iobroker.assistant 0.1.2

package/build/lib/voice/download.js

@@ -0,0 +1,76 @@
+"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.downloadFile = downloadFile;
+exports.extractArchive = extractArchive;
+/**
+ * Small download + extract helpers for the local engines (Piper binary/voices, Vosk models).
+ * Files land in the writable instance data dir so they survive adapter upgrades.
+ */
+const fs = __importStar(require("node:fs"));
+const path = __importStar(require("node:path"));
+const node_child_process_1 = require("node:child_process");
+/** Download a file to `dest` (skips if it already exists). */
+async function downloadFile(url, dest, log) {
+    if (fs.existsSync(dest)) {
+        return;
+    }
+    fs.mkdirSync(path.dirname(dest), { recursive: true });
+    log.info(`Downloading ${path.basename(dest)} …`);
+    const res = await fetch(url);
+    if (!res.ok) {
+        throw new Error(`download ${url} failed: HTTP ${res.status}`);
+    }
+    const buf = Buffer.from(await res.arrayBuffer());
+    // Write to a temp file first so an interrupted download does not leave a truncated "valid" file.
+    const tmp = `${dest}.part`;
+    fs.writeFileSync(tmp, buf);
+    fs.renameSync(tmp, dest);
+    log.info(`Saved ${path.basename(dest)} (${(buf.length / 1e6).toFixed(1)} MB).`);
+}
+/** Extract a `.tar.gz`/`.tgz` (via `tar`) or `.zip` (via `unzip`) into `destDir`. */
+function extractArchive(archive, destDir, log) {
+    fs.mkdirSync(destDir, { recursive: true });
+    const isZip = archive.endsWith('.zip');
+    const cmd = isZip ? 'unzip' : 'tar';
+    const args = isZip ? ['-o', archive, '-d', destDir] : ['-xzf', archive, '-C', destDir];
+    return new Promise((resolve, reject) => {
+        const child = (0, node_child_process_1.spawn)(cmd, args);
+        child.stderr?.on('data', d => log.debug(`${cmd}: ${String(d).trim()}`));
+        child.on('error', e => reject(new Error(`${cmd} failed: ${e.message} — is '${cmd}' installed?`)));
+        child.on('close', code => (code === 0 ? resolve() : reject(new Error(`${cmd} exited with code ${code}`))));
+    });
+}
+//# sourceMappingURL=download.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"download.js","sourceRoot":"","sources":["../../../src/lib/voice/download.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAWA,oCAgBC;AAGD,wCAWC;AAzCD;;;GAGG;AACH,4CAA8B;AAC9B,gDAAkC;AAClC,2DAA2C;AAI3C,8DAA8D;AACvD,KAAK,UAAU,YAAY,CAAC,GAAW,EAAE,IAAY,EAAE,GAAgB;IAC1E,IAAI,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QACtB,OAAO;IACX,CAAC;IACD,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtD,GAAG,CAAC,IAAI,CAAC,eAAe,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjD,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,CAAC;IAC7B,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACV,MAAM,IAAI,KAAK,CAAC,YAAY,GAAG,iBAAiB,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC;IACjD,iGAAiG;IACjG,MAAM,GAAG,GAAG,GAAG,IAAI,OAAO,CAAC;IAC3B,EAAE,CAAC,aAAa,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;IAC3B,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;IACzB,GAAG,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;AACpF,CAAC;AAED,qFAAqF;AACrF,SAAgB,cAAc,CAAC,OAAe,EAAE,OAAe,EAAE,GAAgB;IAC7E,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC3C,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IACvC,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC;IACpC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;IACvF,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACnC,MAAM,KAAK,GAAG,IAAA,0BAAK,EAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAC/B,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,GAAG,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;QACxE,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,GAAG,YAAY,CAAC,CAAC,OAAO,UAAU,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC;QAClG,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,GAAG,qBAAqB,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/G,CAAC,CAAC,CAAC;AACP,CAAC","sourcesContent":["/**\n * Small download + extract helpers for the local engines (Piper binary/voices, Vosk models).\n * Files land in the writable instance data dir so they survive adapter upgrades.\n */\nimport * as fs from 'node:fs';\nimport * as path from 'node:path';\nimport { spawn } from 'node:child_process';\n\nexport type VoiceLogger = Pick<ioBroker.Logger, 'info' | 'warn' | 'debug' | 'error'>;\n\n/** Download a file to `dest` (skips if it already exists). */\nexport async function downloadFile(url: string, dest: string, log: VoiceLogger): Promise<void> {\n    if (fs.existsSync(dest)) {\n        return;\n    }\n    fs.mkdirSync(path.dirname(dest), { recursive: true });\n    log.info(`Downloading ${path.basename(dest)} …`);\n    const res = await fetch(url);\n    if (!res.ok) {\n        throw new Error(`download ${url} failed: HTTP ${res.status}`);\n    }\n    const buf = Buffer.from(await res.arrayBuffer());\n    // Write to a temp file first so an interrupted download does not leave a truncated \"valid\" file.\n    const tmp = `${dest}.part`;\n    fs.writeFileSync(tmp, buf);\n    fs.renameSync(tmp, dest);\n    log.info(`Saved ${path.basename(dest)} (${(buf.length / 1e6).toFixed(1)} MB).`);\n}\n\n/** Extract a `.tar.gz`/`.tgz` (via `tar`) or `.zip` (via `unzip`) into `destDir`. */\nexport function extractArchive(archive: string, destDir: string, log: VoiceLogger): Promise<void> {\n    fs.mkdirSync(destDir, { recursive: true });\n    const isZip = archive.endsWith('.zip');\n    const cmd = isZip ? 'unzip' : 'tar';\n    const args = isZip ? ['-o', archive, '-d', destDir] : ['-xzf', archive, '-C', destDir];\n    return new Promise((resolve, reject) => {\n        const child = spawn(cmd, args);\n        child.stderr?.on('data', d => log.debug(`${cmd}: ${String(d).trim()}`));\n        child.on('error', e => reject(new Error(`${cmd} failed: ${e.message} — is '${cmd}' installed?`)));\n        child.on('close', code => (code === 0 ? resolve() : reject(new Error(`${cmd} exited with code ${code}`))));\n    });\n}\n"]}

package/build/lib/voice/engines.js

@@ -0,0 +1,89 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createSttEngine = createSttEngine;
+exports.createTtsEngine = createTtsEngine;
+exports.listVoices = listVoices;
+exports.listSttModels = listSttModels;
+/**
+ * Factory that builds the configured STT/TTS engine. Cloud engines (OpenAI/Azure/AWS) use resolved
+ * credentials; local engines (Vosk/Piper) download their models on demand into the data dir.
+ * Adding a provider = one new engine class + one `case` here.
+ */
+const stt_1 = require("./stt");
+const tts_1 = require("./tts");
+const azure_1 = require("./azure");
+const aws_1 = require("./aws");
+const vosk_1 = require("./vosk");
+const piper_1 = require("./piper");
+const lang_1 = require("./lang");
+function requireAzure(creds) {
+    if (!creds.azureKey || !creds.azureRegion) {
+        throw new Error('Azure speech needs a subscription key and region');
+    }
+}
+function requireAws(creds) {
+    if (!creds.aws.accessKeyId || !creds.aws.secretAccessKey || !creds.aws.region) {
+        throw new Error('AWS speech needs an access key id, secret access key and region');
+    }
+}
+function requireOpenAi(creds) {
+    if (!creds.openaiKey) {
+        throw new Error('OpenAI speech needs an API key');
+    }
+}
+function createSttEngine(provider, ctx) {
+    switch (provider) {
+        case 'azure':
+            requireAzure(ctx.creds);
+            return new azure_1.AzureStt(ctx.creds.azureKey, ctx.creds.azureRegion);
+        case 'aws':
+            requireAws(ctx.creds);
+            return new aws_1.AwsStt(ctx.creds.aws);
+        case 'vosk':
+            return new vosk_1.VoskStt(ctx.dataDir, ctx.voskModel, ctx.log);
+        case 'piper':
+            throw new Error('Piper is a text-to-speech engine, not speech-to-text');
+        default:
+            requireOpenAi(ctx.creds);
+            return new stt_1.OpenAiStt(ctx.creds.openaiKey, ctx.sttModel || undefined);
+    }
+}
+function createTtsEngine(provider, ctx) {
+    switch (provider) {
+        case 'azure':
+            requireAzure(ctx.creds);
+            return new azure_1.AzureTts(ctx.creds.azureKey, ctx.creds.azureRegion, ctx.voices.azure);
+        case 'aws':
+            requireAws(ctx.creds);
+            return new aws_1.AwsTts(ctx.creds.aws, ctx.voices.aws);
+        case 'piper':
+            return new piper_1.PiperTts(ctx.dataDir, ctx.voices.piper, ctx.log);
+        case 'vosk':
+            throw new Error('Vosk is a speech-to-text engine, not text-to-speech');
+        default:
+            requireOpenAi(ctx.creds);
+            return new tts_1.OpenAiTts(ctx.creds.openaiKey, (ctx.voices.openai || 'alloy'), ctx.ttsModel || undefined);
+    }
+}
+/** OpenAI TTS voices are a fixed set (no list API). */
+const OPENAI_VOICES = ['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'];
+/** List the available TTS voices for a provider (for the settings voice dropdown). */
+async function listVoices(provider, ctx, language) {
+    switch (provider) {
+        case 'azure':
+            requireAzure(ctx.creds);
+            return (0, azure_1.listAzureVoices)(ctx.creds.azureKey, ctx.creds.azureRegion, (0, lang_1.isoToLocale)(language));
+        case 'aws':
+            requireAws(ctx.creds);
+            return (0, aws_1.listPollyVoices)(ctx.creds.aws, (0, lang_1.isoToLocale)(language));
+        case 'piper':
+            return (0, piper_1.listPiperVoices)();
+        default:
+            return OPENAI_VOICES;
+    }
+}
+/** Suggested STT model names for the settings dropdown. Only Vosk has a local model catalogue. */
+function listSttModels(provider, language) {
+    return provider === 'vosk' ? (0, vosk_1.listVoskModels)(language) : [];
+}
+//# sourceMappingURL=engines.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"engines.js","sourceRoot":"","sources":["../../../src/lib/voice/engines.ts"],"names":[],"mappings":";;AAiEA,0CAgBC;AAED,0CAoBC;AAMD,gCAaC;AAGD,sCAEC;AA/HD;;;;GAIG;AACH,+BAAkD;AAClD,+BAAoE;AACpE,mCAA8D;AAC9D,+BAAuE;AACvE,iCAAiD;AACjD,mCAAoD;AACpD,iCAAqC;AAoCrC,SAAS,YAAY,CAAC,KAAuB;IACzC,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACxE,CAAC;AACL,CAAC;AAED,SAAS,UAAU,CAAC,KAAuB;IACvC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,eAAe,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;IACvF,CAAC;AACL,CAAC;AAED,SAAS,aAAa,CAAC,KAAuB;IAC1C,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACnB,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACtD,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnE,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACrC,KAAK,MAAM;YACP,OAAO,IAAI,cAAO,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC5D,KAAK,OAAO;YACR,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;QAC5E;YACI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS,CAAC,CAAC;IAC7E,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrF,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACrD,KAAK,OAAO;YACR,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAChE,KAAK,MAAM;YACP,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QAC3E;YACI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAChB,GAAG,CAAC,KAAK,CAAC,SAAS,EACnB,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,IAAI,OAAO,CAAgB,EAC7C,GAAG,CAAC,QAAQ,IAAI,SAAS,CAC5B,CAAC;IACV,CAAC;AACL,CAAC;AAED,uDAAuD;AACvD,MAAM,aAAa,GAAG,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;AAE5E,sFAAsF;AAC/E,KAAK,UAAU,UAAU,CAAC,QAAwB,EAAE,GAAkB,EAAE,QAAgB;IAC3F,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAA,uBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QAC7F,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAA,qBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QACjE,KAAK,OAAO;YACR,OAAO,IAAA,uBAAe,GAAE,CAAC;QAC7B;YACI,OAAO,aAAa,CAAC;IAC7B,CAAC;AACL,CAAC;AAED,kGAAkG;AAClG,SAAgB,aAAa,CAAC,QAAwB,EAAE,QAAgB;IACpE,OAAO,QAAQ,KAAK,MAAM,CAAC,CAAC,CAAC,IAAA,qBAAc,EAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAC/D,CAAC","sourcesContent":["/**\n * Factory that builds the configured STT/TTS engine. Cloud engines (OpenAI/Azure/AWS) use resolved\n * credentials; local engines (Vosk/Piper) download their models on demand into the data dir.\n * Adding a provider = one new engine class + one `case` here.\n */\nimport { OpenAiStt, type SttEngine } from './stt';\nimport { OpenAiTts, type OpenAiVoice, type TtsEngine } from './tts';\nimport { AzureStt, AzureTts, listAzureVoices } from './azure';\nimport { AwsStt, AwsTts, listPollyVoices, type AwsCreds } from './aws';\nimport { VoskStt, listVoskModels } from './vosk';\nimport { PiperTts, listPiperVoices } from './piper';\nimport { isoToLocale } from './lang';\nimport type { VoiceLogger } from './download';\n\nexport type SpeechProvider = 'openai' | 'azure' | 'aws' | 'vosk' | 'piper';\n\n/** All speech credentials, resolved from config (or the central credential store). */\nexport interface VoiceCredentials {\n    openaiKey: string;\n    azureKey: string;\n    azureRegion: string;\n    aws: AwsCreds;\n}\n\n/** Provider-specific voice ids (each provider names voices differently). */\nexport interface VoiceNames {\n    openai: string; // alloy/echo/…\n    azure: string; // e.g. de-DE-KatjaNeural\n    aws: string; // e.g. Vicki\n    piper: string; // e.g. de_DE-thorsten-medium\n}\n\n/** Everything the factory needs to build any engine. */\nexport interface EngineContext {\n    creds: VoiceCredentials;\n    voices: VoiceNames;\n    /** Writable instance data dir (for the local engines' downloads). */\n    dataDir: string;\n    log: VoiceLogger;\n    /** Local model overrides ('' → default per language). */\n    voskModel: string;\n    /** OpenAI STT model ('' → 'whisper-1'; e.g. 'gpt-4o-transcribe'). */\n    sttModel: string;\n    /** OpenAI TTS model ('' → 'tts-1'; e.g. 'gpt-4o-mini-tts'). */\n    ttsModel: string;\n}\n\nfunction requireAzure(creds: VoiceCredentials): void {\n    if (!creds.azureKey || !creds.azureRegion) {\n        throw new Error('Azure speech needs a subscription key and region');\n    }\n}\n\nfunction requireAws(creds: VoiceCredentials): void {\n    if (!creds.aws.accessKeyId || !creds.aws.secretAccessKey || !creds.aws.region) {\n        throw new Error('AWS speech needs an access key id, secret access key and region');\n    }\n}\n\nfunction requireOpenAi(creds: VoiceCredentials): void {\n    if (!creds.openaiKey) {\n        throw new Error('OpenAI speech needs an API key');\n    }\n}\n\nexport function createSttEngine(provider: SpeechProvider, ctx: EngineContext): SttEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureStt(ctx.creds.azureKey, ctx.creds.azureRegion);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsStt(ctx.creds.aws);\n        case 'vosk':\n            return new VoskStt(ctx.dataDir, ctx.voskModel, ctx.log);\n        case 'piper':\n            throw new Error('Piper is a text-to-speech engine, not speech-to-text');\n        default:\n            requireOpenAi(ctx.creds);\n            return new OpenAiStt(ctx.creds.openaiKey, ctx.sttModel || undefined);\n    }\n}\n\nexport function createTtsEngine(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureTts(ctx.creds.azureKey, ctx.creds.azureRegion, ctx.voices.azure);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsTts(ctx.creds.aws, ctx.voices.aws);\n        case 'piper':\n            return new PiperTts(ctx.dataDir, ctx.voices.piper, ctx.log);\n        case 'vosk':\n            throw new Error('Vosk is a speech-to-text engine, not text-to-speech');\n        default:\n            requireOpenAi(ctx.creds);\n            return new OpenAiTts(\n                ctx.creds.openaiKey,\n                (ctx.voices.openai || 'alloy') as OpenAiVoice,\n                ctx.ttsModel || undefined,\n            );\n    }\n}\n\n/** OpenAI TTS voices are a fixed set (no list API). */\nconst OPENAI_VOICES = ['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'];\n\n/** List the available TTS voices for a provider (for the settings voice dropdown). */\nexport async function listVoices(provider: SpeechProvider, ctx: EngineContext, language: string): Promise<string[]> {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return listAzureVoices(ctx.creds.azureKey, ctx.creds.azureRegion, isoToLocale(language));\n        case 'aws':\n            requireAws(ctx.creds);\n            return listPollyVoices(ctx.creds.aws, isoToLocale(language));\n        case 'piper':\n            return listPiperVoices();\n        default:\n            return OPENAI_VOICES;\n    }\n}\n\n/** Suggested STT model names for the settings dropdown. Only Vosk has a local model catalogue. */\nexport function listSttModels(provider: SpeechProvider, language: string): string[] {\n    return provider === 'vosk' ? listVoskModels(language) : [];\n}\n"]}

package/build/lib/voice/lang.js

@@ -0,0 +1,33 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.isoToLocale = isoToLocale;
+/**
+ * Map an ISO-639-1 language code (or an already-qualified locale) to a BCP-47 locale,
+ * as required by Azure Speech and AWS (Polly/Transcribe). OpenAI/Whisper uses the bare
+ * ISO code, so those engines strip the region again.
+ */
+const LOCALES = {
+    de: 'de-DE',
+    en: 'en-US',
+    ru: 'ru-RU',
+    fr: 'fr-FR',
+    it: 'it-IT',
+    es: 'es-ES',
+    nl: 'nl-NL',
+    pl: 'pl-PL',
+    pt: 'pt-PT',
+    uk: 'uk-UA',
+    zh: 'zh-CN',
+};
+function isoToLocale(lang, fallback = 'en-US') {
+    if (!lang) {
+        return fallback;
+    }
+    if (lang.includes('-')) {
+        // Already a locale (e.g. 'zh-cn') — normalise casing to 'zh-CN'.
+        const [l, r] = lang.split('-');
+        return `${l.toLowerCase()}-${r.toUpperCase()}`;
+    }
+    return LOCALES[lang.toLowerCase()] || fallback;
+}
+//# sourceMappingURL=lang.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"lang.js","sourceRoot":"","sources":["../../../src/lib/voice/lang.ts"],"names":[],"mappings":";;AAmBA,kCAUC;AA7BD;;;;GAIG;AACH,MAAM,OAAO,GAA2B;IACpC,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;IACX,EAAE,EAAE,OAAO;CACd,CAAC;AAEF,SAAgB,WAAW,CAAC,IAAY,EAAE,QAAQ,GAAG,OAAO;IACxD,IAAI,CAAC,IAAI,EAAE,CAAC;QACR,OAAO,QAAQ,CAAC;IACpB,CAAC;IACD,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACrB,iEAAiE;QACjE,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,OAAO,GAAG,CAAC,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;IACnD,CAAC;IACD,OAAO,OAAO,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,IAAI,QAAQ,CAAC;AACnD,CAAC","sourcesContent":["/**\n * Map an ISO-639-1 language code (or an already-qualified locale) to a BCP-47 locale,\n * as required by Azure Speech and AWS (Polly/Transcribe). OpenAI/Whisper uses the bare\n * ISO code, so those engines strip the region again.\n */\nconst LOCALES: Record<string, string> = {\n    de: 'de-DE',\n    en: 'en-US',\n    ru: 'ru-RU',\n    fr: 'fr-FR',\n    it: 'it-IT',\n    es: 'es-ES',\n    nl: 'nl-NL',\n    pl: 'pl-PL',\n    pt: 'pt-PT',\n    uk: 'uk-UA',\n    zh: 'zh-CN',\n};\n\nexport function isoToLocale(lang: string, fallback = 'en-US'): string {\n    if (!lang) {\n        return fallback;\n    }\n    if (lang.includes('-')) {\n        // Already a locale (e.g. 'zh-cn') — normalise casing to 'zh-CN'.\n        const [l, r] = lang.split('-');\n        return `${l.toLowerCase()}-${r.toUpperCase()}`;\n    }\n    return LOCALES[lang.toLowerCase()] || fallback;\n}\n"]}

package/build/lib/voice/piper.js

@@ -0,0 +1,191 @@
+"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.PiperTts = void 0;
+exports.listPiperVoices = listPiperVoices;
+/**
+ * Local TTS via Piper (https://github.com/rhasspy/piper). The binary + a voice model are downloaded
+ * into the instance data dir on first use, then `piper --output_raw` is spawned to synthesise 16-bit
+ * PCM. No cloud, no API key.
+ *
+ * ⚠️ Binary/voice download + spawn are platform-specific — validate on the target device.
+ */
+const fs = __importStar(require("node:fs"));
+const path = __importStar(require("node:path"));
+const node_child_process_1 = require("node:child_process");
+const download_1 = require("./download");
+const PIPER_RELEASE = 'https://github.com/rhasspy/piper/releases/download/2023.11.14-2';
+/** Release asset for the current platform/arch. */
+function piperAsset() {
+    const p = process.platform;
+    const a = process.arch;
+    if (p === 'linux') {
+        if (a === 'arm64') {
+            return 'piper_linux_aarch64.tar.gz';
+        }
+        if (a === 'arm') {
+            return 'piper_linux_armv7l.tar.gz';
+        }
+        return 'piper_linux_x86_64.tar.gz';
+    }
+    if (p === 'darwin') {
+        return a === 'arm64' ? 'piper_macos_aarch64.tar.gz' : 'piper_macos_x64.tar.gz';
+    }
+    if (p === 'win32') {
+        return 'piper_windows_amd64.zip';
+    }
+    throw new Error(`Piper: unsupported platform ${p}/${a}`);
+}
+/** Reasonable default Piper voice per language (rhasspy/piper-voices). */
+const DEFAULT_VOICE = {
+    de: 'de_DE-thorsten-medium',
+    en: 'en_US-lessac-medium',
+    ru: 'ru_RU-irina-medium',
+    fr: 'fr_FR-siwis-medium',
+    it: 'it_IT-riccardo-x_low',
+    es: 'es_ES-sharvard-medium',
+    nl: 'nl_NL-mls-medium',
+    pl: 'pl_PL-darkman-medium',
+    pt: 'pt_PT-tugão-medium',
+    uk: 'uk_UA-ukrainian_tts-medium',
+    zh: 'zh_CN-huayan-medium',
+};
+/** Build the two voice-file URLs (.onnx + .onnx.json) from a voice name like `de_DE-thorsten-medium`. */
+function voiceUrls(voice) {
+    // name = <locale>-<speaker>-<quality>; path = <lang>/<locale>/<speaker>/<quality>/<name>.onnx
+    const [locale, speaker, quality] = voice.split('-');
+    const lang = locale.split('_')[0].toLowerCase();
+    const base = `https://huggingface.co/rhasspy/piper-voices/resolve/main/${lang}/${locale}/${speaker}/${quality}/${voice}`;
+    return { onnx: `${base}.onnx`, json: `${base}.onnx.json` };
+}
+class PiperTts {
+    dataDir;
+    voice;
+    log;
+    binaryPath = '';
+    ready = null;
+    constructor(dataDir, 
+    /** Voice name (e.g. de_DE-thorsten-medium); '' → default per language. */
+    voice, log) {
+        this.dataDir = dataDir;
+        this.voice = voice;
+        this.log = log;
+    }
+    /** Download + unpack the binary once; concurrent calls share one init. */
+    ensure() {
+        if (!this.ready) {
+            this.ready = this.doEnsure().catch(e => {
+                this.ready = null; // allow a retry on the next call
+                throw e;
+            });
+        }
+        return this.ready;
+    }
+    async doEnsure() {
+        // 1. binary
+        const piperDir = path.join(this.dataDir, 'piper');
+        const exe = path.join(piperDir, 'piper', process.platform === 'win32' ? 'piper.exe' : 'piper');
+        if (!fs.existsSync(exe)) {
+            const asset = piperAsset();
+            const archive = path.join(piperDir, asset);
+            await (0, download_1.downloadFile)(`${PIPER_RELEASE}/${asset}`, archive, this.log);
+            await (0, download_1.extractArchive)(archive, piperDir, this.log);
+            if (process.platform !== 'win32') {
+                try {
+                    fs.chmodSync(exe, 0o755);
+                }
+                catch {
+                    /* best effort */
+                }
+            }
+        }
+        this.binaryPath = exe;
+    }
+    voiceFor(lang) {
+        const iso = (lang || 'en').split('-')[0].toLowerCase();
+        return this.voice || DEFAULT_VOICE[iso] || DEFAULT_VOICE.en;
+    }
+    async ensureVoice(voice) {
+        const dir = path.join(this.dataDir, 'piper-voices');
+        const onnx = path.join(dir, `${voice}.onnx`);
+        const json = path.join(dir, `${voice}.onnx.json`);
+        const urls = voiceUrls(voice);
+        await (0, download_1.downloadFile)(urls.onnx, onnx, this.log);
+        await (0, download_1.downloadFile)(urls.json, json, this.log);
+        let sampleRate = 22050;
+        try {
+            const cfg = JSON.parse(fs.readFileSync(json, 'utf8'));
+            sampleRate = cfg.audio?.sample_rate || 22050;
+        }
+        catch {
+            /* keep default */
+        }
+        return { onnx, sampleRate };
+    }
+    /** Warm-up: download the Piper binary + the voice for `lang` so the first command isn't slow. */
+    async prepare(lang) {
+        await this.ensure();
+        await this.ensureVoice(this.voiceFor(lang));
+    }
+    async synthesize(text, lang) {
+        await this.ensure();
+        const voice = this.voiceFor(lang);
+        const { onnx, sampleRate } = await this.ensureVoice(voice);
+        return new Promise((resolve, reject) => {
+            const child = (0, node_child_process_1.spawn)(this.binaryPath, ['--model', onnx, '--output_raw'], {
+                cwd: path.dirname(this.binaryPath), // finds bundled espeak-ng-data / libs
+            });
+            const chunks = [];
+            child.stdout.on('data', (d) => chunks.push(d));
+            child.stderr.on('data', (d) => this.log.debug(`piper: ${String(d).trim()}`));
+            child.on('error', e => reject(new Error(`piper failed: ${e.message}`)));
+            child.on('close', code => {
+                if (code === 0) {
+                    resolve({ pcm: Buffer.concat(chunks), sampleRate });
+                }
+                else {
+                    reject(new Error(`piper exited with code ${code}`));
+                }
+            });
+            child.stdin.end(text);
+        });
+    }
+}
+exports.PiperTts = PiperTts;
+/** Voices offered in the settings dropdown for Piper (the built-in per-language defaults). */
+function listPiperVoices() {
+    return Object.values(DEFAULT_VOICE);
+}
+//# sourceMappingURL=piper.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"piper.js","sourceRoot":"","sources":["../../../src/lib/voice/piper.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6JA,0CAEC;AA/JD;;;;;;GAMG;AACH,4CAA8B;AAC9B,gDAAkC;AAClC,2DAA2C;AAC3C,yCAA4E;AAG5E,MAAM,aAAa,GAAG,iEAAiE,CAAC;AAExF,mDAAmD;AACnD,SAAS,UAAU;IACf,MAAM,CAAC,GAAG,OAAO,CAAC,QAAQ,CAAC;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC;IACvB,IAAI,CAAC,KAAK,OAAO,EAAE,CAAC;QAChB,IAAI,CAAC,KAAK,OAAO,EAAE,CAAC;YAChB,OAAO,4BAA4B,CAAC;QACxC,CAAC;QACD,IAAI,CAAC,KAAK,KAAK,EAAE,CAAC;YACd,OAAO,2BAA2B,CAAC;QACvC,CAAC;QACD,OAAO,2BAA2B,CAAC;IACvC,CAAC;IACD,IAAI,CAAC,KAAK,QAAQ,EAAE,CAAC;QACjB,OAAO,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,4BAA4B,CAAC,CAAC,CAAC,wBAAwB,CAAC;IACnF,CAAC;IACD,IAAI,CAAC,KAAK,OAAO,EAAE,CAAC;QAChB,OAAO,yBAAyB,CAAC;IACrC,CAAC;IACD,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAC7D,CAAC;AAED,0EAA0E;AAC1E,MAAM,aAAa,GAA2B;IAC1C,EAAE,EAAE,uBAAuB;IAC3B,EAAE,EAAE,qBAAqB;IACzB,EAAE,EAAE,oBAAoB;IACxB,EAAE,EAAE,oBAAoB;IACxB,EAAE,EAAE,sBAAsB;IAC1B,EAAE,EAAE,uBAAuB;IAC3B,EAAE,EAAE,kBAAkB;IACtB,EAAE,EAAE,sBAAsB;IAC1B,EAAE,EAAE,oBAAoB;IACxB,EAAE,EAAE,4BAA4B;IAChC,EAAE,EAAE,qBAAqB;CAC5B,CAAC;AAEF,yGAAyG;AACzG,SAAS,SAAS,CAAC,KAAa;IAC5B,8FAA8F;IAC9F,MAAM,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACpD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;IAChD,MAAM,IAAI,GAAG,4DAA4D,IAAI,IAAI,MAAM,IAAI,OAAO,IAAI,OAAO,IAAI,KAAK,EAAE,CAAC;IACzH,OAAO,EAAE,IAAI,EAAE,GAAG,IAAI,OAAO,EAAE,IAAI,EAAE,GAAG,IAAI,YAAY,EAAE,CAAC;AAC/D,CAAC;AAED,MAAa,QAAQ;IAKI;IAEA;IACA;IAPb,UAAU,GAAG,EAAE,CAAC;IAChB,KAAK,GAAyB,IAAI,CAAC;IAE3C,YACqB,OAAe;IAChC,0EAA0E;IACzD,KAAa,EACb,GAAgB;QAHhB,YAAO,GAAP,OAAO,CAAQ;QAEf,UAAK,GAAL,KAAK,CAAQ;QACb,QAAG,GAAH,GAAG,CAAa;IAClC,CAAC;IAEJ,0EAA0E;IAClE,MAAM;QACV,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC;YACd,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;gBACnC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,CAAC,iCAAiC;gBACpD,MAAM,CAAC,CAAC;YACZ,CAAC,CAAC,CAAC;QACP,CAAC;QACD,OAAO,IAAI,CAAC,KAAK,CAAC;IACtB,CAAC;IAEO,KAAK,CAAC,QAAQ;QAClB,YAAY;QACZ,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAClD,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;QAC/F,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YACtB,MAAM,KAAK,GAAG,UAAU,EAAE,CAAC;YAC3B,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YAC3C,MAAM,IAAA,uBAAY,EAAC,GAAG,aAAa,IAAI,KAAK,EAAE,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;YACnE,MAAM,IAAA,yBAAc,EAAC,OAAO,EAAE,QAAQ,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;YAClD,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;gBAC/B,IAAI,CAAC;oBACD,EAAE,CAAC,SAAS,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;gBAC7B,CAAC;gBAAC,MAAM,CAAC;oBACL,iBAAiB;gBACrB,CAAC;YACL,CAAC;QACL,CAAC;QACD,IAAI,CAAC,UAAU,GAAG,GAAG,CAAC;IAC1B,CAAC;IAEO,QAAQ,CAAC,IAAY;QACzB,MAAM,GAAG,GAAG,CAAC,IAAI,IAAI,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;QACvD,OAAO,IAAI,CAAC,KAAK,IAAI,aAAa,CAAC,GAAG,CAAC,IAAI,aAAa,CAAC,EAAE,CAAC;IAChE,CAAC;IAEO,KAAK,CAAC,WAAW,CAAC,KAAa;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;QACpD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,OAAO,CAAC,CAAC;QAC7C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,YAAY,CAAC,CAAC;QAClD,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAC9B,MAAM,IAAA,uBAAY,EAAC,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QAC9C,MAAM,IAAA,uBAAY,EAAC,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QAC9C,IAAI,UAAU,GAAG,KAAK,CAAC;QACvB,IAAI,CAAC;YACD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAAyC,CAAC;YAC9F,UAAU,GAAG,GAAG,CAAC,KAAK,EAAE,WAAW,IAAI,KAAK,CAAC;QACjD,CAAC;QAAC,MAAM,CAAC;YACL,kBAAkB;QACtB,CAAC;QACD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;IAChC,CAAC;IAED,iGAAiG;IACjG,KAAK,CAAC,OAAO,CAAC,IAAY;QACtB,MAAM,IAAI,CAAC,MAAM,EAAE,CAAC;QACpB,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC;IAChD,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,IAAY,EAAE,IAAY;QACvC,MAAM,IAAI,CAAC,MAAM,EAAE,CAAC;QACpB,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QAClC,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,MAAM,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAE3D,OAAO,IAAI,OAAO,CAAY,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YAC9C,MAAM,KAAK,GAAG,IAAA,0BAAK,EAAC,IAAI,CAAC,UAAU,EAAE,CAAC,SAAS,EAAE,IAAI,EAAE,cAAc,CAAC,EAAE;gBACpE,GAAG,EAAE,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,sCAAsC;aAC7E,CAAC,CAAC;YACH,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;YACvD,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;YACrF,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;YACxE,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE;gBACrB,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;oBACb,OAAO,CAAC,EAAE,GAAG,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC;gBACxD,CAAC;qBAAM,CAAC;oBACJ,MAAM,CAAC,IAAI,KAAK,CAAC,0BAA0B,IAAI,EAAE,CAAC,CAAC,CAAC;gBACxD,CAAC;YACL,CAAC,CAAC,CAAC;YACH,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1B,CAAC,CAAC,CAAC;IACP,CAAC;CACJ;AA7FD,4BA6FC;AAED,8FAA8F;AAC9F,SAAgB,eAAe;IAC3B,OAAO,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC;AACxC,CAAC","sourcesContent":["/**\n * Local TTS via Piper (https://github.com/rhasspy/piper). The binary + a voice model are downloaded\n * into the instance data dir on first use, then `piper --output_raw` is spawned to synthesise 16-bit\n * PCM. No cloud, no API key.\n *\n * ⚠️ Binary/voice download + spawn are platform-specific — validate on the target device.\n */\nimport * as fs from 'node:fs';\nimport * as path from 'node:path';\nimport { spawn } from 'node:child_process';\nimport { downloadFile, extractArchive, type VoiceLogger } from './download';\nimport type { TtsEngine, TtsResult } from './tts';\n\nconst PIPER_RELEASE = 'https://github.com/rhasspy/piper/releases/download/2023.11.14-2';\n\n/** Release asset for the current platform/arch. */\nfunction piperAsset(): string {\n    const p = process.platform;\n    const a = process.arch;\n    if (p === 'linux') {\n        if (a === 'arm64') {\n            return 'piper_linux_aarch64.tar.gz';\n        }\n        if (a === 'arm') {\n            return 'piper_linux_armv7l.tar.gz';\n        }\n        return 'piper_linux_x86_64.tar.gz';\n    }\n    if (p === 'darwin') {\n        return a === 'arm64' ? 'piper_macos_aarch64.tar.gz' : 'piper_macos_x64.tar.gz';\n    }\n    if (p === 'win32') {\n        return 'piper_windows_amd64.zip';\n    }\n    throw new Error(`Piper: unsupported platform ${p}/${a}`);\n}\n\n/** Reasonable default Piper voice per language (rhasspy/piper-voices). */\nconst DEFAULT_VOICE: Record<string, string> = {\n    de: 'de_DE-thorsten-medium',\n    en: 'en_US-lessac-medium',\n    ru: 'ru_RU-irina-medium',\n    fr: 'fr_FR-siwis-medium',\n    it: 'it_IT-riccardo-x_low',\n    es: 'es_ES-sharvard-medium',\n    nl: 'nl_NL-mls-medium',\n    pl: 'pl_PL-darkman-medium',\n    pt: 'pt_PT-tugão-medium',\n    uk: 'uk_UA-ukrainian_tts-medium',\n    zh: 'zh_CN-huayan-medium',\n};\n\n/** Build the two voice-file URLs (.onnx + .onnx.json) from a voice name like `de_DE-thorsten-medium`. */\nfunction voiceUrls(voice: string): { onnx: string; json: string } {\n    // name = <locale>-<speaker>-<quality>; path = <lang>/<locale>/<speaker>/<quality>/<name>.onnx\n    const [locale, speaker, quality] = voice.split('-');\n    const lang = locale.split('_')[0].toLowerCase();\n    const base = `https://huggingface.co/rhasspy/piper-voices/resolve/main/${lang}/${locale}/${speaker}/${quality}/${voice}`;\n    return { onnx: `${base}.onnx`, json: `${base}.onnx.json` };\n}\n\nexport class PiperTts implements TtsEngine {\n    private binaryPath = '';\n    private ready: Promise<void> | null = null;\n\n    constructor(\n        private readonly dataDir: string,\n        /** Voice name (e.g. de_DE-thorsten-medium); '' → default per language. */\n        private readonly voice: string,\n        private readonly log: VoiceLogger,\n    ) {}\n\n    /** Download + unpack the binary once; concurrent calls share one init. */\n    private ensure(): Promise<void> {\n        if (!this.ready) {\n            this.ready = this.doEnsure().catch(e => {\n                this.ready = null; // allow a retry on the next call\n                throw e;\n            });\n        }\n        return this.ready;\n    }\n\n    private async doEnsure(): Promise<void> {\n        // 1. binary\n        const piperDir = path.join(this.dataDir, 'piper');\n        const exe = path.join(piperDir, 'piper', process.platform === 'win32' ? 'piper.exe' : 'piper');\n        if (!fs.existsSync(exe)) {\n            const asset = piperAsset();\n            const archive = path.join(piperDir, asset);\n            await downloadFile(`${PIPER_RELEASE}/${asset}`, archive, this.log);\n            await extractArchive(archive, piperDir, this.log);\n            if (process.platform !== 'win32') {\n                try {\n                    fs.chmodSync(exe, 0o755);\n                } catch {\n                    /* best effort */\n                }\n            }\n        }\n        this.binaryPath = exe;\n    }\n\n    private voiceFor(lang: string): string {\n        const iso = (lang || 'en').split('-')[0].toLowerCase();\n        return this.voice || DEFAULT_VOICE[iso] || DEFAULT_VOICE.en;\n    }\n\n    private async ensureVoice(voice: string): Promise<{ onnx: string; sampleRate: number }> {\n        const dir = path.join(this.dataDir, 'piper-voices');\n        const onnx = path.join(dir, `${voice}.onnx`);\n        const json = path.join(dir, `${voice}.onnx.json`);\n        const urls = voiceUrls(voice);\n        await downloadFile(urls.onnx, onnx, this.log);\n        await downloadFile(urls.json, json, this.log);\n        let sampleRate = 22050;\n        try {\n            const cfg = JSON.parse(fs.readFileSync(json, 'utf8')) as { audio?: { sample_rate?: number } };\n            sampleRate = cfg.audio?.sample_rate || 22050;\n        } catch {\n            /* keep default */\n        }\n        return { onnx, sampleRate };\n    }\n\n    /** Warm-up: download the Piper binary + the voice for `lang` so the first command isn't slow. */\n    async prepare(lang: string): Promise<void> {\n        await this.ensure();\n        await this.ensureVoice(this.voiceFor(lang));\n    }\n\n    async synthesize(text: string, lang: string): Promise<TtsResult> {\n        await this.ensure();\n        const voice = this.voiceFor(lang);\n        const { onnx, sampleRate } = await this.ensureVoice(voice);\n\n        return new Promise<TtsResult>((resolve, reject) => {\n            const child = spawn(this.binaryPath, ['--model', onnx, '--output_raw'], {\n                cwd: path.dirname(this.binaryPath), // finds bundled espeak-ng-data / libs\n            });\n            const chunks: Buffer[] = [];\n            child.stdout.on('data', (d: Buffer) => chunks.push(d));\n            child.stderr.on('data', (d: Buffer) => this.log.debug(`piper: ${String(d).trim()}`));\n            child.on('error', e => reject(new Error(`piper failed: ${e.message}`)));\n            child.on('close', code => {\n                if (code === 0) {\n                    resolve({ pcm: Buffer.concat(chunks), sampleRate });\n                } else {\n                    reject(new Error(`piper exited with code ${code}`));\n                }\n            });\n            child.stdin.end(text);\n        });\n    }\n}\n\n/** Voices offered in the settings dropdown for Piper (the built-in per-language defaults). */\nexport function listPiperVoices(): string[] {\n    return Object.values(DEFAULT_VOICE);\n}\n"]}

package/build/lib/voice/protocol.js

@@ -0,0 +1,35 @@
+"use strict";
+/**
+ * Voice-satellite UDP wire protocol — compatible with the Hannah satellite
+ * (`C:\iot\Hannah\satellite-pi\satellite.py`) so existing ESP/Python satellites keep working.
+ *
+ * One UDP socket, first byte is the packet type:
+ *   0x01 CONTROL — UTF-8 JSON message (both directions)
+ *   0x02 AUDIO   — raw 16 kHz mono 16-bit signed LE PCM (satellite → server)
+ *   0x03 TTS     — raw mono 16-bit signed LE PCM (server → satellite)
+ *
+ * ⚠️ KEEP IN SYNC with the copy in the standalone `@iobroker/assistant-satellite` repo.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AUDIO_SAMPLE_RATE = exports.TYPE_TTS = exports.TYPE_AUDIO = exports.TYPE_CONTROL = void 0;
+exports.encodeControl = encodeControl;
+exports.encodeTts = encodeTts;
+exports.decodePacket = decodePacket;
+exports.TYPE_CONTROL = 0x01;
+exports.TYPE_AUDIO = 0x02;
+exports.TYPE_TTS = 0x03;
+/** Audio on the wire from the satellite is always 16 kHz, mono, 16-bit signed little-endian PCM. */
+exports.AUDIO_SAMPLE_RATE = 16000;
+/** Prefix a control JSON message with the CONTROL type byte. */
+function encodeControl(msg) {
+    return Buffer.concat([Buffer.from([exports.TYPE_CONTROL]), Buffer.from(JSON.stringify(msg), 'utf8')]);
+}
+/** Prefix a raw PCM chunk with the TTS type byte. */
+function encodeTts(pcm) {
+    return Buffer.concat([Buffer.from([exports.TYPE_TTS]), pcm]);
+}
+/** Split an incoming datagram into its type byte and payload (a view onto the original buffer). */
+function decodePacket(data) {
+    return { type: data[0], payload: data.subarray(1) };
+}
+//# sourceMappingURL=protocol.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"protocol.js","sourceRoot":"","sources":["../../../src/lib/voice/protocol.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;GAUG;;;AA2BH,sCAEC;AAGD,8BAEC;AAGD,oCAEC;AArCY,QAAA,YAAY,GAAG,IAAI,CAAC;AACpB,QAAA,UAAU,GAAG,IAAI,CAAC;AAClB,QAAA,QAAQ,GAAG,IAAI,CAAC;AAE7B,oGAAoG;AACvF,QAAA,iBAAiB,GAAG,KAAK,CAAC;AAmBvC,gEAAgE;AAChE,SAAgB,aAAa,CAAC,GAAgB;IAC1C,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,oBAAY,CAAC,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;AAClG,CAAC;AAED,qDAAqD;AACrD,SAAgB,SAAS,CAAC,GAAW;IACjC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,gBAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;AACzD,CAAC;AAED,mGAAmG;AACnG,SAAgB,YAAY,CAAC,IAAY;IACrC,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;AACxD,CAAC","sourcesContent":["/**\n * Voice-satellite UDP wire protocol — compatible with the Hannah satellite\n * (`C:\\iot\\Hannah\\satellite-pi\\satellite.py`) so existing ESP/Python satellites keep working.\n *\n * One UDP socket, first byte is the packet type:\n *   0x01 CONTROL — UTF-8 JSON message (both directions)\n *   0x02 AUDIO   — raw 16 kHz mono 16-bit signed LE PCM (satellite → server)\n *   0x03 TTS     — raw mono 16-bit signed LE PCM (server → satellite)\n *\n * ⚠️ KEEP IN SYNC with the copy in the standalone `@iobroker/assistant-satellite` repo.\n */\n\nexport const TYPE_CONTROL = 0x01;\nexport const TYPE_AUDIO = 0x02;\nexport const TYPE_TTS = 0x03;\n\n/** Audio on the wire from the satellite is always 16 kHz, mono, 16-bit signed little-endian PCM. */\nexport const AUDIO_SAMPLE_RATE = 16000;\n\n/** Control messages sent by a satellite to the server. */\nexport type SatToServer =\n    | { type: 'register'; device: string; room?: string; listen_port?: number }\n    | { type: 'heartbeat'; device: string }\n    | { type: 'audio_end'; device: string };\n\n/** Satellite lifecycle/status as reflected to the satellite (drives its LED) and to ioBroker states. */\nexport type SatelliteState = 'idle' | 'listening' | 'processing' | 'speaking';\n\n/** Control messages sent by the server to a satellite. */\nexport type ServerToSat =\n    | { type: 'registered'; ok: boolean }\n    | { type: 'heartbeat_ack' }\n    | { type: 'reregister' }\n    | { type: 'tts_end'; sample_rate: number }\n    | { type: 'status'; state: SatelliteState };\n\n/** Prefix a control JSON message with the CONTROL type byte. */\nexport function encodeControl(msg: ServerToSat): Buffer {\n    return Buffer.concat([Buffer.from([TYPE_CONTROL]), Buffer.from(JSON.stringify(msg), 'utf8')]);\n}\n\n/** Prefix a raw PCM chunk with the TTS type byte. */\nexport function encodeTts(pcm: Buffer): Buffer {\n    return Buffer.concat([Buffer.from([TYPE_TTS]), pcm]);\n}\n\n/** Split an incoming datagram into its type byte and payload (a view onto the original buffer). */\nexport function decodePacket(data: Buffer): { type: number; payload: Buffer } {\n    return { type: data[0], payload: data.subarray(1) };\n}\n"]}

package/build/lib/voice/stt.js

@@ -0,0 +1,116 @@
+"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.OpenAiStt = void 0;
+exports.hintsToPrompt = hintsToPrompt;
+exports.pcmToWav = pcmToWav;
+/**
+ * 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.
+ */
+const openai_1 = __importStar(require("openai"));
+/**
+ * Join vocabulary hints into a Whisper `prompt` string (a soft bias toward these terms), de-duplicated and
+ * length-bounded to stay within the model's ~224-token prompt budget. Returns '' when there are no hints.
+ */
+function hintsToPrompt(hints, maxChars = 800) {
+    if (!hints?.length) {
+        return '';
+    }
+    const seen = new Set();
+    const parts = [];
+    let len = 0;
+    for (const h of hints) {
+        const term = (h || '').trim();
+        const key = term.toLowerCase();
+        if (!term || seen.has(key)) {
+            continue;
+        }
+        if (len + term.length + 2 > maxChars) {
+            break;
+        }
+        seen.add(key);
+        parts.push(term);
+        len += term.length + 2;
+    }
+    return parts.join(', ');
+}
+/** Wrap raw mono 16-bit-signed-LE PCM into a minimal 44-byte WAV container. */
+function pcmToWav(pcm, sampleRate) {
+    const numChannels = 1;
+    const bitsPerSample = 16;
+    const blockAlign = (numChannels * bitsPerSample) / 8;
+    const byteRate = sampleRate * blockAlign;
+    const header = Buffer.alloc(44);
+    header.write('RIFF', 0);
+    header.writeUInt32LE(36 + pcm.length, 4);
+    header.write('WAVE', 8);
+    header.write('fmt ', 12);
+    header.writeUInt32LE(16, 16); // PCM fmt chunk size
+    header.writeUInt16LE(1, 20); // audio format = PCM
+    header.writeUInt16LE(numChannels, 22);
+    header.writeUInt32LE(sampleRate, 24);
+    header.writeUInt32LE(byteRate, 28);
+    header.writeUInt16LE(blockAlign, 32);
+    header.writeUInt16LE(bitsPerSample, 34);
+    header.write('data', 36);
+    header.writeUInt32LE(pcm.length, 40);
+    return Buffer.concat([header, pcm]);
+}
+/** Cloud STT via the OpenAI transcription API (Whisper). */
+class OpenAiStt {
+    client;
+    model;
+    constructor(apiKey, model = 'whisper-1') {
+        this.client = new openai_1.default({ apiKey });
+        this.model = model;
+    }
+    async transcribe(pcm, sampleRate, lang, hints) {
+        const file = await (0, openai_1.toFile)(pcmToWav(pcm, sampleRate), 'audio.wav', { type: 'audio/wav' });
+        // A prompt biases Whisper toward the given domain vocabulary (device/room names) without excluding
+        // other words — improves recognition of proper nouns the NLU then needs to match.
+        const prompt = hintsToPrompt(hints);
+        const res = await this.client.audio.transcriptions.create({
+            file,
+            model: this.model,
+            // Whisper wants the bare ISO-639-1 code; strip any region (de-DE → de). '' → auto-detect.
+            language: (lang || '').split('-')[0] || undefined,
+            ...(prompt ? { prompt } : {}),
+        });
+        return (res.text || '').trim();
+    }
+}
+exports.OpenAiStt = OpenAiStt;
+//# sourceMappingURL=stt.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"stt.js","sourceRoot":"","sources":["../../../src/lib/voice/stt.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwBA,sCAqBC;AAGD,4BAoBC;AApED;;;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,+EAA+E;AAC/E,SAAgB,QAAQ,CAAC,GAAW,EAAE,UAAkB;IACpD,MAAM,WAAW,GAAG,CAAC,CAAC;IACtB,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 mono 16-bit-signed-LE PCM into a minimal 44-byte WAV container. */\nexport function pcmToWav(pcm: Buffer, sampleRate: number): Buffer {\n    const numChannels = 1;\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"]}