@lumiastream/wakeword 1.1.9 → 1.2.0

package/lib/voice.js

@@ -48,6 +48,7 @@ function unpacked(p) {
 }
 
 const UNKNOWN_TOKEN = "[unk]";
+const UNKNOWN_TOKEN_NORMALIZED = "unk";
 const normalizePhrase = (phrase = "") => phrase.trim().toLowerCase();
 const toBool = (v = "") =>
 	["1", "true", "yes", "y"].includes(`${v}`.trim().toLowerCase());
@@ -61,6 +62,23 @@ const tokenize = (phrase = "") =>
 		.filter(Boolean);
 const tokensEqual = (a = [], b = []) =>
 	a.length === b.length && a.every((token, idx) => token === b[idx]);
+const trimUnknownBoundaryTokens = (tokens = []) => {
+	let start = 0;
+	let end = tokens.length;
+	while (
+		start < end &&
+		(tokens[start] === UNKNOWN_TOKEN_NORMALIZED || tokens[start] === UNKNOWN_TOKEN)
+	) {
+		start += 1;
+	}
+	while (
+		end > start &&
+		(tokens[end - 1] === UNKNOWN_TOKEN_NORMALIZED || tokens[end - 1] === UNKNOWN_TOKEN)
+	) {
+		end -= 1;
+	}
+	return tokens.slice(start, end);
+};
 const tokensContainSequence = (tokens = [], phraseTokens = []) => {
 	if (!phraseTokens.length || tokens.length < phraseTokens.length) return false;
 	for (let i = 0; i <= tokens.length - phraseTokens.length; i += 1) {
@@ -283,8 +301,12 @@ function handle(processedWord, averageConfidence, originalText, options = {}) {
 
 	const normalizedProcessed = normalizePhrase(processedWord);
 	const normalizedOriginal = normalizePhrase(originalText);
-	const processedTokens = tokenize(normalizedProcessed);
-	const originalTokens = tokenize(normalizedOriginal);
+	const processedTokens = trimUnknownBoundaryTokens(
+		tokenize(normalizedProcessed),
+	);
+	const originalTokens = trimUnknownBoundaryTokens(
+		tokenize(normalizedOriginal),
+	);
 	const matches = new Set();
 	const confidentCommands = new Set();
 
@@ -292,10 +314,16 @@ function handle(processedWord, averageConfidence, originalText, options = {}) {
 		if (!tokens?.length) return;
 		const hits = MATCH_SENTENCE
 			? allowedCommands.filter((command) =>
-					tokensContainSequence(tokens, tokenize(command)),
+					tokensContainSequence(
+						tokens,
+						trimUnknownBoundaryTokens(tokenize(command)),
+					),
 				)
 			: allowedCommands.filter((command) =>
-					tokensEqual(tokens, tokenize(command)),
+					tokensEqual(
+						tokens,
+						trimUnknownBoundaryTokens(tokenize(command)),
+					),
 				);
 		hits.forEach((hit) => matches.add(hit));
 	};
@@ -303,7 +331,7 @@ function handle(processedWord, averageConfidence, originalText, options = {}) {
 	// Only allow sentence matches for commands that were confidently recognized.
 	if (normalizedProcessed) {
 		COMMANDS.forEach((command) => {
-			const commandTokens = tokenize(command);
+			const commandTokens = trimUnknownBoundaryTokens(tokenize(command));
 			const isMatch = MATCH_SENTENCE
 				? tokensContainSequence(processedTokens, commandTokens)
 				: tokensEqual(processedTokens, commandTokens);

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lumiastream/wakeword",
-	"version": "1.1.9",
+	"version": "1.2.0",
 	"type": "module",
 	"main": "lib/index.js",
 	"files": [