iobroker.assistant 1.0.3 → 1.2.0

package/build/lib/voice/wakeWords.js

@@ -0,0 +1,284 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.WAKE_WORD_URL_PREFIX = exports.WAKE_WORD_DIR = exports.WAKE_WORD_ID_RE = exports.MAX_MODEL_BYTES = void 0;
+exports.slugify = slugify;
+exports.isTfliteModel = isTfliteModel;
+exports.readTfliteIo = readTfliteIo;
+exports.microModelProblem = microModelProblem;
+exports.prepareWakeWord = prepareWakeWord;
+exports.loadWakeWord = loadWakeWord;
+/**
+ * Custom wake words for ESPHome satellites — models the user uploads, offered to the devices as
+ * "external wake words".
+ *
+ * The ESPHome API has no upload message. Instead the client lists models in every
+ * `VoiceAssistantConfigurationRequest` (`external_wake_words`: id, phrase, languages, type, size,
+ * SHA-256 and the URL of a JSON config). The device adds them to its available list, and only when one
+ * is selected does it fetch the config from `url` and the model from the **same directory** as
+ * `<id>.tflite` — query and fragment are dropped — storing both as `<id>.json` / `<id>.tflite` next to
+ * its built-in models. Verified in the ThirdReality firmware (`linux-voice-assistant-cpp`,
+ * `Satellite.cpp` `ResolveWakeWordConfig`, `audio/ExternalWakeWord.cpp`). Consequences baked in here:
+ *
+ * - **Only microWakeWord models** (`model_type: "micro"`): the firmware skips any other type.
+ * - The device loads the model named by the JSON's `model` field, relative to the JSON — so the JSON we
+ *   serve always says `<id>.tflite`, whatever the uploaded file was called.
+ * - Once downloaded, a model is a *local* one on the device and wins over the external offer by id. A
+ *   changed model under the same id would never reach a device that already has it, so the id carries
+ *   a short hash of the model: a new upload is a new wake word.
+ * - Downloaded models stay on the device; the API has no way to remove them.
+ */
+const node_crypto_1 = require("node:crypto");
+/** microWakeWord models are 50–200 KB; anything near this is not one. */
+exports.MAX_MODEL_BYTES = 1024 * 1024;
+/** Ids are file names on the device and path segments in our URL. */
+exports.WAKE_WORD_ID_RE = /^[a-z0-9_]{1,64}$/;
+/** Folder in the instance's file storage (`assistant.0/wakewords/`). */
+exports.WAKE_WORD_DIR = 'wakewords';
+/** Where the media server serves them: `<prefix><id>.json` and `<prefix><id>.tflite`. */
+exports.WAKE_WORD_URL_PREFIX = '/wakeword/';
+/** Defaults of the official v2 models, used when a model comes without its JSON. */
+const DEFAULT_MICRO = { probability_cutoff: 0.97, sliding_window_size: 5 };
+/** `hey_hannah.tflite` → `hey_hannah`; anything that is not a-z/0-9 becomes `_`. */
+function slugify(value) {
+    return value
+        .replace(/^.*[\\/]/, '')
+        .toLowerCase()
+        .replace(/\.(tflite|json)$/, '')
+        .normalize('NFKD')
+        .replace(/[̀-ͯ]/g, '')
+        .replace(/ß/g, 'ss')
+        .replace(/[^a-z0-9]+/g, '_')
+        .replace(/^_+|_+$/g, '')
+        .slice(0, 40)
+        .replace(/_+$/, '');
+}
+/** `hey_hannah` → `Hey Hannah` — a phrase for a model that came without its JSON. */
+function phraseFromFileName(fileName) {
+    const base = fileName.replace(/^.*[\\/]/, '').replace(/\.(tflite|json)$/i, '');
+    return base
+        .split(/[_\-\s]+/)
+        .filter(Boolean)
+        .map(w => w[0].toUpperCase() + w.slice(1))
+        .join(' ');
+}
+/** A TensorFlow Lite flatbuffer carries the file identifier `TFL3` at byte 4. */
+function isTfliteModel(model) {
+    return model.length > 8 && model.toString('latin1', 4, 8) === 'TFL3';
+}
+/** microWakeWord feeds 40 mel features per frame (`MicroFeatures::kFeatureSize` on the device). */
+const MICRO_FEATURES = 40;
+/** openWakeWord classifiers take 96-value speech embeddings, e.g. input `[1, 16, 96]`. */
+const OPEN_WAKE_WORD_FEATURES = 96;
+/** Shapes longer than this are not a wake word model; it also bounds the loop over a broken file. */
+const MAX_DIMENSIONS = 8;
+/**
+ * Read the first input and output tensor of the main graph — the two the device checks before it runs a
+ * model. Walks the flatbuffer by hand (schema: `tensorflow/compiler/mlir/lite/schema/schema.fbs`), which
+ * spares a TensorFlow dependency for four fields. `null` when the file is not a readable model.
+ */
+function readTfliteIo(model) {
+    // Field positions in the schema's tables.
+    const MODEL_SUBGRAPHS = 2;
+    const SUBGRAPH_TENSORS = 0;
+    const SUBGRAPH_INPUTS = 1;
+    const SUBGRAPH_OUTPUTS = 2;
+    const TENSOR_SHAPE = 0;
+    const TENSOR_TYPE = 1;
+    const TENSOR_QUANTIZATION = 4;
+    const QUANTIZATION_SCALE = 2;
+    /** Position of field `index` of the table at `table`, 0 when it is absent. */
+    const field = (table, index) => {
+        const vtable = table - model.readInt32LE(table);
+        const offset = 4 + 2 * index;
+        if (offset >= model.readUInt16LE(vtable)) {
+            return 0;
+        }
+        const at = model.readUInt16LE(vtable + offset);
+        return at ? table + at : 0;
+    };
+    /** Follow the offset stored in field `index` to the table, vector or string it points at. */
+    const ref = (table, index) => {
+        const at = field(table, index);
+        return at ? at + model.readUInt32LE(at) : 0;
+    };
+    const length = (vector) => (vector ? model.readUInt32LE(vector) : 0);
+    /** Element `k` of a vector of tables. */
+    const tableAt = (vector, k) => {
+        const at = vector + 4 + 4 * k;
+        return at + model.readUInt32LE(at);
+    };
+    const intAt = (vector, k) => model.readInt32LE(vector + 4 + 4 * k);
+    const tensor = (subgraph, list) => {
+        const tensors = ref(subgraph, SUBGRAPH_TENSORS);
+        if (!length(list)) {
+            return null;
+        }
+        const index = intAt(list, 0);
+        if (index < 0 || index >= length(tensors)) {
+            return null;
+        }
+        const t = tableAt(tensors, index);
+        const dims = ref(t, TENSOR_SHAPE);
+        if (length(dims) > MAX_DIMENSIONS) {
+            return null;
+        }
+        const shape = Array.from({ length: length(dims) }, (_, k) => intAt(dims, k));
+        const typeAt = field(t, TENSOR_TYPE);
+        const quantization = ref(t, TENSOR_QUANTIZATION);
+        const scales = quantization ? ref(quantization, QUANTIZATION_SCALE) : 0;
+        return {
+            shape,
+            type: typeAt ? model.readUInt8(typeAt) : 0,
+            quantized: length(scales) > 0 && model.readFloatLE(scales + 4) !== 0,
+        };
+    };
+    if (!isTfliteModel(model)) {
+        return null;
+    }
+    try {
+        const subgraphs = ref(model.readUInt32LE(0), MODEL_SUBGRAPHS);
+        if (!length(subgraphs)) {
+            return null;
+        }
+        const main = tableAt(subgraphs, 0);
+        const input = tensor(main, ref(main, SUBGRAPH_INPUTS));
+        const output = tensor(main, ref(main, SUBGRAPH_OUTPUTS));
+        return input && output ? { input, output } : null;
+    }
+    catch {
+        return null; // an offset pointed outside the file
+    }
+}
+/**
+ * Why the device cannot run this model, or `null` when it can. Mirrors the checks of the firmware's
+ * `MicroWakeWord::FromConfig`: input `[1, stride, 40]`, quantized input and output. An openWakeWord
+ * model is named as such — it is the most likely mix-up, both come as `.tflite`, and the device runs
+ * openWakeWord only when its firmware was started in that mode (then without the built-in models).
+ *
+ * @param model the `.tflite` file
+ */
+function microModelProblem(model) {
+    if (!isTfliteModel(model)) {
+        return 'the model file is not a TensorFlow Lite model (.tflite)';
+    }
+    const io = readTfliteIo(model);
+    if (!io) {
+        return 'the model file is damaged — it could not be read as a TensorFlow Lite model';
+    }
+    const shape = io.input.shape;
+    if (shape[shape.length - 1] === OPEN_WAKE_WORD_FEATURES) {
+        return 'this is an openWakeWord model — the device runs only microWakeWord models, so the wake word has to be trained with microWakeWord';
+    }
+    if (shape.length !== 3 || shape[2] !== MICRO_FEATURES) {
+        return `this is not a microWakeWord model (input [${shape.join(', ')}], a microWakeWord model takes [1, n, ${MICRO_FEATURES}])`;
+    }
+    if (!io.input.quantized || !io.output.quantized) {
+        return 'the model is not quantized — the device runs only quantized microWakeWord models';
+    }
+    return null;
+}
+function parseConfig(config) {
+    if (config === undefined || config === null || config === '') {
+        return {};
+    }
+    if (typeof config === 'object') {
+        return config;
+    }
+    let parsed;
+    try {
+        parsed = JSON.parse(config);
+    }
+    catch (e) {
+        throw new Error(`the JSON file is not valid JSON: ${e.message}`);
+    }
+    if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) {
+        throw new Error('the JSON file must contain an object');
+    }
+    return parsed;
+}
+function languageList(value) {
+    return Array.isArray(value) ? [...new Set(value.map(v => String(v).trim().toLowerCase()).filter(Boolean))] : [];
+}
+/**
+ * Validate an upload and build the wake word the devices will download. Throws with a message meant for
+ * the user when the files cannot work on the device.
+ *
+ * @param upload what was uploaded
+ * @param defaultLanguage language to record when neither the upload nor its JSON names one
+ */
+function prepareWakeWord(upload, defaultLanguage = '') {
+    const model = upload.model;
+    if (!model?.length) {
+        throw new Error('no model file (.tflite) was uploaded');
+    }
+    if (model.length > exports.MAX_MODEL_BYTES) {
+        throw new Error(`the model is ${Math.round(model.length / 1024)} KB — microWakeWord models are well below ${exports.MAX_MODEL_BYTES / 1024} KB`);
+    }
+    const problem = microModelProblem(model);
+    if (problem) {
+        throw new Error(problem);
+    }
+    const source = parseConfig(upload.config);
+    const type = typeof source.type === 'string' ? source.type : '';
+    if (type && type !== 'micro') {
+        throw new Error(`this is a "${type}" model — the device only accepts microWakeWord models ("micro") from outside`);
+    }
+    const fileName = upload.fileName || (typeof source.model === 'string' ? source.model : '');
+    const phrase = (upload.phrase?.trim() ||
+        (typeof source.wake_word === 'string' ? source.wake_word.trim() : '') ||
+        phraseFromFileName(fileName)).slice(0, 64);
+    if (!phrase) {
+        throw new Error('the wake word needs a phrase (what you say to wake the device)');
+    }
+    let languages = languageList(upload.languages);
+    if (!languages.length) {
+        languages = languageList(source.trained_languages);
+    }
+    if (!languages.length && defaultLanguage) {
+        languages = [defaultLanguage.toLowerCase()];
+    }
+    const sha256 = (0, node_crypto_1.createHash)('sha256').update(model).digest('hex');
+    const id = `${slugify(fileName) || slugify(phrase) || 'wake_word'}_${sha256.slice(0, 8)}`;
+    const micro = source.micro && typeof source.micro === 'object' ? source.micro : {};
+    const config = {
+        ...source,
+        type: 'micro',
+        wake_word: phrase,
+        model: `${id}.tflite`,
+        trained_languages: languages,
+        version: typeof source.version === 'number' ? source.version : 2,
+        micro: { ...DEFAULT_MICRO, ...micro },
+    };
+    return {
+        id,
+        phrase,
+        languages,
+        config: Buffer.from(`${JSON.stringify(config, null, 2)}\n`),
+        model,
+        size: model.length,
+        sha256,
+    };
+}
+/**
+ * Rebuild a stored wake word from its two files (adapter start). The JSON is the one we wrote, so it is
+ * trusted for the phrase; size and hash are recomputed from the model rather than stored.
+ */
+function loadWakeWord(id, config, model) {
+    if (!exports.WAKE_WORD_ID_RE.test(id)) {
+        throw new Error(`invalid wake word id "${id}"`);
+    }
+    if (!isTfliteModel(model)) {
+        throw new Error(`${id}.tflite is not a TensorFlow Lite model`);
+    }
+    const parsed = parseConfig(config.toString('utf8'));
+    return {
+        id,
+        phrase: typeof parsed.wake_word === 'string' ? parsed.wake_word : id,
+        languages: languageList(parsed.trained_languages),
+        config,
+        model,
+        size: model.length,
+        sha256: (0, node_crypto_1.createHash)('sha256').update(model).digest('hex'),
+    };
+}
+//# sourceMappingURL=wakeWords.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"wakeWords.js","sourceRoot":"","sources":["../../../src/lib/voice/wakeWords.ts"],"names":[],"mappings":";;;AAmEA,0BAYC;AAaD,sCAEC;AAwBD,oCA0EC;AAUD,8CAmBC;AAgCD,0CAgEC;AAMD,oCAiBC;AApVD;;;;;;;;;;;;;;;;;;;GAmBG;AACH,6CAAyC;AAEzC,yEAAyE;AAC5D,QAAA,eAAe,GAAG,IAAI,GAAG,IAAI,CAAC;AAE3C,qEAAqE;AACxD,QAAA,eAAe,GAAG,mBAAmB,CAAC;AAEnD,wEAAwE;AAC3D,QAAA,aAAa,GAAG,WAAW,CAAC;AAEzC,yFAAyF;AAC5E,QAAA,oBAAoB,GAAG,YAAY,CAAC;AAEjD,oFAAoF;AACpF,MAAM,aAAa,GAAG,EAAE,kBAAkB,EAAE,IAAI,EAAE,mBAAmB,EAAE,CAAC,EAAE,CAAC;AA+B3E,oFAAoF;AACpF,SAAgB,OAAO,CAAC,KAAa;IACjC,OAAO,KAAK;SACP,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,WAAW,EAAE;SACb,OAAO,CAAC,kBAAkB,EAAE,EAAE,CAAC;SAC/B,SAAS,CAAC,MAAM,CAAC;SACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC;SACrB,OAAO,CAAC,IAAI,EAAE,IAAI,CAAC;SACnB,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;SACZ,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;AAC5B,CAAC;AAED,qFAAqF;AACrF,SAAS,kBAAkB,CAAC,QAAgB;IACxC,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,EAAE,EAAE,CAAC,CAAC;IAC/E,OAAO,IAAI;SACN,KAAK,CAAC,UAAU,CAAC;SACjB,MAAM,CAAC,OAAO,CAAC;SACf,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;SACzC,IAAI,CAAC,GAAG,CAAC,CAAC;AACnB,CAAC;AAED,iFAAiF;AACjF,SAAgB,aAAa,CAAC,KAAa;IACvC,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC,KAAK,MAAM,CAAC;AACzE,CAAC;AAED,mGAAmG;AACnG,MAAM,cAAc,GAAG,EAAE,CAAC;AAE1B,0FAA0F;AAC1F,MAAM,uBAAuB,GAAG,EAAE,CAAC;AAEnC,qGAAqG;AACrG,MAAM,cAAc,GAAG,CAAC,CAAC;AAUzB;;;;GAIG;AACH,SAAgB,YAAY,CAAC,KAAa;IACtC,0CAA0C;IAC1C,MAAM,eAAe,GAAG,CAAC,CAAC;IAC1B,MAAM,gBAAgB,GAAG,CAAC,CAAC;IAC3B,MAAM,eAAe,GAAG,CAAC,CAAC;IAC1B,MAAM,gBAAgB,GAAG,CAAC,CAAC;IAC3B,MAAM,YAAY,GAAG,CAAC,CAAC;IACvB,MAAM,WAAW,GAAG,CAAC,CAAC;IACtB,MAAM,mBAAmB,GAAG,CAAC,CAAC;IAC9B,MAAM,kBAAkB,GAAG,CAAC,CAAC;IAE7B,8EAA8E;IAC9E,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,KAAa,EAAU,EAAE;QACnD,MAAM,MAAM,GAAG,KAAK,GAAG,KAAK,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC;QAC7B,IAAI,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,MAAM,CAAC,EAAE,CAAC;YACvC,OAAO,CAAC,CAAC;QACb,CAAC;QACD,MAAM,EAAE,GAAG,KAAK,CAAC,YAAY,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;QAC/C,OAAO,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/B,CAAC,CAAC;IACF,6FAA6F;IAC7F,MAAM,GAAG,GAAG,CAAC,KAAa,EAAE,KAAa,EAAU,EAAE;QACjD,MAAM,EAAE,GAAG,KAAK,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;QAC/B,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,YAAY,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAChD,CAAC,CAAC;IACF,MAAM,MAAM,GAAG,CAAC,MAAc,EAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACrF,yCAAyC;IACzC,MAAM,OAAO,GAAG,CAAC,MAAc,EAAE,CAAS,EAAU,EAAE;QAClD,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC9B,OAAO,EAAE,GAAG,KAAK,CAAC,YAAY,CAAC,EAAE,CAAC,CAAC;IACvC,CAAC,CAAC;IACF,MAAM,KAAK,GAAG,CAAC,MAAc,EAAE,CAAS,EAAU,EAAE,CAAC,KAAK,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IAE3F,MAAM,MAAM,GAAG,CAAC,QAAgB,EAAE,IAAY,EAAuB,EAAE;QACnE,MAAM,OAAO,GAAG,GAAG,CAAC,QAAQ,EAAE,gBAAgB,CAAC,CAAC;QAChD,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC;YAChB,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,CAAC,IAAI,KAAK,IAAI,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;YACxC,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,MAAM,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAClC,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,EAAE,YAAY,CAAC,CAAC;QAClC,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,cAAc,EAAE,CAAC;YAChC,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAC7E,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC;QACrC,MAAM,YAAY,GAAG,GAAG,CAAC,CAAC,EAAE,mBAAmB,CAAC,CAAC;QACjD,MAAM,MAAM,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,CAAC,YAAY,EAAE,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACxE,OAAO;YACH,KAAK;YACL,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;YAC1C,SAAS,EAAE,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC;SACvE,CAAC;IACN,CAAC,CAAC;IAEF,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,IAAI,CAAC;IAChB,CAAC;IACD,IAAI,CAAC;QACD,MAAM,SAAS,GAAG,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,eAAe,CAAC,CAAC;QAC9D,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,EAAE,CAAC;YACrB,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,MAAM,IAAI,GAAG,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,eAAe,CAAC,CAAC,CAAC;QACvD,MAAM,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,gBAAgB,CAAC,CAAC,CAAC;QACzD,OAAO,KAAK,IAAI,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IACtD,CAAC;IAAC,MAAM,CAAC;QACL,OAAO,IAAI,CAAC,CAAC,qCAAqC;IACtD,CAAC;AACL,CAAC;AAED;;;;;;;GAOG;AACH,SAAgB,iBAAiB,CAAC,KAAa;IAC3C,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,yDAAyD,CAAC;IACrE,CAAC;IACD,MAAM,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;IAC/B,IAAI,CAAC,EAAE,EAAE,CAAC;QACN,OAAO,6EAA6E,CAAC;IACzF,CAAC;IACD,MAAM,KAAK,GAAG,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC;IAC7B,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,uBAAuB,EAAE,CAAC;QACtD,OAAO,kIAAkI,CAAC;IAC9I,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,cAAc,EAAE,CAAC;QACpD,OAAO,6CAA6C,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,yCAAyC,cAAc,IAAI,CAAC;IACpI,CAAC;IACD,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;QAC9C,OAAO,kFAAkF,CAAC;IAC9F,CAAC;IACD,OAAO,IAAI,CAAC;AAChB,CAAC;AAED,SAAS,WAAW,CAAC,MAAgC;IACjD,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,IAAI,IAAI,MAAM,KAAK,EAAE,EAAE,CAAC;QAC3D,OAAO,EAAE,CAAC;IACd,CAAC;IACD,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;QAC7B,OAAO,MAAM,CAAC;IAClB,CAAC;IACD,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACD,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,MAAM,IAAI,KAAK,CAAC,oCAAqC,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;IAChF,CAAC;IACD,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QACjE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAC5D,CAAC;IACD,OAAO,MAAiC,CAAC;AAC7C,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAChC,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AACpH,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,eAAe,CAAC,MAAsB,EAAE,eAAe,GAAG,EAAE;IACxE,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;IAC3B,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC;QACjB,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,GAAG,uBAAe,EAAE,CAAC;QACjC,MAAM,IAAI,KAAK,CACX,gBAAgB,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,6CAA6C,uBAAe,GAAG,IAAI,KAAK,CAC1H,CAAC;IACN,CAAC;IACD,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IACzC,IAAI,OAAO,EAAE,CAAC;QACV,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;IAC7B,CAAC;IAED,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,IAAI,GAAG,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;IAChE,IAAI,IAAI,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;QAC3B,MAAM,IAAI,KAAK,CACX,cAAc,IAAI,+EAA+E,CACpG,CAAC;IACN,CAAC;IAED,MAAM,QAAQ,GAAG,MAAM,CAAC,QAAQ,IAAI,CAAC,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC3F,MAAM,MAAM,GAAG,CACX,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE;QACrB,CAAC,OAAO,MAAM,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACrE,kBAAkB,CAAC,QAAQ,CAAC,CAC/B,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACf,IAAI,CAAC,MAAM,EAAE,CAAC;QACV,MAAM,IAAI,KAAK,CAAC,gEAAgE,CAAC,CAAC;IACtF,CAAC;IAED,IAAI,SAAS,GAAG,YAAY,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;IAC/C,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC;QACpB,SAAS,GAAG,YAAY,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;IACvD,CAAC;IACD,IAAI,CAAC,SAAS,CAAC,MAAM,IAAI,eAAe,EAAE,CAAC;QACvC,SAAS,GAAG,CAAC,eAAe,CAAC,WAAW,EAAE,CAAC,CAAC;IAChD,CAAC;IAED,MAAM,MAAM,GAAG,IAAA,wBAAU,EAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAChE,MAAM,EAAE,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,IAAI,OAAO,CAAC,MAAM,CAAC,IAAI,WAAW,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;IAE1F,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAE,MAAM,CAAC,KAAiC,CAAC,CAAC,CAAC,EAAE,CAAC;IAChH,MAAM,MAAM,GAAG;QACX,GAAG,MAAM;QACT,IAAI,EAAE,OAAO;QACb,SAAS,EAAE,MAAM;QACjB,KAAK,EAAE,GAAG,EAAE,SAAS;QACrB,iBAAiB,EAAE,SAAS;QAC5B,OAAO,EAAE,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAChE,KAAK,EAAE,EAAE,GAAG,aAAa,EAAE,GAAG,KAAK,EAAE;KACxC,CAAC;IAEF,OAAO;QACH,EAAE;QACF,MAAM;QACN,SAAS;QACT,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC;QAC3D,KAAK;QACL,IAAI,EAAE,KAAK,CAAC,MAAM;QAClB,MAAM;KACT,CAAC;AACN,CAAC;AAED;;;GAGG;AACH,SAAgB,YAAY,CAAC,EAAU,EAAE,MAAc,EAAE,KAAa;IAClE,IAAI,CAAC,uBAAe,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CAAC,yBAAyB,EAAE,GAAG,CAAC,CAAC;IACpD,CAAC;IACD,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,GAAG,EAAE,wCAAwC,CAAC,CAAC;IACnE,CAAC;IACD,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;IACpD,OAAO;QACH,EAAE;QACF,MAAM,EAAE,OAAO,MAAM,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE;QACpE,SAAS,EAAE,YAAY,CAAC,MAAM,CAAC,iBAAiB,CAAC;QACjD,MAAM;QACN,KAAK;QACL,IAAI,EAAE,KAAK,CAAC,MAAM;QAClB,MAAM,EAAE,IAAA,wBAAU,EAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;KAC3D,CAAC;AACN,CAAC","sourcesContent":["/**\n * Custom wake words for ESPHome satellites — models the user uploads, offered to the devices as\n * \"external wake words\".\n *\n * The ESPHome API has no upload message. Instead the client lists models in every\n * `VoiceAssistantConfigurationRequest` (`external_wake_words`: id, phrase, languages, type, size,\n * SHA-256 and the URL of a JSON config). The device adds them to its available list, and only when one\n * is selected does it fetch the config from `url` and the model from the **same directory** as\n * `<id>.tflite` — query and fragment are dropped — storing both as `<id>.json` / `<id>.tflite` next to\n * its built-in models. Verified in the ThirdReality firmware (`linux-voice-assistant-cpp`,\n * `Satellite.cpp` `ResolveWakeWordConfig`, `audio/ExternalWakeWord.cpp`). Consequences baked in here:\n *\n * - **Only microWakeWord models** (`model_type: \"micro\"`): the firmware skips any other type.\n * - The device loads the model named by the JSON's `model` field, relative to the JSON — so the JSON we\n *   serve always says `<id>.tflite`, whatever the uploaded file was called.\n * - Once downloaded, a model is a *local* one on the device and wins over the external offer by id. A\n *   changed model under the same id would never reach a device that already has it, so the id carries\n *   a short hash of the model: a new upload is a new wake word.\n * - Downloaded models stay on the device; the API has no way to remove them.\n */\nimport { createHash } from 'node:crypto';\n\n/** microWakeWord models are 50–200 KB; anything near this is not one. */\nexport const MAX_MODEL_BYTES = 1024 * 1024;\n\n/** Ids are file names on the device and path segments in our URL. */\nexport const WAKE_WORD_ID_RE = /^[a-z0-9_]{1,64}$/;\n\n/** Folder in the instance's file storage (`assistant.0/wakewords/`). */\nexport const WAKE_WORD_DIR = 'wakewords';\n\n/** Where the media server serves them: `<prefix><id>.json` and `<prefix><id>.tflite`. */\nexport const WAKE_WORD_URL_PREFIX = '/wakeword/';\n\n/** Defaults of the official v2 models, used when a model comes without its JSON. */\nconst DEFAULT_MICRO = { probability_cutoff: 0.97, sliding_window_size: 5 };\n\n/** One uploaded wake word, ready to be offered and served. */\nexport interface CustomWakeWord {\n    id: string;\n    /** What the user says, e.g. \"Hey Hannah\". */\n    phrase: string;\n    /** ISO-639-1 codes the model was trained on. */\n    languages: string[];\n    /** The JSON config exactly as the device will store it. */\n    config: Buffer;\n    /** The `.tflite` model. */\n    model: Buffer;\n    size: number;\n    /** SHA-256 of the model, lower-case hex — the device compares it with its cached copy. */\n    sha256: string;\n}\n\nexport interface WakeWordUpload {\n    /** The `.tflite` file. */\n    model: Buffer;\n    /** Name of the uploaded model file — the readable half of the id. */\n    fileName?: string;\n    /** The `.json` that came with the model, if any. */\n    config?: string | Record<string, unknown>;\n    /** Overrides the phrase from the JSON. */\n    phrase?: string;\n    /** Overrides the languages from the JSON. */\n    languages?: string[];\n}\n\n/** `hey_hannah.tflite` → `hey_hannah`; anything that is not a-z/0-9 becomes `_`. */\nexport function slugify(value: string): string {\n    return value\n        .replace(/^.*[\\\\/]/, '')\n        .toLowerCase()\n        .replace(/\\.(tflite|json)$/, '')\n        .normalize('NFKD')\n        .replace(/[̀-ͯ]/g, '')\n        .replace(/ß/g, 'ss')\n        .replace(/[^a-z0-9]+/g, '_')\n        .replace(/^_+|_+$/g, '')\n        .slice(0, 40)\n        .replace(/_+$/, '');\n}\n\n/** `hey_hannah` → `Hey Hannah` — a phrase for a model that came without its JSON. */\nfunction phraseFromFileName(fileName: string): string {\n    const base = fileName.replace(/^.*[\\\\/]/, '').replace(/\\.(tflite|json)$/i, '');\n    return base\n        .split(/[_\\-\\s]+/)\n        .filter(Boolean)\n        .map(w => w[0].toUpperCase() + w.slice(1))\n        .join(' ');\n}\n\n/** A TensorFlow Lite flatbuffer carries the file identifier `TFL3` at byte 4. */\nexport function isTfliteModel(model: Buffer): boolean {\n    return model.length > 8 && model.toString('latin1', 4, 8) === 'TFL3';\n}\n\n/** microWakeWord feeds 40 mel features per frame (`MicroFeatures::kFeatureSize` on the device). */\nconst MICRO_FEATURES = 40;\n\n/** openWakeWord classifiers take 96-value speech embeddings, e.g. input `[1, 16, 96]`. */\nconst OPEN_WAKE_WORD_FEATURES = 96;\n\n/** Shapes longer than this are not a wake word model; it also bounds the loop over a broken file. */\nconst MAX_DIMENSIONS = 8;\n\nexport interface TfliteTensor {\n    shape: number[];\n    /** TFLite `TensorType`: 0 float32, 3 uint8, 9 int8, … */\n    type: number;\n    /** Has a non-zero quantization scale. */\n    quantized: boolean;\n}\n\n/**\n * Read the first input and output tensor of the main graph — the two the device checks before it runs a\n * model. Walks the flatbuffer by hand (schema: `tensorflow/compiler/mlir/lite/schema/schema.fbs`), which\n * spares a TensorFlow dependency for four fields. `null` when the file is not a readable model.\n */\nexport function readTfliteIo(model: Buffer): { input: TfliteTensor; output: TfliteTensor } | null {\n    // Field positions in the schema's tables.\n    const MODEL_SUBGRAPHS = 2;\n    const SUBGRAPH_TENSORS = 0;\n    const SUBGRAPH_INPUTS = 1;\n    const SUBGRAPH_OUTPUTS = 2;\n    const TENSOR_SHAPE = 0;\n    const TENSOR_TYPE = 1;\n    const TENSOR_QUANTIZATION = 4;\n    const QUANTIZATION_SCALE = 2;\n\n    /** Position of field `index` of the table at `table`, 0 when it is absent. */\n    const field = (table: number, index: number): number => {\n        const vtable = table - model.readInt32LE(table);\n        const offset = 4 + 2 * index;\n        if (offset >= model.readUInt16LE(vtable)) {\n            return 0;\n        }\n        const at = model.readUInt16LE(vtable + offset);\n        return at ? table + at : 0;\n    };\n    /** Follow the offset stored in field `index` to the table, vector or string it points at. */\n    const ref = (table: number, index: number): number => {\n        const at = field(table, index);\n        return at ? at + model.readUInt32LE(at) : 0;\n    };\n    const length = (vector: number): number => (vector ? model.readUInt32LE(vector) : 0);\n    /** Element `k` of a vector of tables. */\n    const tableAt = (vector: number, k: number): number => {\n        const at = vector + 4 + 4 * k;\n        return at + model.readUInt32LE(at);\n    };\n    const intAt = (vector: number, k: number): number => model.readInt32LE(vector + 4 + 4 * k);\n\n    const tensor = (subgraph: number, list: number): TfliteTensor | null => {\n        const tensors = ref(subgraph, SUBGRAPH_TENSORS);\n        if (!length(list)) {\n            return null;\n        }\n        const index = intAt(list, 0);\n        if (index < 0 || index >= length(tensors)) {\n            return null;\n        }\n        const t = tableAt(tensors, index);\n        const dims = ref(t, TENSOR_SHAPE);\n        if (length(dims) > MAX_DIMENSIONS) {\n            return null;\n        }\n        const shape = Array.from({ length: length(dims) }, (_, k) => intAt(dims, k));\n        const typeAt = field(t, TENSOR_TYPE);\n        const quantization = ref(t, TENSOR_QUANTIZATION);\n        const scales = quantization ? ref(quantization, QUANTIZATION_SCALE) : 0;\n        return {\n            shape,\n            type: typeAt ? model.readUInt8(typeAt) : 0,\n            quantized: length(scales) > 0 && model.readFloatLE(scales + 4) !== 0,\n        };\n    };\n\n    if (!isTfliteModel(model)) {\n        return null;\n    }\n    try {\n        const subgraphs = ref(model.readUInt32LE(0), MODEL_SUBGRAPHS);\n        if (!length(subgraphs)) {\n            return null;\n        }\n        const main = tableAt(subgraphs, 0);\n        const input = tensor(main, ref(main, SUBGRAPH_INPUTS));\n        const output = tensor(main, ref(main, SUBGRAPH_OUTPUTS));\n        return input && output ? { input, output } : null;\n    } catch {\n        return null; // an offset pointed outside the file\n    }\n}\n\n/**\n * Why the device cannot run this model, or `null` when it can. Mirrors the checks of the firmware's\n * `MicroWakeWord::FromConfig`: input `[1, stride, 40]`, quantized input and output. An openWakeWord\n * model is named as such — it is the most likely mix-up, both come as `.tflite`, and the device runs\n * openWakeWord only when its firmware was started in that mode (then without the built-in models).\n *\n * @param model the `.tflite` file\n */\nexport function microModelProblem(model: Buffer): string | null {\n    if (!isTfliteModel(model)) {\n        return 'the model file is not a TensorFlow Lite model (.tflite)';\n    }\n    const io = readTfliteIo(model);\n    if (!io) {\n        return 'the model file is damaged — it could not be read as a TensorFlow Lite model';\n    }\n    const shape = io.input.shape;\n    if (shape[shape.length - 1] === OPEN_WAKE_WORD_FEATURES) {\n        return 'this is an openWakeWord model — the device runs only microWakeWord models, so the wake word has to be trained with microWakeWord';\n    }\n    if (shape.length !== 3 || shape[2] !== MICRO_FEATURES) {\n        return `this is not a microWakeWord model (input [${shape.join(', ')}], a microWakeWord model takes [1, n, ${MICRO_FEATURES}])`;\n    }\n    if (!io.input.quantized || !io.output.quantized) {\n        return 'the model is not quantized — the device runs only quantized microWakeWord models';\n    }\n    return null;\n}\n\nfunction parseConfig(config: WakeWordUpload['config']): Record<string, unknown> {\n    if (config === undefined || config === null || config === '') {\n        return {};\n    }\n    if (typeof config === 'object') {\n        return config;\n    }\n    let parsed: unknown;\n    try {\n        parsed = JSON.parse(config);\n    } catch (e) {\n        throw new Error(`the JSON file is not valid JSON: ${(e as Error).message}`);\n    }\n    if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) {\n        throw new Error('the JSON file must contain an object');\n    }\n    return parsed as Record<string, unknown>;\n}\n\nfunction languageList(value: unknown): string[] {\n    return Array.isArray(value) ? [...new Set(value.map(v => String(v).trim().toLowerCase()).filter(Boolean))] : [];\n}\n\n/**\n * Validate an upload and build the wake word the devices will download. Throws with a message meant for\n * the user when the files cannot work on the device.\n *\n * @param upload what was uploaded\n * @param defaultLanguage language to record when neither the upload nor its JSON names one\n */\nexport function prepareWakeWord(upload: WakeWordUpload, defaultLanguage = ''): CustomWakeWord {\n    const model = upload.model;\n    if (!model?.length) {\n        throw new Error('no model file (.tflite) was uploaded');\n    }\n    if (model.length > MAX_MODEL_BYTES) {\n        throw new Error(\n            `the model is ${Math.round(model.length / 1024)} KB — microWakeWord models are well below ${MAX_MODEL_BYTES / 1024} KB`,\n        );\n    }\n    const problem = microModelProblem(model);\n    if (problem) {\n        throw new Error(problem);\n    }\n\n    const source = parseConfig(upload.config);\n    const type = typeof source.type === 'string' ? source.type : '';\n    if (type && type !== 'micro') {\n        throw new Error(\n            `this is a \"${type}\" model — the device only accepts microWakeWord models (\"micro\") from outside`,\n        );\n    }\n\n    const fileName = upload.fileName || (typeof source.model === 'string' ? source.model : '');\n    const phrase = (\n        upload.phrase?.trim() ||\n        (typeof source.wake_word === 'string' ? source.wake_word.trim() : '') ||\n        phraseFromFileName(fileName)\n    ).slice(0, 64);\n    if (!phrase) {\n        throw new Error('the wake word needs a phrase (what you say to wake the device)');\n    }\n\n    let languages = languageList(upload.languages);\n    if (!languages.length) {\n        languages = languageList(source.trained_languages);\n    }\n    if (!languages.length && defaultLanguage) {\n        languages = [defaultLanguage.toLowerCase()];\n    }\n\n    const sha256 = createHash('sha256').update(model).digest('hex');\n    const id = `${slugify(fileName) || slugify(phrase) || 'wake_word'}_${sha256.slice(0, 8)}`;\n\n    const micro = source.micro && typeof source.micro === 'object' ? (source.micro as Record<string, unknown>) : {};\n    const config = {\n        ...source,\n        type: 'micro',\n        wake_word: phrase,\n        model: `${id}.tflite`,\n        trained_languages: languages,\n        version: typeof source.version === 'number' ? source.version : 2,\n        micro: { ...DEFAULT_MICRO, ...micro },\n    };\n\n    return {\n        id,\n        phrase,\n        languages,\n        config: Buffer.from(`${JSON.stringify(config, null, 2)}\\n`),\n        model,\n        size: model.length,\n        sha256,\n    };\n}\n\n/**\n * Rebuild a stored wake word from its two files (adapter start). The JSON is the one we wrote, so it is\n * trusted for the phrase; size and hash are recomputed from the model rather than stored.\n */\nexport function loadWakeWord(id: string, config: Buffer, model: Buffer): CustomWakeWord {\n    if (!WAKE_WORD_ID_RE.test(id)) {\n        throw new Error(`invalid wake word id \"${id}\"`);\n    }\n    if (!isTfliteModel(model)) {\n        throw new Error(`${id}.tflite is not a TensorFlow Lite model`);\n    }\n    const parsed = parseConfig(config.toString('utf8'));\n    return {\n        id,\n        phrase: typeof parsed.wake_word === 'string' ? parsed.wake_word : id,\n        languages: languageList(parsed.trained_languages),\n        config,\n        model,\n        size: model.length,\n        sha256: createHash('sha256').update(model).digest('hex'),\n    };\n}\n"]}

package/build/lib/voice/wyoming.js

@@ -225,10 +225,13 @@ class WyomingConnection {
             if (!text) {
                 return;
             }
-            const answer = (await this.opts.answer(text, sttMs)).trim();
+            const audioMs = Math.round((pcm.length / 2 / this.audioRate) * 1000);
+            const answer = (await this.opts.answer(text, sttMs, audioMs)).trim();
             this.opts.log.info(`Wyoming A: ${answer}`);
-            if (answer) {
-                await this.speak(answer);
+            const spoken = answer ? await this.speak(answer) : null;
+            if (spoken) {
+                // Only a reply counts — a bare `synthesize` from the client is no request of ours.
+                this.opts.onSpoken?.(spoken);
             }
         }
         catch (e) {
@@ -236,22 +239,27 @@ class WyomingConnection {
             this.write('error', { text: e.message });
         }
     }
+    /** Synthesise and stream `text`; returns how long synthesis took and how long the audio plays. */
     async speak(text) {
         if (!text) {
-            return;
+            return null;
         }
         try {
+            const ttsStart = Date.now();
             const { pcm, sampleRate } = await this.opts.tts.synthesize(text, this.lang);
+            const spoken = { ms: Date.now() - ttsStart, audioMs: (pcm.length / 2 / sampleRate) * 1000 };
             const meta = { rate: sampleRate, width: 2, channels: 1 };
             this.write('audio-start', meta);
             for (let off = 0; off < pcm.length; off += AUDIO_CHUNK) {
                 this.write('audio-chunk', meta, pcm.subarray(off, off + AUDIO_CHUNK));
             }
             this.write('audio-stop', {});
+            return spoken;
         }
         catch (e) {
             this.opts.log.error(`Wyoming TTS failed: ${e.message}`);
             this.write('error', { text: e.message });
+            return null;
         }
     }
     /** Capabilities advertised in response to `describe`. */

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

@@ -1 +1 @@
-{"version":3,"file":"wyoming.js","sourceRoot":"","sources":["../../../src/lib/voice/wyoming.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAAA;;;;;;;;;;;GAWG;AACH,8CAAgC;AAGhC,6CAAqC;AAsBrC,6CAA6C;AAC7C,MAAM,WAAW,GAAG,IAAI,CAAC;AAEzB,MAAa,aAAa;IAIO;IAHrB,MAAM,GAAsB,IAAI,CAAC;IACxB,KAAK,GAAG,IAAI,GAAG,EAAc,CAAC;IAE/C,YAA6B,IAA0B;QAA1B,SAAI,GAAJ,IAAI,CAAsB;IAAG,CAAC;IAE3D,KAAK;QACD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YACnC,MAAM,MAAM,GAAG,GAAG,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC;YACjE,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;YACrB,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC7B,MAAM,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACnF,MAAM,IAAI,GACN,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,IAAI,CAAC,IAAI,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;YACrG,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,GAAG,EAAE;gBACrC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,mCAAmC,IAAI,IAAI,SAAS,IAAI,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;gBAC7F,OAAO,EAAE,CAAC;YACd,CAAC,CAAC,CAAC;QACP,CAAC,CAAC,CAAC;IACP,CAAC;IAED,KAAK,CAAC,IAAI;QACN,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACzB,CAAC,CAAC,OAAO,EAAE,CAAC;QAChB,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAC3B,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC;QACnB,IAAI,MAAM,EAAE,CAAC;YACT,MAAM,IAAI,OAAO,CAAO,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QAC9D,CAAC;IACL,CAAC;IAEO,YAAY,CAAC,IAAgB;QACjC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACrB,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QAChD,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QAChD,IAAI,iBAAiB,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC;CACJ;AAvCD,sCAuCC;AAED,4DAA4D;AAC5D,MAAM,iBAAiB;IASE;IACA;IATb,GAAG,GAAG,oBAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACtB,QAAQ,GAAG,KAAK,CAAC;IACjB,WAAW,GAAa,EAAE,CAAC;IAC3B,SAAS,GAAG,KAAK,CAAC;IAC1B,0FAA0F;IAClF,IAAI,CAAS;IAErB,YACqB,IAAgB,EAChB,IAA0B;QAD1B,SAAI,GAAJ,IAAI,CAAY;QAChB,SAAI,GAAJ,IAAI,CAAsB;QAE3C,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC1B,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE;YAChB,IAAI,CAAC,GAAG,GAAG,oBAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;YACxC,KAAK,IAAI,CAAC,KAAK,EAAE,CAAC;QACtB,CAAC,CAAC,CAAC;IACP,CAAC;IAED,uFAAuF;IAC/E,KAAK,CAAC,KAAK;QACf,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAChB,OAAO;QACX,CAAC;QACD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACrB,IAAI,CAAC;YACD,SAAS,CAAC;gBACN,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO;gBAC1C,IAAI,EAAE,GAAG,CAAC,EAAE,CAAC;oBACT,MAAM;gBACV,CAAC;gBACD,IAAI,MAAkF,CAAC;gBACvF,IAAI,CAAC;oBACD,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;gBACnE,CAAC;gBAAC,MAAM,CAAC;oBACL,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,kBAAkB;oBACxD,SAAS;gBACb,CAAC;gBACD,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,cAAc,IAAI,CAAC,CAAC,CAAC;gBACtD,MAAM,KAAK,GAAG,EAAE,GAAG,CAAC,GAAG,UAAU,CAAC;gBAClC,IAAI,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,KAAK,EAAE,CAAC;oBAC1B,MAAM,CAAC,mCAAmC;gBAC9C,CAAC;gBACD,MAAM,OAAO,GAAG,UAAU,CAAC,CAAC,CAAC,oBAAM,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBACvF,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;gBACpC,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACvF,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,6BAA8B,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC5E,CAAC;gBAAS,CAAC;YACP,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC;QAC1B,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,IAAY,EAAE,IAA8B,EAAE,OAAgB;QACxE,MAAM,MAAM,GAA4B,EAAE,IAAI,EAAE,CAAC;QACjD,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,IAAI,OAAO,EAAE,CAAC;YACV,MAAM,CAAC,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAC/C,IAAI,OAAO,EAAE,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,MAAM,CAAC,EAAW;QAC5B,QAAQ,EAAE,CAAC,IAAI,EAAE,CAAC;YACd,KAAK,UAAU;gBACX,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;gBAChC,MAAM;YACV,kFAAkF;YAClF,KAAK,YAAY,CAAC;YAClB,KAAK,eAAe,CAAC;YACrB,KAAK,aAAa;gBACd,IAAI,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC;oBACpB,IAAI,CAAC,IAAI;wBACL,OAAO,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;4BAChC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC;4BAClC,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,QAAmB,CAAC,QAAQ,EAAE,CAAC;gBACtD,CAAC;gBACD,IAAI,EAAE,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC;oBAChB,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC;gBAC5D,CAAC;gBACD,IAAI,CAAC,WAAW,GAAG,EAAE,CAAC;gBACtB,MAAM;YACV,KAAK,aAAa;gBACd,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC;oBACb,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,IAAI,EAAE,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC;wBAC5C,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC;oBAC5D,CAAC;oBACD,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC;gBACtC,CAAC;gBACD,MAAM;YACV,yEAAyE;YACzE,KAAK,YAAY,CAAC;YAClB,KAAK,eAAe;gBAChB,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;gBACzB,MAAM;YACV,KAAK,YAAY;gBACb,IAAI,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC;oBACpB,IAAI,CAAC,IAAI;wBACL,OAAO,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;4BAChC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC;4BAClC,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,QAAmB,CAAC,QAAQ,EAAE,CAAC;gBACtD,CAAC;gBACD,MAAM,IAAI,CAAC,KAAK,CACZ,EAAE,CAAC,IAAI,EAAE,IAAI;oBACT,CAAC,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,IAAI,KAAK,QAAQ;wBAC/B,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC;wBAC9B,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,IAAe,CAAC,QAAQ,EAAE;oBACzC,CAAC,CAAC,EAAE,CACX,CAAC;gBACF,MAAM;YACV,KAAK,MAAM;gBACP,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBAC5B,MAAM;QACd,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,WAAW;QACrB,MAAM,GAAG,GAAG,oBAAM,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC5C,IAAI,CAAC,WAAW,GAAG,EAAE,CAAC;QACtB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC;YACd,OAAO;QACX,CAAC;QACD,IAAI,CAAC;YACD,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;YAC1E,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC5B,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;YACnF,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,CAAC;YACpC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,cAAc,IAAI,IAAI,SAAS,EAAE,CAAC,CAAC;YACtD,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC;YACnC,IAAI,CAAC,IAAI,EAAE,CAAC;gBACR,OAAO;YACX,CAAC;YACD,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YAC5D,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,cAAc,MAAM,EAAE,CAAC,CAAC;YAC3C,IAAI,MAAM,EAAE,CAAC;gBACT,MAAM,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YAC7B,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,4BAA6B,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YACxE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,IAAI,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QACxD,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,KAAK,CAAC,IAAY;QAC5B,IAAI,CAAC,IAAI,EAAE,CAAC;YACR,OAAO;QACX,CAAC;QACD,IAAI,CAAC;YACD,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;YAC5E,MAAM,IAAI,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;YACzD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;YAChC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,GAAG,CAAC,MAAM,EAAE,GAAG,IAAI,WAAW,EAAE,CAAC;gBACrD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,WAAW,CAAC,CAAC,CAAC;YAC1E,CAAC;YACD,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;QACjC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,uBAAwB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YACnE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,IAAI,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QACxD,CAAC;IACL,CAAC;IAED,yDAAyD;IACjD,IAAI;QACR,MAAM,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACxD,MAAM,WAAW,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,gDAAgD,EAAE,CAAC;QAChG,OAAO;YACH,GAAG,EAAE;gBACD;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,mCAAmC;oBAChD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;YACD,GAAG,EAAE;gBACD;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,mCAAmC;oBAChD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;YACD,MAAM,EAAE;gBACJ;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,oCAAoC;oBACjD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;SACJ,CAAC;IACN,CAAC;CACJ","sourcesContent":["/**\n * Wyoming protocol endpoint (https://github.com/rhasspy/wyoming) — a TCP server that bridges Wyoming\n * events to the existing STT → answer → TTS pipeline, so a `wyoming-satellite`, Home Assistant Voice PE\n * or ESPHome voice device can stream to this adapter directly.\n *\n * Wire format: one JSON header line terminated by '\\n', optionally followed by `payload_length` raw\n * bytes (audio). We handle: describe → info; audio-start/chunk/stop → STT → answer → TTS (streamed back\n * as audio-*); synthesize → TTS.\n *\n * ⚠️ Framing is unit-tested (loopback); interop with real HA Voice PE / wyoming-satellite still needs\n * on-device validation.\n */\nimport * as net from 'node:net';\nimport type { SttEngine } from './stt';\nimport type { TtsEngine } from './tts';\nimport { Buffer } from 'node:buffer';\n\nexport interface WyomingServerOptions {\n    port: number;\n    bindAddress?: string;\n    /** ISO-639-1 language hint for STT/TTS. */\n    language: string;\n    stt: SttEngine;\n    tts: TtsEngine;\n    /** `sttMs`: how long recognition took (for the journal). */\n    answer: (question: string, sttMs: number) => Promise<string>;\n    /** Optional STT vocabulary hints (device/room names) to bias recognition; re-read per utterance. */\n    getHints?: () => string[] | Promise<string[]>;\n    log: ioBroker.Logger;\n}\n\ninterface WyEvent {\n    type: string;\n    data?: Record<string, unknown>;\n    payload?: Buffer;\n}\n\n/** Bytes of PCM per outgoing audio-chunk. */\nconst AUDIO_CHUNK = 8192;\n\nexport class WyomingServer {\n    private server: net.Server | null = null;\n    private readonly conns = new Set<net.Socket>();\n\n    constructor(private readonly opts: WyomingServerOptions) {}\n\n    start(): Promise<void> {\n        return new Promise((resolve, reject) => {\n            const server = net.createServer(sock => this.onConnection(sock));\n            this.server = server;\n            server.once('error', reject);\n            server.on('error', e => this.opts.log.error(`Wyoming server error: ${e.message}`));\n            const addr =\n                this.opts.bindAddress && this.opts.bindAddress !== '0.0.0.0' ? this.opts.bindAddress : undefined;\n            server.listen(this.opts.port, addr, () => {\n                this.opts.log.info(`Wyoming server listening on TCP ${addr || '0.0.0.0'}:${this.opts.port}`);\n                resolve();\n            });\n        });\n    }\n\n    async stop(): Promise<void> {\n        for (const s of this.conns) {\n            s.destroy();\n        }\n        this.conns.clear();\n        const server = this.server;\n        this.server = null;\n        if (server) {\n            await new Promise<void>(res => server.close(() => res()));\n        }\n    }\n\n    private onConnection(sock: net.Socket): void {\n        this.conns.add(sock);\n        sock.on('close', () => this.conns.delete(sock));\n        sock.on('error', () => this.conns.delete(sock));\n        new WyomingConnection(sock, this.opts);\n    }\n}\n\n/** One client connection: frame parser + event handling. */\nclass WyomingConnection {\n    private buf = Buffer.alloc(0);\n    private draining = false;\n    private audioChunks: Buffer[] = [];\n    private audioRate = 16000;\n    /** Per-utterance language (overridable via `transcribe`/`synthesize` `data.language`). */\n    private lang: string;\n\n    constructor(\n        private readonly sock: net.Socket,\n        private readonly opts: WyomingServerOptions,\n    ) {\n        this.lang = opts.language;\n        sock.on('data', d => {\n            this.buf = Buffer.concat([this.buf, d]);\n            void this.drain();\n        });\n    }\n\n    /** Pull complete events (header line + optional payload) out of the receive buffer. */\n    private async drain(): Promise<void> {\n        if (this.draining) {\n            return;\n        }\n        this.draining = true;\n        try {\n            for (;;) {\n                const nl = this.buf.indexOf(0x0a); // '\\n'\n                if (nl < 0) {\n                    break;\n                }\n                let header: { type?: string; data?: Record<string, unknown>; payload_length?: number };\n                try {\n                    header = JSON.parse(this.buf.subarray(0, nl).toString('utf8'));\n                } catch {\n                    this.buf = this.buf.subarray(nl + 1); // skip a bad line\n                    continue;\n                }\n                const payloadLen = Number(header.payload_length || 0);\n                const total = nl + 1 + payloadLen;\n                if (this.buf.length < total) {\n                    break; // wait for the rest of the payload\n                }\n                const payload = payloadLen ? Buffer.from(this.buf.subarray(nl + 1, total)) : undefined;\n                this.buf = this.buf.subarray(total);\n                await this.handle({ type: String(header.type || ''), data: header.data, payload });\n            }\n        } catch (e) {\n            this.opts.log.warn(`Wyoming connection error: ${(e as Error).message}`);\n        } finally {\n            this.draining = false;\n        }\n    }\n\n    private write(type: string, data?: Record<string, unknown>, payload?: Buffer): void {\n        const header: Record<string, unknown> = { type };\n        if (data !== undefined) {\n            header.data = data;\n        }\n        if (payload) {\n            header.payload_length = payload.length;\n        }\n        this.sock.write(`${JSON.stringify(header)}\\n`);\n        if (payload) {\n            this.sock.write(payload);\n        }\n    }\n\n    private async handle(ev: WyEvent): Promise<void> {\n        switch (ev.type) {\n            case 'describe':\n                this.write('info', this.info());\n                break;\n            // ASR request / start-of-speech (VAD) — reset the buffer, optional language hint.\n            case 'transcribe':\n            case 'voice-started':\n            case 'audio-start':\n                if (ev.data?.language) {\n                    this.lang =\n                        typeof ev.data.language === 'object'\n                            ? JSON.stringify(ev.data.language)\n                            : (ev.data.language as string).toString();\n                }\n                if (ev.data?.rate) {\n                    this.audioRate = Number(ev.data.rate) || this.audioRate;\n                }\n                this.audioChunks = [];\n                break;\n            case 'audio-chunk':\n                if (ev.payload) {\n                    if (!this.audioChunks.length && ev.data?.rate) {\n                        this.audioRate = Number(ev.data.rate) || this.audioRate;\n                    }\n                    this.audioChunks.push(ev.payload);\n                }\n                break;\n            // End-of-speech — either the explicit audio-stop or a VAD voice-stopped.\n            case 'audio-stop':\n            case 'voice-stopped':\n                await this.onUtterance();\n                break;\n            case 'synthesize':\n                if (ev.data?.language) {\n                    this.lang =\n                        typeof ev.data.language === 'object'\n                            ? JSON.stringify(ev.data.language)\n                            : (ev.data.language as string).toString();\n                }\n                await this.speak(\n                    ev.data?.text\n                        ? typeof ev.data?.text === 'object'\n                            ? JSON.stringify(ev.data.text)\n                            : (ev.data.text as string).toString()\n                        : '',\n                );\n                break;\n            case 'ping':\n                this.write('pong', ev.data);\n                break;\n        }\n    }\n\n    private async onUtterance(): Promise<void> {\n        const pcm = Buffer.concat(this.audioChunks);\n        this.audioChunks = [];\n        if (!pcm.length) {\n            return;\n        }\n        try {\n            const hints = this.opts.getHints ? await this.opts.getHints() : undefined;\n            const sttStart = Date.now();\n            const text = await this.opts.stt.transcribe(pcm, this.audioRate, this.lang, hints);\n            const sttMs = Date.now() - sttStart;\n            this.opts.log.info(`Wyoming Q: ${text || '(empty)'}`);\n            this.write('transcript', { text });\n            if (!text) {\n                return;\n            }\n            const answer = (await this.opts.answer(text, sttMs)).trim();\n            this.opts.log.info(`Wyoming A: ${answer}`);\n            if (answer) {\n                await this.speak(answer);\n            }\n        } catch (e) {\n            this.opts.log.error(`Wyoming pipeline failed: ${(e as Error).message}`);\n            this.write('error', { text: (e as Error).message });\n        }\n    }\n\n    private async speak(text: string): Promise<void> {\n        if (!text) {\n            return;\n        }\n        try {\n            const { pcm, sampleRate } = await this.opts.tts.synthesize(text, this.lang);\n            const meta = { rate: sampleRate, width: 2, channels: 1 };\n            this.write('audio-start', meta);\n            for (let off = 0; off < pcm.length; off += AUDIO_CHUNK) {\n                this.write('audio-chunk', meta, pcm.subarray(off, off + AUDIO_CHUNK));\n            }\n            this.write('audio-stop', {});\n        } catch (e) {\n            this.opts.log.error(`Wyoming TTS failed: ${(e as Error).message}`);\n            this.write('error', { text: (e as Error).message });\n        }\n    }\n\n    /** Capabilities advertised in response to `describe`. */\n    private info(): Record<string, unknown> {\n        const lang = (this.opts.language || 'en').split('-')[0];\n        const attribution = { name: 'ioBroker', url: 'https://github.com/ioBroker/ioBroker.assistant' };\n        return {\n            asr: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant speech-to-text',\n                    version: '1.0',\n                    models: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n            tts: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant text-to-speech',\n                    version: '1.0',\n                    voices: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n            handle: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant intent handling',\n                    version: '1.0',\n                    models: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n        };\n    }\n}\n"]}
+{"version":3,"file":"wyoming.js","sourceRoot":"","sources":["../../../src/lib/voice/wyoming.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAAA;;;;;;;;;;;GAWG;AACH,8CAAgC;AAGhC,6CAAqC;AAwBrC,6CAA6C;AAC7C,MAAM,WAAW,GAAG,IAAI,CAAC;AAEzB,MAAa,aAAa;IAIO;IAHrB,MAAM,GAAsB,IAAI,CAAC;IACxB,KAAK,GAAG,IAAI,GAAG,EAAc,CAAC;IAE/C,YAA6B,IAA0B;QAA1B,SAAI,GAAJ,IAAI,CAAsB;IAAG,CAAC;IAE3D,KAAK;QACD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YACnC,MAAM,MAAM,GAAG,GAAG,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC;YACjE,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;YACrB,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC7B,MAAM,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YACnF,MAAM,IAAI,GACN,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,IAAI,CAAC,IAAI,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;YACrG,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,GAAG,EAAE;gBACrC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,mCAAmC,IAAI,IAAI,SAAS,IAAI,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;gBAC7F,OAAO,EAAE,CAAC;YACd,CAAC,CAAC,CAAC;QACP,CAAC,CAAC,CAAC;IACP,CAAC;IAED,KAAK,CAAC,IAAI;QACN,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACzB,CAAC,CAAC,OAAO,EAAE,CAAC;QAChB,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAC3B,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC;QACnB,IAAI,MAAM,EAAE,CAAC;YACT,MAAM,IAAI,OAAO,CAAO,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QAC9D,CAAC;IACL,CAAC;IAEO,YAAY,CAAC,IAAgB;QACjC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACrB,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QAChD,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;QAChD,IAAI,iBAAiB,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC;CACJ;AAvCD,sCAuCC;AAED,4DAA4D;AAC5D,MAAM,iBAAiB;IASE;IACA;IATb,GAAG,GAAG,oBAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACtB,QAAQ,GAAG,KAAK,CAAC;IACjB,WAAW,GAAa,EAAE,CAAC;IAC3B,SAAS,GAAG,KAAK,CAAC;IAC1B,0FAA0F;IAClF,IAAI,CAAS;IAErB,YACqB,IAAgB,EAChB,IAA0B;QAD1B,SAAI,GAAJ,IAAI,CAAY;QAChB,SAAI,GAAJ,IAAI,CAAsB;QAE3C,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC1B,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE;YAChB,IAAI,CAAC,GAAG,GAAG,oBAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;YACxC,KAAK,IAAI,CAAC,KAAK,EAAE,CAAC;QACtB,CAAC,CAAC,CAAC;IACP,CAAC;IAED,uFAAuF;IAC/E,KAAK,CAAC,KAAK;QACf,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAChB,OAAO;QACX,CAAC;QACD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACrB,IAAI,CAAC;YACD,SAAS,CAAC;gBACN,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO;gBAC1C,IAAI,EAAE,GAAG,CAAC,EAAE,CAAC;oBACT,MAAM;gBACV,CAAC;gBACD,IAAI,MAAkF,CAAC;gBACvF,IAAI,CAAC;oBACD,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;gBACnE,CAAC;gBAAC,MAAM,CAAC;oBACL,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,kBAAkB;oBACxD,SAAS;gBACb,CAAC;gBACD,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,cAAc,IAAI,CAAC,CAAC,CAAC;gBACtD,MAAM,KAAK,GAAG,EAAE,GAAG,CAAC,GAAG,UAAU,CAAC;gBAClC,IAAI,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,KAAK,EAAE,CAAC;oBAC1B,MAAM,CAAC,mCAAmC;gBAC9C,CAAC;gBACD,MAAM,OAAO,GAAG,UAAU,CAAC,CAAC,CAAC,oBAAM,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBACvF,IAAI,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;gBACpC,MAAM,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YACvF,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,6BAA8B,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC5E,CAAC;gBAAS,CAAC;YACP,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC;QAC1B,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,IAAY,EAAE,IAA8B,EAAE,OAAgB;QACxE,MAAM,MAAM,GAA4B,EAAE,IAAI,EAAE,CAAC;QACjD,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,IAAI,OAAO,EAAE,CAAC;YACV,MAAM,CAAC,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAC/C,IAAI,OAAO,EAAE,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,MAAM,CAAC,EAAW;QAC5B,QAAQ,EAAE,CAAC,IAAI,EAAE,CAAC;YACd,KAAK,UAAU;gBACX,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;gBAChC,MAAM;YACV,kFAAkF;YAClF,KAAK,YAAY,CAAC;YAClB,KAAK,eAAe,CAAC;YACrB,KAAK,aAAa;gBACd,IAAI,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC;oBACpB,IAAI,CAAC,IAAI;wBACL,OAAO,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;4BAChC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC;4BAClC,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,QAAmB,CAAC,QAAQ,EAAE,CAAC;gBACtD,CAAC;gBACD,IAAI,EAAE,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC;oBAChB,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC;gBAC5D,CAAC;gBACD,IAAI,CAAC,WAAW,GAAG,EAAE,CAAC;gBACtB,MAAM;YACV,KAAK,aAAa;gBACd,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC;oBACb,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,IAAI,EAAE,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC;wBAC5C,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC;oBAC5D,CAAC;oBACD,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC;gBACtC,CAAC;gBACD,MAAM;YACV,yEAAyE;YACzE,KAAK,YAAY,CAAC;YAClB,KAAK,eAAe;gBAChB,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;gBACzB,MAAM;YACV,KAAK,YAAY;gBACb,IAAI,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC;oBACpB,IAAI,CAAC,IAAI;wBACL,OAAO,EAAE,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;4BAChC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC;4BAClC,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,QAAmB,CAAC,QAAQ,EAAE,CAAC;gBACtD,CAAC;gBACD,MAAM,IAAI,CAAC,KAAK,CACZ,EAAE,CAAC,IAAI,EAAE,IAAI;oBACT,CAAC,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,IAAI,KAAK,QAAQ;wBAC/B,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC;wBAC9B,CAAC,CAAE,EAAE,CAAC,IAAI,CAAC,IAAe,CAAC,QAAQ,EAAE;oBACzC,CAAC,CAAC,EAAE,CACX,CAAC;gBACF,MAAM;YACV,KAAK,MAAM;gBACP,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC;gBAC5B,MAAM;QACd,CAAC;IACL,CAAC;IAEO,KAAK,CAAC,WAAW;QACrB,MAAM,GAAG,GAAG,oBAAM,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC5C,IAAI,CAAC,WAAW,GAAG,EAAE,CAAC;QACtB,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC;YACd,OAAO;QACX,CAAC;QACD,IAAI,CAAC;YACD,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;YAC1E,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC5B,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;YACnF,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,CAAC;YACpC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,cAAc,IAAI,IAAI,SAAS,EAAE,CAAC,CAAC;YACtD,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC;YACnC,IAAI,CAAC,IAAI,EAAE,CAAC;gBACR,OAAO;YACX,CAAC;YACD,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC;YACrE,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;YACrE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,cAAc,MAAM,EAAE,CAAC,CAAC;YAC3C,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YACxD,IAAI,MAAM,EAAE,CAAC;gBACT,mFAAmF;gBACnF,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC;YACjC,CAAC;QACL,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,4BAA6B,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YACxE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,IAAI,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QACxD,CAAC;IACL,CAAC;IAED,kGAAkG;IAC1F,KAAK,CAAC,KAAK,CAAC,IAAY;QAC5B,IAAI,CAAC,IAAI,EAAE,CAAC;YACR,OAAO,IAAI,CAAC;QAChB,CAAC;QACD,IAAI,CAAC;YACD,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC5B,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;YAC5E,MAAM,MAAM,GAAG,EAAE,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,GAAG,UAAU,CAAC,GAAG,IAAI,EAAE,CAAC;YAC5F,MAAM,IAAI,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;YACzD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;YAChC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,GAAG,CAAC,MAAM,EAAE,GAAG,IAAI,WAAW,EAAE,CAAC;gBACrD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,GAAG,WAAW,CAAC,CAAC,CAAC;YAC1E,CAAC;YACD,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;YAC7B,OAAO,MAAM,CAAC;QAClB,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACT,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,uBAAwB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YACnE,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,IAAI,EAAG,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;YACpD,OAAO,IAAI,CAAC;QAChB,CAAC;IACL,CAAC;IAED,yDAAyD;IACjD,IAAI;QACR,MAAM,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACxD,MAAM,WAAW,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,gDAAgD,EAAE,CAAC;QAChG,OAAO;YACH,GAAG,EAAE;gBACD;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,mCAAmC;oBAChD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;YACD,GAAG,EAAE;gBACD;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,mCAAmC;oBAChD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;YACD,MAAM,EAAE;gBACJ;oBACI,IAAI,EAAE,oBAAoB;oBAC1B,WAAW;oBACX,SAAS,EAAE,IAAI;oBACf,WAAW,EAAE,oCAAoC;oBACjD,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE;wBACJ;4BACI,IAAI,EAAE,SAAS;4BACf,WAAW;4BACX,SAAS,EAAE,IAAI;4BACf,WAAW,EAAE,EAAE;4BACf,OAAO,EAAE,KAAK;4BACd,SAAS,EAAE,CAAC,IAAI,CAAC;yBACpB;qBACJ;iBACJ;aACJ;SACJ,CAAC;IACN,CAAC;CACJ","sourcesContent":["/**\n * Wyoming protocol endpoint (https://github.com/rhasspy/wyoming) — a TCP server that bridges Wyoming\n * events to the existing STT → answer → TTS pipeline, so a `wyoming-satellite`, Home Assistant Voice PE\n * or ESPHome voice device can stream to this adapter directly.\n *\n * Wire format: one JSON header line terminated by '\\n', optionally followed by `payload_length` raw\n * bytes (audio). We handle: describe → info; audio-start/chunk/stop → STT → answer → TTS (streamed back\n * as audio-*); synthesize → TTS.\n *\n * ⚠️ Framing is unit-tested (loopback); interop with real HA Voice PE / wyoming-satellite still needs\n * on-device validation.\n */\nimport * as net from 'node:net';\nimport type { SttEngine } from './stt';\nimport type { TtsEngine } from './tts';\nimport { Buffer } from 'node:buffer';\n\nexport interface WyomingServerOptions {\n    port: number;\n    bindAddress?: string;\n    /** ISO-639-1 language hint for STT/TTS. */\n    language: string;\n    stt: SttEngine;\n    tts: TtsEngine;\n    /** `sttMs`: how long recognition took, `audioMs`: length of the recording (both for the journal). */\n    answer: (question: string, sttMs: number, audioMs: number) => Promise<string>;\n    /** The reply was synthesised: how long that took and how long it plays (for the request journal). */\n    onSpoken?: (info: { ms: number; audioMs: number }) => void;\n    /** Optional STT vocabulary hints (device/room names) to bias recognition; re-read per utterance. */\n    getHints?: () => string[] | Promise<string[]>;\n    log: ioBroker.Logger;\n}\n\ninterface WyEvent {\n    type: string;\n    data?: Record<string, unknown>;\n    payload?: Buffer;\n}\n\n/** Bytes of PCM per outgoing audio-chunk. */\nconst AUDIO_CHUNK = 8192;\n\nexport class WyomingServer {\n    private server: net.Server | null = null;\n    private readonly conns = new Set<net.Socket>();\n\n    constructor(private readonly opts: WyomingServerOptions) {}\n\n    start(): Promise<void> {\n        return new Promise((resolve, reject) => {\n            const server = net.createServer(sock => this.onConnection(sock));\n            this.server = server;\n            server.once('error', reject);\n            server.on('error', e => this.opts.log.error(`Wyoming server error: ${e.message}`));\n            const addr =\n                this.opts.bindAddress && this.opts.bindAddress !== '0.0.0.0' ? this.opts.bindAddress : undefined;\n            server.listen(this.opts.port, addr, () => {\n                this.opts.log.info(`Wyoming server listening on TCP ${addr || '0.0.0.0'}:${this.opts.port}`);\n                resolve();\n            });\n        });\n    }\n\n    async stop(): Promise<void> {\n        for (const s of this.conns) {\n            s.destroy();\n        }\n        this.conns.clear();\n        const server = this.server;\n        this.server = null;\n        if (server) {\n            await new Promise<void>(res => server.close(() => res()));\n        }\n    }\n\n    private onConnection(sock: net.Socket): void {\n        this.conns.add(sock);\n        sock.on('close', () => this.conns.delete(sock));\n        sock.on('error', () => this.conns.delete(sock));\n        new WyomingConnection(sock, this.opts);\n    }\n}\n\n/** One client connection: frame parser + event handling. */\nclass WyomingConnection {\n    private buf = Buffer.alloc(0);\n    private draining = false;\n    private audioChunks: Buffer[] = [];\n    private audioRate = 16000;\n    /** Per-utterance language (overridable via `transcribe`/`synthesize` `data.language`). */\n    private lang: string;\n\n    constructor(\n        private readonly sock: net.Socket,\n        private readonly opts: WyomingServerOptions,\n    ) {\n        this.lang = opts.language;\n        sock.on('data', d => {\n            this.buf = Buffer.concat([this.buf, d]);\n            void this.drain();\n        });\n    }\n\n    /** Pull complete events (header line + optional payload) out of the receive buffer. */\n    private async drain(): Promise<void> {\n        if (this.draining) {\n            return;\n        }\n        this.draining = true;\n        try {\n            for (;;) {\n                const nl = this.buf.indexOf(0x0a); // '\\n'\n                if (nl < 0) {\n                    break;\n                }\n                let header: { type?: string; data?: Record<string, unknown>; payload_length?: number };\n                try {\n                    header = JSON.parse(this.buf.subarray(0, nl).toString('utf8'));\n                } catch {\n                    this.buf = this.buf.subarray(nl + 1); // skip a bad line\n                    continue;\n                }\n                const payloadLen = Number(header.payload_length || 0);\n                const total = nl + 1 + payloadLen;\n                if (this.buf.length < total) {\n                    break; // wait for the rest of the payload\n                }\n                const payload = payloadLen ? Buffer.from(this.buf.subarray(nl + 1, total)) : undefined;\n                this.buf = this.buf.subarray(total);\n                await this.handle({ type: String(header.type || ''), data: header.data, payload });\n            }\n        } catch (e) {\n            this.opts.log.warn(`Wyoming connection error: ${(e as Error).message}`);\n        } finally {\n            this.draining = false;\n        }\n    }\n\n    private write(type: string, data?: Record<string, unknown>, payload?: Buffer): void {\n        const header: Record<string, unknown> = { type };\n        if (data !== undefined) {\n            header.data = data;\n        }\n        if (payload) {\n            header.payload_length = payload.length;\n        }\n        this.sock.write(`${JSON.stringify(header)}\\n`);\n        if (payload) {\n            this.sock.write(payload);\n        }\n    }\n\n    private async handle(ev: WyEvent): Promise<void> {\n        switch (ev.type) {\n            case 'describe':\n                this.write('info', this.info());\n                break;\n            // ASR request / start-of-speech (VAD) — reset the buffer, optional language hint.\n            case 'transcribe':\n            case 'voice-started':\n            case 'audio-start':\n                if (ev.data?.language) {\n                    this.lang =\n                        typeof ev.data.language === 'object'\n                            ? JSON.stringify(ev.data.language)\n                            : (ev.data.language as string).toString();\n                }\n                if (ev.data?.rate) {\n                    this.audioRate = Number(ev.data.rate) || this.audioRate;\n                }\n                this.audioChunks = [];\n                break;\n            case 'audio-chunk':\n                if (ev.payload) {\n                    if (!this.audioChunks.length && ev.data?.rate) {\n                        this.audioRate = Number(ev.data.rate) || this.audioRate;\n                    }\n                    this.audioChunks.push(ev.payload);\n                }\n                break;\n            // End-of-speech — either the explicit audio-stop or a VAD voice-stopped.\n            case 'audio-stop':\n            case 'voice-stopped':\n                await this.onUtterance();\n                break;\n            case 'synthesize':\n                if (ev.data?.language) {\n                    this.lang =\n                        typeof ev.data.language === 'object'\n                            ? JSON.stringify(ev.data.language)\n                            : (ev.data.language as string).toString();\n                }\n                await this.speak(\n                    ev.data?.text\n                        ? typeof ev.data?.text === 'object'\n                            ? JSON.stringify(ev.data.text)\n                            : (ev.data.text as string).toString()\n                        : '',\n                );\n                break;\n            case 'ping':\n                this.write('pong', ev.data);\n                break;\n        }\n    }\n\n    private async onUtterance(): Promise<void> {\n        const pcm = Buffer.concat(this.audioChunks);\n        this.audioChunks = [];\n        if (!pcm.length) {\n            return;\n        }\n        try {\n            const hints = this.opts.getHints ? await this.opts.getHints() : undefined;\n            const sttStart = Date.now();\n            const text = await this.opts.stt.transcribe(pcm, this.audioRate, this.lang, hints);\n            const sttMs = Date.now() - sttStart;\n            this.opts.log.info(`Wyoming Q: ${text || '(empty)'}`);\n            this.write('transcript', { text });\n            if (!text) {\n                return;\n            }\n            const audioMs = Math.round((pcm.length / 2 / this.audioRate) * 1000);\n            const answer = (await this.opts.answer(text, sttMs, audioMs)).trim();\n            this.opts.log.info(`Wyoming A: ${answer}`);\n            const spoken = answer ? await this.speak(answer) : null;\n            if (spoken) {\n                // Only a reply counts — a bare `synthesize` from the client is no request of ours.\n                this.opts.onSpoken?.(spoken);\n            }\n        } catch (e) {\n            this.opts.log.error(`Wyoming pipeline failed: ${(e as Error).message}`);\n            this.write('error', { text: (e as Error).message });\n        }\n    }\n\n    /** Synthesise and stream `text`; returns how long synthesis took and how long the audio plays. */\n    private async speak(text: string): Promise<{ ms: number; audioMs: number } | null> {\n        if (!text) {\n            return null;\n        }\n        try {\n            const ttsStart = Date.now();\n            const { pcm, sampleRate } = await this.opts.tts.synthesize(text, this.lang);\n            const spoken = { ms: Date.now() - ttsStart, audioMs: (pcm.length / 2 / sampleRate) * 1000 };\n            const meta = { rate: sampleRate, width: 2, channels: 1 };\n            this.write('audio-start', meta);\n            for (let off = 0; off < pcm.length; off += AUDIO_CHUNK) {\n                this.write('audio-chunk', meta, pcm.subarray(off, off + AUDIO_CHUNK));\n            }\n            this.write('audio-stop', {});\n            return spoken;\n        } catch (e) {\n            this.opts.log.error(`Wyoming TTS failed: ${(e as Error).message}`);\n            this.write('error', { text: (e as Error).message });\n            return null;\n        }\n    }\n\n    /** Capabilities advertised in response to `describe`. */\n    private info(): Record<string, unknown> {\n        const lang = (this.opts.language || 'en').split('-')[0];\n        const attribution = { name: 'ioBroker', url: 'https://github.com/ioBroker/ioBroker.assistant' };\n        return {\n            asr: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant speech-to-text',\n                    version: '1.0',\n                    models: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n            tts: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant text-to-speech',\n                    version: '1.0',\n                    voices: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n            handle: [\n                {\n                    name: 'iobroker-assistant',\n                    attribution,\n                    installed: true,\n                    description: 'ioBroker.assistant intent handling',\n                    version: '1.0',\n                    models: [\n                        {\n                            name: 'default',\n                            attribution,\n                            installed: true,\n                            description: '',\n                            version: '1.0',\n                            languages: [lang],\n                        },\n                    ],\n                },\n            ],\n        };\n    }\n}\n"]}