iobroker.assistant 1.4.5 → 1.4.7

package/build/lib/voice/ffmpegStatic.js

@@ -0,0 +1,137 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.FFMPEG_STATIC_RELEASE = void 0;
+exports.ffmpegStaticAsset = ffmpegStaticAsset;
+exports.ffmpegStaticPath = ffmpegStaticPath;
+exports.downloadFfmpegStatic = downloadFfmpegStatic;
+/**
+ * A static ffmpeg for hosts without one: the binaries of ffmpeg-static (github.com/eugeneware/ffmpeg-static),
+ * downloaded on request ("Download ffmpeg" on the Voice tab) into the instance data dir.
+ *
+ * Deliberately not the npm package: its install script downloads 20-30 MB for every installation of the adapter,
+ * also where nobody uses voice, and newer npm versions no longer run install scripts without asking - then the
+ * binary is silently missing. The release and the SHA-256 of every file are pinned here, so a replaced file is
+ * refused instead of executed.
+ */
+const node_crypto_1 = require("node:crypto");
+const fs = __importStar(require("node:fs"));
+const path = __importStar(require("node:path"));
+const node_stream_1 = require("node:stream");
+const promises_1 = require("node:stream/promises");
+const node_zlib_1 = require("node:zlib");
+/** The ffmpeg-static release. The binaries are newer than its name (Linux arm64: ffmpeg 7.0.2). */
+exports.FFMPEG_STATIC_RELEASE = 'b6.1.1';
+const BASE_URL = `https://github.com/eugeneware/ffmpeg-static/releases/download/${exports.FFMPEG_STATIC_RELEASE}`;
+/** SHA-256 of the gzipped binaries, as GitHub reports them for the release. Key = `<platform>-<arch>` of Node. */
+const ASSETS = {
+    'darwin-arm64': '8923876afa8db5585022d7860ec7e589af192f441c56793971276d450ed3bbfa',
+    'darwin-x64': '929b375c1182d956c51f7ac25e0b2b0411fb01f6f407aa15c9758efeb4242106',
+    'linux-arm': '64b115a12f0ab77c277e3c418aae8b40ef881e75e746a0e2d066a206b9bc5172',
+    'linux-arm64': '754a678672298bc68156adff58aa7385a592c2b30b1d0ae8750c45c915c4bac0',
+    'linux-ia32': '169b27c078a8ecedb814cac67afccf15a9868d63e9d74ef86088adefaa500d00',
+    'linux-x64': 'bfe8a8fc511530457b528c48d77b5737527b504a3797a9bc4866aeca69c2dffa',
+    'win32-x64': '8883a3dffbd0a16cf4ef95206ea05283f78908dbfb118f73c83f4951dcc06d77',
+};
+/**
+ * The file for this machine, or `null` if ffmpeg-static has none (e.g. Windows on ARM, FreeBSD).
+ *
+ * @param platform Node's `process.platform`
+ * @param arch Node's `process.arch`
+ */
+function ffmpegStaticAsset(platform = process.platform, arch = process.arch) {
+    const key = `${platform}-${arch}`;
+    return ASSETS[key] ? { name: `ffmpeg-${key}.gz`, sha256: ASSETS[key] } : null;
+}
+/**
+ * Where the downloaded binary lives.
+ *
+ * @param dataDir the instance data dir
+ * @param platform Node's `process.platform`
+ */
+function ffmpegStaticPath(dataDir, platform = process.platform) {
+    return path.join(dataDir, 'ffmpeg', platform === 'win32' ? 'ffmpeg.exe' : 'ffmpeg');
+}
+/**
+ * Download, check and unpack the static ffmpeg into `<dataDir>/ffmpeg/`. Replaces an older download.
+ *
+ * @param dataDir the instance data dir
+ * @param options logger; the rest is for tests
+ * @returns the path of the binary
+ */
+async function downloadFfmpegStatic(dataDir, options = {}) {
+    const platform = options.platform ?? process.platform;
+    const asset = options.asset === undefined ? ffmpegStaticAsset(platform) : options.asset;
+    if (!asset) {
+        throw new Error(`there is no ffmpeg download for ${process.platform}-${process.arch} - install ffmpeg yourself and enter its path`);
+    }
+    const url = `${options.baseUrl ?? BASE_URL}/${asset.name}`;
+    const dest = ffmpegStaticPath(dataDir, platform);
+    const tmp = `${dest}.part`;
+    fs.mkdirSync(path.dirname(dest), { recursive: true });
+    const res = await fetch(url, { signal: AbortSignal.timeout(options.timeoutMs ?? 10 * 60_000) });
+    if (!res.ok || !res.body) {
+        throw new Error(`download of ${url} failed: HTTP ${res.status}`);
+    }
+    const size = Number(res.headers.get('content-length')) || 0;
+    options.log?.info(`Downloading ffmpeg (${asset.name}${size ? `, ${(size / 1e6).toFixed(1)} MB` : ''}) …`);
+    // The hash is taken over the gzipped bytes - that is what GitHub publishes.
+    const hash = (0, node_crypto_1.createHash)('sha256');
+    const tap = new node_stream_1.Transform({
+        transform(chunk, _encoding, done) {
+            hash.update(chunk);
+            done(null, chunk);
+        },
+    });
+    try {
+        await (0, promises_1.pipeline)(node_stream_1.Readable.fromWeb(res.body), tap, (0, node_zlib_1.createGunzip)(), fs.createWriteStream(tmp, { mode: 0o755 }));
+        const actual = hash.digest('hex');
+        if (actual !== asset.sha256) {
+            throw new Error(`${asset.name} has the wrong checksum (${actual}) - not used`);
+        }
+        if (platform !== 'win32') {
+            fs.chmodSync(tmp, 0o755);
+        }
+        fs.rmSync(dest, { force: true });
+        fs.renameSync(tmp, dest);
+    }
+    catch (e) {
+        fs.rmSync(tmp, { force: true });
+        throw e;
+    }
+    options.log?.info(`ffmpeg saved to ${dest}`);
+    return dest;
+}
+//# sourceMappingURL=ffmpegStatic.js.map

package/build/lib/voice/ffmpegStatic.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ffmpegStatic.js","sourceRoot":"","sources":["../../../src/lib/voice/ffmpegStatic.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA+CA,8CAMC;AAQD,4CAEC;AAoBD,oDAkDC;AArID;;;;;;;;GAQG;AACH,6CAAyC;AACzC,4CAA8B;AAC9B,gDAAkC;AAClC,6CAAkD;AAElD,mDAAgD;AAChD,yCAAyC;AAIzC,mGAAmG;AACtF,QAAA,qBAAqB,GAAG,QAAQ,CAAC;AAE9C,MAAM,QAAQ,GAAG,iEAAiE,6BAAqB,EAAE,CAAC;AAE1G,kHAAkH;AAClH,MAAM,MAAM,GAA2B;IACnC,cAAc,EAAE,kEAAkE;IAClF,YAAY,EAAE,kEAAkE;IAChF,WAAW,EAAE,kEAAkE;IAC/E,aAAa,EAAE,kEAAkE;IACjF,YAAY,EAAE,kEAAkE;IAChF,WAAW,EAAE,kEAAkE;IAC/E,WAAW,EAAE,kEAAkE;CAClF,CAAC;AAQF;;;;;GAKG;AACH,SAAgB,iBAAiB,CAC7B,WAAmB,OAAO,CAAC,QAAQ,EACnC,OAAe,OAAO,CAAC,IAAI;IAE3B,MAAM,GAAG,GAAG,GAAG,QAAQ,IAAI,IAAI,EAAE,CAAC;IAClC,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,GAAG,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;AAClF,CAAC;AAED;;;;;GAKG;AACH,SAAgB,gBAAgB,CAAC,OAAe,EAAE,WAAmB,OAAO,CAAC,QAAQ;IACjF,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;AACxF,CAAC;AAaD;;;;;;GAMG;AACI,KAAK,UAAU,oBAAoB,CAAC,OAAe,EAAE,UAAiC,EAAE;IAC3F,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,QAAQ,CAAC;IACtD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC;IACxF,IAAI,CAAC,KAAK,EAAE,CAAC;QACT,MAAM,IAAI,KAAK,CACX,mCAAmC,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,IAAI,+CAA+C,CACrH,CAAC;IACN,CAAC;IACD,MAAM,GAAG,GAAG,GAAG,OAAO,CAAC,OAAO,IAAI,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;IAC3D,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IACjD,MAAM,GAAG,GAAG,GAAG,IAAI,OAAO,CAAC;IAC3B,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAEtD,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,IAAI,EAAE,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC;IAChG,IAAI,CAAC,GAAG,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CAAC,eAAe,GAAG,iBAAiB,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IACrE,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC;IAC5D,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,uBAAuB,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IAE1G,4EAA4E;IAC5E,MAAM,IAAI,GAAG,IAAA,wBAAU,EAAC,QAAQ,CAAC,CAAC;IAClC,MAAM,GAAG,GAAG,IAAI,uBAAS,CAAC;QACtB,SAAS,CAAC,KAAa,EAAE,SAAS,EAAE,IAAI;YACpC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACnB,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACtB,CAAC;KACJ,CAAC,CAAC;IACH,IAAI,CAAC;QACD,MAAM,IAAA,mBAAQ,EACV,sBAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,IAAyB,CAAC,EAC/C,GAAG,EACH,IAAA,wBAAY,GAAE,EACd,EAAE,CAAC,iBAAiB,CAAC,GAAG,EAAE,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAC7C,CAAC;QACF,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAClC,IAAI,MAAM,KAAK,KAAK,CAAC,MAAM,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,CAAC,IAAI,4BAA4B,MAAM,cAAc,CAAC,CAAC;QACnF,CAAC;QACD,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;YACvB,EAAE,CAAC,SAAS,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QAC7B,CAAC;QACD,EAAE,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACjC,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;IAC7B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,EAAE,CAAC,MAAM,CAAC,GAAG,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAChC,MAAM,CAAC,CAAC;IACZ,CAAC;IACD,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,mBAAmB,IAAI,EAAE,CAAC,CAAC;IAC7C,OAAO,IAAI,CAAC;AAChB,CAAC","sourcesContent":["/**\n * A static ffmpeg for hosts without one: the binaries of ffmpeg-static (github.com/eugeneware/ffmpeg-static),\n * downloaded on request (\"Download ffmpeg\" on the Voice tab) into the instance data dir.\n *\n * Deliberately not the npm package: its install script downloads 20-30 MB for every installation of the adapter,\n * also where nobody uses voice, and newer npm versions no longer run install scripts without asking - then the\n * binary is silently missing. The release and the SHA-256 of every file are pinned here, so a replaced file is\n * refused instead of executed.\n */\nimport { createHash } from 'node:crypto';\nimport * as fs from 'node:fs';\nimport * as path from 'node:path';\nimport { Readable, Transform } from 'node:stream';\nimport type { ReadableStream as WebReadableStream } from 'node:stream/web';\nimport { pipeline } from 'node:stream/promises';\nimport { createGunzip } from 'node:zlib';\n\nimport type { VoiceLogger } from './download';\n\n/** The ffmpeg-static release. The binaries are newer than its name (Linux arm64: ffmpeg 7.0.2). */\nexport const FFMPEG_STATIC_RELEASE = 'b6.1.1';\n\nconst BASE_URL = `https://github.com/eugeneware/ffmpeg-static/releases/download/${FFMPEG_STATIC_RELEASE}`;\n\n/** SHA-256 of the gzipped binaries, as GitHub reports them for the release. Key = `<platform>-<arch>` of Node. */\nconst ASSETS: Record<string, string> = {\n    'darwin-arm64': '8923876afa8db5585022d7860ec7e589af192f441c56793971276d450ed3bbfa',\n    'darwin-x64': '929b375c1182d956c51f7ac25e0b2b0411fb01f6f407aa15c9758efeb4242106',\n    'linux-arm': '64b115a12f0ab77c277e3c418aae8b40ef881e75e746a0e2d066a206b9bc5172',\n    'linux-arm64': '754a678672298bc68156adff58aa7385a592c2b30b1d0ae8750c45c915c4bac0',\n    'linux-ia32': '169b27c078a8ecedb814cac67afccf15a9868d63e9d74ef86088adefaa500d00',\n    'linux-x64': 'bfe8a8fc511530457b528c48d77b5737527b504a3797a9bc4866aeca69c2dffa',\n    'win32-x64': '8883a3dffbd0a16cf4ef95206ea05283f78908dbfb118f73c83f4951dcc06d77',\n};\n\n/** A file of the release: its name and the SHA-256 of the gzipped content. */\nexport interface FfmpegAsset {\n    name: string;\n    sha256: string;\n}\n\n/**\n * The file for this machine, or `null` if ffmpeg-static has none (e.g. Windows on ARM, FreeBSD).\n *\n * @param platform Node's `process.platform`\n * @param arch Node's `process.arch`\n */\nexport function ffmpegStaticAsset(\n    platform: string = process.platform,\n    arch: string = process.arch,\n): FfmpegAsset | null {\n    const key = `${platform}-${arch}`;\n    return ASSETS[key] ? { name: `ffmpeg-${key}.gz`, sha256: ASSETS[key] } : null;\n}\n\n/**\n * Where the downloaded binary lives.\n *\n * @param dataDir the instance data dir\n * @param platform Node's `process.platform`\n */\nexport function ffmpegStaticPath(dataDir: string, platform: string = process.platform): string {\n    return path.join(dataDir, 'ffmpeg', platform === 'win32' ? 'ffmpeg.exe' : 'ffmpeg');\n}\n\n/** Options of {@link downloadFfmpegStatic} - besides the logger only for tests. */\nexport interface FfmpegDownloadOptions {\n    log?: VoiceLogger;\n    /** Release URL without the file name */\n    baseUrl?: string;\n    /** Instead of the file for this machine */\n    asset?: FfmpegAsset | null;\n    platform?: string;\n    timeoutMs?: number;\n}\n\n/**\n * Download, check and unpack the static ffmpeg into `<dataDir>/ffmpeg/`. Replaces an older download.\n *\n * @param dataDir the instance data dir\n * @param options logger; the rest is for tests\n * @returns the path of the binary\n */\nexport async function downloadFfmpegStatic(dataDir: string, options: FfmpegDownloadOptions = {}): Promise<string> {\n    const platform = options.platform ?? process.platform;\n    const asset = options.asset === undefined ? ffmpegStaticAsset(platform) : options.asset;\n    if (!asset) {\n        throw new Error(\n            `there is no ffmpeg download for ${process.platform}-${process.arch} - install ffmpeg yourself and enter its path`,\n        );\n    }\n    const url = `${options.baseUrl ?? BASE_URL}/${asset.name}`;\n    const dest = ffmpegStaticPath(dataDir, platform);\n    const tmp = `${dest}.part`;\n    fs.mkdirSync(path.dirname(dest), { recursive: true });\n\n    const res = await fetch(url, { signal: AbortSignal.timeout(options.timeoutMs ?? 10 * 60_000) });\n    if (!res.ok || !res.body) {\n        throw new Error(`download of ${url} failed: HTTP ${res.status}`);\n    }\n    const size = Number(res.headers.get('content-length')) || 0;\n    options.log?.info(`Downloading ffmpeg (${asset.name}${size ? `, ${(size / 1e6).toFixed(1)} MB` : ''}) …`);\n\n    // The hash is taken over the gzipped bytes - that is what GitHub publishes.\n    const hash = createHash('sha256');\n    const tap = new Transform({\n        transform(chunk: Buffer, _encoding, done): void {\n            hash.update(chunk);\n            done(null, chunk);\n        },\n    });\n    try {\n        await pipeline(\n            Readable.fromWeb(res.body as WebReadableStream),\n            tap,\n            createGunzip(),\n            fs.createWriteStream(tmp, { mode: 0o755 }),\n        );\n        const actual = hash.digest('hex');\n        if (actual !== asset.sha256) {\n            throw new Error(`${asset.name} has the wrong checksum (${actual}) - not used`);\n        }\n        if (platform !== 'win32') {\n            fs.chmodSync(tmp, 0o755);\n        }\n        fs.rmSync(dest, { force: true });\n        fs.renameSync(tmp, dest);\n    } catch (e) {\n        fs.rmSync(tmp, { force: true });\n        throw e;\n    }\n    options.log?.info(`ffmpeg saved to ${dest}`);\n    return dest;\n}\n"]}

package/build/lib/voice/stt.js

@@ -36,6 +36,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.OpenAiStt = void 0;
 exports.hintsToPrompt = hintsToPrompt;
 exports.pcmToWav = pcmToWav;
+exports.normalizeSttBaseUrl = normalizeSttBaseUrl;
 /**
  * Speech-to-text engines. V1 ships the cloud engine (OpenAI Whisper); local engines
  * (Vosk / whisper.cpp / Wyoming) plug in behind the same `SttEngine` interface later.
@@ -88,12 +89,43 @@ function pcmToWav(pcm, sampleRate, numChannels = 1) {
     header.writeUInt32LE(pcm.length, 40);
     return Buffer.concat([header, pcm]);
 }
-/** Cloud STT via the OpenAI transcription API (Whisper). */
+/**
+ * Base URL of an OpenAI-compatible transcription server as the SDK wants it. Practically every such server
+ * (speaches/faster-whisper-server, LocalAI, vLLM, Groq) serves `/v1/audio/transcriptions`, so an address typed
+ * without a path gets `/v1`; one with a path is taken as it is.
+ */
+function normalizeSttBaseUrl(raw) {
+    const url = raw.trim().replace(/\/+$/, '');
+    if (!url) {
+        return '';
+    }
+    try {
+        const parsed = new URL(url);
+        return parsed.pathname === '/' || parsed.pathname === '' ? `${url}/v1` : url;
+    }
+    catch {
+        throw new Error(`speech recognition base URL "${raw}" is not a URL (e.g. http://192.168.1.10:8000/v1)`);
+    }
+}
+/**
+ * STT via the OpenAI transcription API (Whisper) — or any server that speaks it, e.g. a local
+ * whisper-large-v3-turbo next to ioBroker (`baseUrl`).
+ */
 class OpenAiStt {
     client;
     model;
-    constructor(apiKey, model = 'whisper-1') {
-        this.client = new openai_1.default({ apiKey });
+    /**
+     * @param apiKey OpenAI key, or the key of the server at `baseUrl` ('' for a server without one)
+     * @param model model id; on an own server as that server names it
+     * @param baseUrl OpenAI-compatible server instead of api.openai.com ('' = OpenAI)
+     */
+    constructor(apiKey, model = 'whisper-1', baseUrl = '') {
+        const baseURL = normalizeSttBaseUrl(baseUrl);
+        this.client = baseURL
+            ? // The SDK refuses an empty key; a local server ignores it. A server that hangs must not hold the
+                // satellite for the SDK's default ten minutes.
+                new openai_1.default({ apiKey: apiKey || 'none', baseURL, timeout: 60_000, maxRetries: 1 })
+            : new openai_1.default({ apiKey });
         this.model = model;
     }
     async transcribe(pcm, sampleRate, lang, hints) {

package/build/lib/voice/stt.js.map

@@ -1 +1 @@
-{"version":3,"file":"stt.js","sourceRoot":"","sources":["../../../src/lib/voice/stt.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwBA,sCAqBC;AAGD,4BAmBC;AAnED;;;GAGG;AACH,iDAAwC;AAgBxC;;;GAGG;AACH,SAAgB,aAAa,CAAC,KAA2B,EAAE,QAAQ,GAAG,GAAG;IACrE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC;QACjB,OAAO,EAAE,CAAC;IACd,CAAC;IACD,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACpB,MAAM,IAAI,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAC9B,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAC/B,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,SAAS;QACb,CAAC;QACD,IAAI,GAAG,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,QAAQ,EAAE,CAAC;YACnC,MAAM;QACV,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,GAAG,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC5B,CAAC;AAED,wGAAwG;AACxG,SAAgB,QAAQ,CAAC,GAAW,EAAE,UAAkB,EAAE,WAAW,GAAG,CAAC;IACrE,MAAM,aAAa,GAAG,EAAE,CAAC;IACzB,MAAM,UAAU,GAAG,CAAC,WAAW,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;IACrD,MAAM,QAAQ,GAAG,UAAU,GAAG,UAAU,CAAC;IACzC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAChC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACxB,MAAM,CAAC,aAAa,CAAC,EAAE,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACzC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACxB,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACzB,MAAM,CAAC,aAAa,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,qBAAqB;IACnD,MAAM,CAAC,aAAa,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,qBAAqB;IAClD,MAAM,CAAC,aAAa,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IACtC,MAAM,CAAC,aAAa,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACrC,MAAM,CAAC,aAAa,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACnC,MAAM,CAAC,aAAa,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACrC,MAAM,CAAC,aAAa,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC;IACxC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACzB,MAAM,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACrC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC;AACxC,CAAC;AAED,4DAA4D;AAC5D,MAAa,SAAS;IACD,MAAM,CAAS;IACf,KAAK,CAAS;IAE/B,YAAY,MAAc,EAAE,KAAK,GAAG,WAAW;QAC3C,IAAI,CAAC,MAAM,GAAG,IAAI,gBAAM,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;QACrC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;IACvB,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,GAAW,EAAE,UAAkB,EAAE,IAAY,EAAE,KAAgB;QAC5E,MAAM,IAAI,GAAG,MAAM,IAAA,eAAM,EAAC,QAAQ,CAAC,GAAG,EAAE,UAAU,CAAC,EAAE,WAAW,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;QACzF,mGAAmG;QACnG,kFAAkF;QAClF,MAAM,MAAM,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;QACpC,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC;YACtD,IAAI;YACJ,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,0FAA0F;YAC1F,QAAQ,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,SAAS;YACjD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChC,CAAC,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACnC,CAAC;CACJ;AAvBD,8BAuBC","sourcesContent":["/**\n * Speech-to-text engines. V1 ships the cloud engine (OpenAI Whisper); local engines\n * (Vosk / whisper.cpp / Wyoming) plug in behind the same `SttEngine` interface later.\n */\nimport OpenAI, { toFile } from 'openai';\n\nexport interface SttEngine {\n    /**\n     * Transcribe raw mono 16-bit PCM. `lang` is an ISO-639-1 hint ('' = auto-detect).\n     *\n     * `hints` is an optional domain vocabulary (device / room / scene names …) to bias recognition toward,\n     * so the engine better transcribes those proper nouns. Only engines that support *soft* biasing use it\n     * (OpenAI Whisper `prompt`, Azure phrase list); engines where a word list would be a hard constraint or\n     * needs pre-registration (Vosk grammar, AWS custom vocabulary) ignore it.\n     */\n    transcribe(pcm: Buffer, sampleRate: number, lang: string, hints?: string[]): Promise<string>;\n    /** Optional warm-up (install/download engine + model for `lang`) so the first request isn't slow. */\n    prepare?(lang: string): Promise<void>;\n}\n\n/**\n * Join vocabulary hints into a Whisper `prompt` string (a soft bias toward these terms), de-duplicated and\n * length-bounded to stay within the model's ~224-token prompt budget. Returns '' when there are no hints.\n */\nexport function hintsToPrompt(hints: string[] | undefined, maxChars = 800): string {\n    if (!hints?.length) {\n        return '';\n    }\n    const seen = new Set<string>();\n    const parts: string[] = [];\n    let len = 0;\n    for (const h of hints) {\n        const term = (h || '').trim();\n        const key = term.toLowerCase();\n        if (!term || seen.has(key)) {\n            continue;\n        }\n        if (len + term.length + 2 > maxChars) {\n            break;\n        }\n        seen.add(key);\n        parts.push(term);\n        len += term.length + 2;\n    }\n    return parts.join(', ');\n}\n\n/** Wrap raw 16-bit-signed-LE PCM (mono, or interleaved stereo) into a minimal 44-byte WAV container. */\nexport function pcmToWav(pcm: Buffer, sampleRate: number, numChannels = 1): Buffer {\n    const bitsPerSample = 16;\n    const blockAlign = (numChannels * bitsPerSample) / 8;\n    const byteRate = sampleRate * blockAlign;\n    const header = Buffer.alloc(44);\n    header.write('RIFF', 0);\n    header.writeUInt32LE(36 + pcm.length, 4);\n    header.write('WAVE', 8);\n    header.write('fmt ', 12);\n    header.writeUInt32LE(16, 16); // PCM fmt chunk size\n    header.writeUInt16LE(1, 20); // audio format = PCM\n    header.writeUInt16LE(numChannels, 22);\n    header.writeUInt32LE(sampleRate, 24);\n    header.writeUInt32LE(byteRate, 28);\n    header.writeUInt16LE(blockAlign, 32);\n    header.writeUInt16LE(bitsPerSample, 34);\n    header.write('data', 36);\n    header.writeUInt32LE(pcm.length, 40);\n    return Buffer.concat([header, pcm]);\n}\n\n/** Cloud STT via the OpenAI transcription API (Whisper). */\nexport class OpenAiStt implements SttEngine {\n    private readonly client: OpenAI;\n    private readonly model: string;\n\n    constructor(apiKey: string, model = 'whisper-1') {\n        this.client = new OpenAI({ apiKey });\n        this.model = model;\n    }\n\n    async transcribe(pcm: Buffer, sampleRate: number, lang: string, hints?: string[]): Promise<string> {\n        const file = await toFile(pcmToWav(pcm, sampleRate), 'audio.wav', { type: 'audio/wav' });\n        // A prompt biases Whisper toward the given domain vocabulary (device/room names) without excluding\n        // other words — improves recognition of proper nouns the NLU then needs to match.\n        const prompt = hintsToPrompt(hints);\n        const res = await this.client.audio.transcriptions.create({\n            file,\n            model: this.model,\n            // Whisper wants the bare ISO-639-1 code; strip any region (de-DE → de). '' → auto-detect.\n            language: (lang || '').split('-')[0] || undefined,\n            ...(prompt ? { prompt } : {}),\n        });\n        return (res.text || '').trim();\n    }\n}\n"]}
+{"version":3,"file":"stt.js","sourceRoot":"","sources":["../../../src/lib/voice/stt.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwBA,sCAqBC;AAGD,4BAmBC;AAOD,kDAWC;AArFD;;;GAGG;AACH,iDAAwC;AAgBxC;;;GAGG;AACH,SAAgB,aAAa,CAAC,KAA2B,EAAE,QAAQ,GAAG,GAAG;IACrE,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC;QACjB,OAAO,EAAE,CAAC;IACd,CAAC;IACD,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;IAC/B,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACpB,MAAM,IAAI,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAC9B,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QAC/B,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,SAAS;QACb,CAAC;QACD,IAAI,GAAG,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,QAAQ,EAAE,CAAC;YACnC,MAAM;QACV,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,GAAG,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC5B,CAAC;AAED,wGAAwG;AACxG,SAAgB,QAAQ,CAAC,GAAW,EAAE,UAAkB,EAAE,WAAW,GAAG,CAAC;IACrE,MAAM,aAAa,GAAG,EAAE,CAAC;IACzB,MAAM,UAAU,GAAG,CAAC,WAAW,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;IACrD,MAAM,QAAQ,GAAG,UAAU,GAAG,UAAU,CAAC;IACzC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAChC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACxB,MAAM,CAAC,aAAa,CAAC,EAAE,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACzC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACxB,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACzB,MAAM,CAAC,aAAa,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,qBAAqB;IACnD,MAAM,CAAC,aAAa,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,qBAAqB;IAClD,MAAM,CAAC,aAAa,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;IACtC,MAAM,CAAC,aAAa,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACrC,MAAM,CAAC,aAAa,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;IACnC,MAAM,CAAC,aAAa,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IACrC,MAAM,CAAC,aAAa,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC;IACxC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACzB,MAAM,CAAC,aAAa,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACrC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC;AACxC,CAAC;AAED;;;;GAIG;AACH,SAAgB,mBAAmB,CAAC,GAAW;IAC3C,MAAM,GAAG,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC3C,IAAI,CAAC,GAAG,EAAE,CAAC;QACP,OAAO,EAAE,CAAC;IACd,CAAC;IACD,IAAI,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;QAC5B,OAAO,MAAM,CAAC,QAAQ,KAAK,GAAG,IAAI,MAAM,CAAC,QAAQ,KAAK,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;IACjF,CAAC;IAAC,MAAM,CAAC;QACL,MAAM,IAAI,KAAK,CAAC,gCAAgC,GAAG,mDAAmD,CAAC,CAAC;IAC5G,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,MAAa,SAAS;IACD,MAAM,CAAS;IACf,KAAK,CAAS;IAE/B;;;;OAIG;IACH,YAAY,MAAc,EAAE,KAAK,GAAG,WAAW,EAAE,OAAO,GAAG,EAAE;QACzD,MAAM,OAAO,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;QAC7C,IAAI,CAAC,MAAM,GAAG,OAAO;YACjB,CAAC,CAAC,iGAAiG;gBACjG,+CAA+C;gBAC/C,IAAI,gBAAM,CAAC,EAAE,MAAM,EAAE,MAAM,IAAI,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC,EAAE,CAAC;YACnF,CAAC,CAAC,IAAI,gBAAM,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;QAC7B,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;IACvB,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,GAAW,EAAE,UAAkB,EAAE,IAAY,EAAE,KAAgB;QAC5E,MAAM,IAAI,GAAG,MAAM,IAAA,eAAM,EAAC,QAAQ,CAAC,GAAG,EAAE,UAAU,CAAC,EAAE,WAAW,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;QACzF,mGAAmG;QACnG,kFAAkF;QAClF,MAAM,MAAM,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC;QACpC,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC;YACtD,IAAI;YACJ,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,0FAA0F;YAC1F,QAAQ,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,SAAS;YACjD,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChC,CAAC,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACnC,CAAC;CACJ;AAjCD,8BAiCC","sourcesContent":["/**\n * Speech-to-text engines. V1 ships the cloud engine (OpenAI Whisper); local engines\n * (Vosk / whisper.cpp / Wyoming) plug in behind the same `SttEngine` interface later.\n */\nimport OpenAI, { toFile } from 'openai';\n\nexport interface SttEngine {\n    /**\n     * Transcribe raw mono 16-bit PCM. `lang` is an ISO-639-1 hint ('' = auto-detect).\n     *\n     * `hints` is an optional domain vocabulary (device / room / scene names …) to bias recognition toward,\n     * so the engine better transcribes those proper nouns. Only engines that support *soft* biasing use it\n     * (OpenAI Whisper `prompt`, Azure phrase list); engines where a word list would be a hard constraint or\n     * needs pre-registration (Vosk grammar, AWS custom vocabulary) ignore it.\n     */\n    transcribe(pcm: Buffer, sampleRate: number, lang: string, hints?: string[]): Promise<string>;\n    /** Optional warm-up (install/download engine + model for `lang`) so the first request isn't slow. */\n    prepare?(lang: string): Promise<void>;\n}\n\n/**\n * Join vocabulary hints into a Whisper `prompt` string (a soft bias toward these terms), de-duplicated and\n * length-bounded to stay within the model's ~224-token prompt budget. Returns '' when there are no hints.\n */\nexport function hintsToPrompt(hints: string[] | undefined, maxChars = 800): string {\n    if (!hints?.length) {\n        return '';\n    }\n    const seen = new Set<string>();\n    const parts: string[] = [];\n    let len = 0;\n    for (const h of hints) {\n        const term = (h || '').trim();\n        const key = term.toLowerCase();\n        if (!term || seen.has(key)) {\n            continue;\n        }\n        if (len + term.length + 2 > maxChars) {\n            break;\n        }\n        seen.add(key);\n        parts.push(term);\n        len += term.length + 2;\n    }\n    return parts.join(', ');\n}\n\n/** Wrap raw 16-bit-signed-LE PCM (mono, or interleaved stereo) into a minimal 44-byte WAV container. */\nexport function pcmToWav(pcm: Buffer, sampleRate: number, numChannels = 1): Buffer {\n    const bitsPerSample = 16;\n    const blockAlign = (numChannels * bitsPerSample) / 8;\n    const byteRate = sampleRate * blockAlign;\n    const header = Buffer.alloc(44);\n    header.write('RIFF', 0);\n    header.writeUInt32LE(36 + pcm.length, 4);\n    header.write('WAVE', 8);\n    header.write('fmt ', 12);\n    header.writeUInt32LE(16, 16); // PCM fmt chunk size\n    header.writeUInt16LE(1, 20); // audio format = PCM\n    header.writeUInt16LE(numChannels, 22);\n    header.writeUInt32LE(sampleRate, 24);\n    header.writeUInt32LE(byteRate, 28);\n    header.writeUInt16LE(blockAlign, 32);\n    header.writeUInt16LE(bitsPerSample, 34);\n    header.write('data', 36);\n    header.writeUInt32LE(pcm.length, 40);\n    return Buffer.concat([header, pcm]);\n}\n\n/**\n * Base URL of an OpenAI-compatible transcription server as the SDK wants it. Practically every such server\n * (speaches/faster-whisper-server, LocalAI, vLLM, Groq) serves `/v1/audio/transcriptions`, so an address typed\n * without a path gets `/v1`; one with a path is taken as it is.\n */\nexport function normalizeSttBaseUrl(raw: string): string {\n    const url = raw.trim().replace(/\\/+$/, '');\n    if (!url) {\n        return '';\n    }\n    try {\n        const parsed = new URL(url);\n        return parsed.pathname === '/' || parsed.pathname === '' ? `${url}/v1` : url;\n    } catch {\n        throw new Error(`speech recognition base URL \"${raw}\" is not a URL (e.g. http://192.168.1.10:8000/v1)`);\n    }\n}\n\n/**\n * STT via the OpenAI transcription API (Whisper) — or any server that speaks it, e.g. a local\n * whisper-large-v3-turbo next to ioBroker (`baseUrl`).\n */\nexport class OpenAiStt implements SttEngine {\n    private readonly client: OpenAI;\n    private readonly model: string;\n\n    /**\n     * @param apiKey OpenAI key, or the key of the server at `baseUrl` ('' for a server without one)\n     * @param model model id; on an own server as that server names it\n     * @param baseUrl OpenAI-compatible server instead of api.openai.com ('' = OpenAI)\n     */\n    constructor(apiKey: string, model = 'whisper-1', baseUrl = '') {\n        const baseURL = normalizeSttBaseUrl(baseUrl);\n        this.client = baseURL\n            ? // The SDK refuses an empty key; a local server ignores it. A server that hangs must not hold the\n              // satellite for the SDK's default ten minutes.\n              new OpenAI({ apiKey: apiKey || 'none', baseURL, timeout: 60_000, maxRetries: 1 })\n            : new OpenAI({ apiKey });\n        this.model = model;\n    }\n\n    async transcribe(pcm: Buffer, sampleRate: number, lang: string, hints?: string[]): Promise<string> {\n        const file = await toFile(pcmToWav(pcm, sampleRate), 'audio.wav', { type: 'audio/wav' });\n        // A prompt biases Whisper toward the given domain vocabulary (device/room names) without excluding\n        // other words — improves recognition of proper nouns the NLU then needs to match.\n        const prompt = hintsToPrompt(hints);\n        const res = await this.client.audio.transcriptions.create({\n            file,\n            model: this.model,\n            // Whisper wants the bare ISO-639-1 code; strip any region (de-DE → de). '' → auto-detect.\n            language: (lang || '').split('-')[0] || undefined,\n            ...(prompt ? { prompt } : {}),\n        });\n        return (res.text || '').trim();\n    }\n}\n"]}