,15 @@ afterEach(() => { }); describe("offline Piper voice download", () => { - test("uses the high-quality lessac voice path", () => { - expect(VOICE).toBe("/piper/en_US-lessac-high"); + test("uses the commercially reviewed replacement voice path", () => { + expect(VOICE).toBe("https://voice.shadow.wang/piper/en_US-ljspeech-medium"); expect(VOICE_FILES).toEqual([ - "/piper/en_US-lessac-high.onnx.part-00", - "/piper/en_US-lessac-high.onnx.part-01", - "/piper/en_US-lessac-high.onnx.json", + "https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx", + "https://voice.shadow.wang/piper/en_US-ljspeech-medium.onnx.json", ]); }); - test("treats every split model part and config as one cached voice", async () => { - const cachedFiles = new Set(VOICE_FILES); - const cache = { - match: vi.fn((url) => Promise.resolve(cachedFiles.has(url) ? new Response("cached") : undefined)), - }; - vi.stubGlobal("caches", { open: vi.fn().mockResolvedValue(cache) }); - - await expect(isVoiceCached()).resolves.toBe(true); - expect(cache.match.mock.calls.map(([url]) => url)).toEqual(VOICE_FILES); - }); - - test("reports aggregate progress using bundled sizes when the response omits Content-Length", async () => { + test("reports received bytes when the response omits Content-Length", async () => { const cache = { match: vi.fn().mockResolvedValue(undefined), put: vi.fn().mockResolvedValue(undefined), @@ -65,7 +53,7 @@ describe("offline Piper voice download", () => { vi.fn((url, options) => { if (options?.method === "HEAD") return Promise.resolve(responseWithChunks([])); return Promise.resolve( - url.includes(".onnx.part-") + url.endsWith(".onnx") ? responseWithChunks([Uint8Array.from([1, 2]), Uint8Array.from([3, 4, 5])]) : responseWithChunks([Uint8Array.from([6])]) ); @@ -75,40 +63,8 @@ describe("offline Piper voice download", () => { await downloadVoice((received, total) => progress.push([received, total])); + expect(progress).toContainEqual([5, 0]); expect(progress.some(([received]) => received > 0)).toBe(true); - expect(progress.some(([received, total]) => received > 0 && total > 100_000_000)).toBe(true); - }); - - test("shows a percentage in the download dialog when Content-Length is unavailable", async () => { - Object.defineProperty(HTMLDialogElement.prototype, "showModal", { - configurable: true, - value() { - this.open = true; - }, - }); - Object.defineProperty(HTMLDialogElement.prototype, "close", { - configurable: true, - value() { - this.open = false; - }, - }); - - const cache = { - match: vi.fn().mockResolvedValue(undefined), - put: vi.fn().mockResolvedValue(undefined), - }; - vi.stubGlobal("caches", { open: vi.fn().mockResolvedValue(cache) }); - vi.stubGlobal("fetch", vi.fn(() => Promise.resolve(responseWithChunks([Uint8Array.from([1])])))); - - const result = askDownloadVoice(); - const dialog = document.querySelector("#shadow-voice-dialog"); - dialog.querySelector('[data-action="ok"]').click(); - - await vi.waitFor(() => { - expect(dialog.querySelector(".voice-dialog-pct").textContent).toMatch(/^\d+%$/); - }); - expect(dialog.querySelector(".voice-dialog-bar i").classList.contains("indeterminate")).toBe(false); - await expect(result).resolves.toBe("ok"); }); test("reports aggregate bytes across the model and config files", async () => { @@ -120,11 +76,11 @@ describe("offline Piper voice download", () => { vi.stubGlobal( "fetch", vi.fn((url) => { - const isModelPart = url.includes(".onnx.part-"); + const isModel = url.endsWith(".onnx"); return Promise.resolve( responseWithChunks( - isModelPart ? [Uint8Array.from([1, 2, 3]), Uint8Array.from([4, 5])] : [Uint8Array.from([6])], - { "content-length": isModelPart ? "5" : "1" } + isModel ? [Uint8Array.from([1, 2, 3]), Uint8Array.from([4, 5])] : [Uint8Array.from([6])], + { "content-length": isModel ? "5" : "1" } ) ); }) @@ -133,11 +89,9 @@ describe("offline Piper voice download", () => { await downloadVoice((received, total) => progress.push([received, total])); - expect(progress).toContainEqual([3, 11]); - expect(progress).toContainEqual([5, 11]); - expect(progress).toContainEqual([8, 11]); - expect(progress).toContainEqual([10, 11]); - expect(progress.at(-1)).toEqual([11, 11]); + expect(progress).toContainEqual([3, 6]); + expect(progress).toContainEqual([5, 6]); + expect(progress.at(-1)).toEqual([6, 6]); }); test("rejects when speech synthesis never settles", async () => { diff --git a/vercel.json b/vercel.json index 117588e..acbabac 100644 --- a/vercel.json +++ b/vercel.json @@ -21,7 +21,7 @@ }, { "key": "Content-Security-Policy", - "value": "default-src 'self'; script-src 'self' 'wasm-unsafe-eval'; style-src 'self'; connect-src 'self' blob: https://dutepjyocxcvecmsrtfp.supabase.co wss://dutepjyocxcvecmsrtfp.supabase.co; media-src 'self' blob:; img-src 'self' data:; font-src 'self'; object-src 'none'; base-uri 'self'; frame-ancestors 'none'; form-action 'self'; upgrade-insecure-requests" + "value": "default-src 'self'; script-src 'self' 'wasm-unsafe-eval'; style-src 'self'; connect-src 'self' blob: https://voice.shadow.wang https://dutepjyocxcvecmsrtfp.supabase.co wss://dutepjyocxcvecmsrtfp.supabase.co; media-src 'self' blob:; img-src 'self' data:; font-src 'self'; object-src 'none'; base-uri 'self'; frame-ancestors 'none'; form-action 'self'; upgrade-insecure-requests" }, { "key": "Strict-Transport-Security",