@lumiastream/wakeword 1.1.5-alpha.2 → 1.1.5-alpha.4

package/lib/voice.js

@@ -5,6 +5,14 @@ import { fileURLToPath } from "node:url";
 import { existsSync, chmodSync } from "node:fs";
 import readline from "node:readline";
 
+/* ------------------------------------------------------------------ */
+/* 0. Helpers                                                         */
+/* ------------------------------------------------------------------ */
+const here = dirname(fileURLToPath(import.meta.url));
+
+const ASAR_TOKEN = "app.asar";
+const ASAR_UNPACKED_TOKEN = "app.asar.unpacked";
+
 // Ensure native libs can load from app.asar.unpacked when packaged
 const maybeUnpackedPath = (libPath) => {
 	if (typeof libPath !== "string") return libPath;
@@ -33,14 +41,6 @@ koffi.load = (libPath, ...rest) => {
 
 const { Model, Recognizer, setLogLevel } = await import("vosk-koffi");
 
-/* ------------------------------------------------------------------ */
-/* 0. Helpers                                                         */
-/* ------------------------------------------------------------------ */
-const here = dirname(fileURLToPath(import.meta.url));
-
-const ASAR_TOKEN = "app.asar";
-const ASAR_UNPACKED_TOKEN = "app.asar.unpacked";
-
 function unpacked(p) {
 	if (!p || !p.includes(ASAR_TOKEN)) return p;
 	if (p.includes(ASAR_UNPACKED_TOKEN)) return p;
@@ -156,6 +156,9 @@ const LOG_PARTIAL =
 	["1", "true", "yes"].includes(
 		(process.env.WAKEWORD_LOG_PARTIAL || "").toLowerCase()
 	);
+let LOG_FINAL = ["1", "true", "yes"].includes(
+	(process.env.WAKEWORD_LOG_FINAL || "").toLowerCase()
+);
 let lastLevelLog = 0;
 
 function logAudioLevel(buf) {
@@ -222,25 +225,49 @@ mic.on("data", (buf) => {
 function handle(processedWord, averageConfidence, originalText) {
 	if (!processedWord && !originalText) return;
 
+	const finalSentence =
+		typeof originalText === "string" && originalText.trim()
+			? originalText.trim()
+			: (processedWord ?? "").toString().trim();
+	if (LOG_FINAL && finalSentence) {
+		process.stdout?.write(`final|${finalSentence}\n`);
+	}
+
 	const normalizedProcessed = normalizePhrase(processedWord);
 	const normalizedOriginal = normalizePhrase(originalText);
 	const matches = new Set();
+	const confidentCommands = new Set();
 
-	const findMatches = (text) => {
+	const findMatches = (text, allowedCommands = COMMANDS) => {
 		if (!text || text.includes(UNKNOWN_TOKEN)) return;
 		const hits = MATCH_SENTENCE
-			? COMMANDS.filter((command) => text.includes(command))
-			: COMMANDS.filter((command) => text === command);
+			? allowedCommands.filter((command) => text.includes(command))
+			: allowedCommands.filter((command) => text === command);
 		hits.forEach((hit) => matches.add(hit));
 	};
 
-	// Try the filtered text first, then fall back to the raw sentence for sentence matching
+	// Only allow sentence matches for commands that were confidently recognized.
+	if (normalizedProcessed) {
+		COMMANDS.forEach((command) => {
+			const isMatch = MATCH_SENTENCE
+				? normalizedProcessed.includes(command)
+				: normalizedProcessed === command;
+			if (isMatch) {
+				confidentCommands.add(command);
+			}
+		});
+	}
+
+	// Try the filtered text first, then fall back to the raw sentence only for confident commands.
 	findMatches(normalizedProcessed);
-	findMatches(normalizedOriginal);
+	findMatches(normalizedOriginal, [...confidentCommands]);
 
 	if (!matches.size) return;
 
 	matches.forEach((match) => {
+		if (finalSentence) {
+			process.stdout?.write(`sentence|${finalSentence}\n`);
+		}
 		process.stdout?.write(`voice|${match}\n`);
 		process.stdout?.write(`confidence|${averageConfidence}\n`);
 	});
@@ -252,13 +279,21 @@ const rl = readline.createInterface({ input: process.stdin, terminal: false });
 
 rl.on("line", (line) => {
 	const trimmed = line.trim();
-	if (!trimmed.startsWith("update,") && !trimmed.startsWith("confidence,"))
+	if (
+		!trimmed.startsWith("update,") &&
+		!trimmed.startsWith("confidence,") &&
+		!trimmed.startsWith("debug,")
+	)
 		return;
 
 	if (trimmed.startsWith("confidence,")) {
 		WORD_CONFIDENCE_THRESHOLD = Number(trimmed.split(",")[1]);
 		return;
 	}
+	if (trimmed.startsWith("debug,")) {
+		LOG_FINAL = toBool(trimmed.split(",")[1]);
+		return;
+	}
 
 	const phrases = trimmed
 		.split(",")

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lumiastream/wakeword",
-	"version": "1.1.5-alpha.2",
+	"version": "1.1.5-alpha.4",
 	"type": "module",
 	"main": "lib/index.js",
 	"files": [