fix(sherpa): sanitize Qwen ASR JSON control characters

extraResources/sherpa/README.md

@@ -3,8 +3,10 @@
 封装 JS 复制自 npm `sherpa-onnx-node@1.13.2`（Apache-2.0）。
 
 - `vendor/addon.js` 已替换为自定义加载器：从环境变量 `SHERPA_ONNX_LIB_DIR`
-  用 `process.dlopen` 加载 `sherpa-onnx.node`（其余 `vendor/*.js` 原样保留，
-  它们都经 `require('./addon.js')` 取原生模块）。
+  用 `process.dlopen` 加载 `sherpa-onnx.node`；vendor 文件都经
+  `require('./addon.js')` 取原生模块。
+- `vendor/json-result.js` 由 `non-streaming-asr.js` 使用，处理原生识别结果中
+  的未转义控制字符；其余 `vendor/*.js` 原样保留。
 - `vendor/addon-static-import.js` 在自定义加载器下不再被引用（保留以便升级对照）。
 
 原生库**不在此处**，按需下载到 `userData/sherpa-onnx/current/`

extraResources/sherpa/vendor/json-result.js

@@ -0,0 +1,74 @@
+'use strict';
+
+// Native sherpa results are JSON strings. Some Qwen outputs may contain raw
+// C0 control characters inside the transcript text, which JSON.parse rejects.
+function sanitizeJsonControlCharacters(jsonStr) {
+  let sanitized = '';
+  let inString = false;
+  let escaped = false;
+
+  for (const char of String(jsonStr)) {
+    const code = char.charCodeAt(0);
+
+    if (inString) {
+      if (escaped) {
+        sanitized += char;
+        escaped = false;
+        continue;
+      }
+      if (char === '\\') {
+        sanitized += char;
+        escaped = true;
+        continue;
+      }
+      if (char === '"') {
+        sanitized += char;
+        inString = false;
+        continue;
+      }
+      if (code <= 0x1f) {
+        sanitized += escapeControlCharacter(char);
+        continue;
+      }
+      sanitized += char;
+      continue;
+    }
+
+    if (char === '"') {
+      inString = true;
+      sanitized += char;
+    } else if (code > 0x1f || char === '\t' || char === '\n' || char === '\r') {
+      // Tabs, line feeds, and carriage returns are valid JSON whitespace
+      // outside strings. Other C0 controls are never valid JSON tokens.
+      sanitized += char;
+    }
+  }
+
+  return sanitized;
+}
+
+function escapeControlCharacter(char) {
+  switch (char) {
+    case '\b':
+      return '\\b';
+    case '\f':
+      return '\\f';
+    case '\n':
+      return '\\n';
+    case '\r':
+      return '\\r';
+    case '\t':
+      return '\\t';
+    default:
+      return `\\u${char.charCodeAt(0).toString(16).padStart(4, '0')}`;
+  }
+}
+
+function parseJsonResult(jsonStr) {
+  return JSON.parse(sanitizeJsonControlCharacters(jsonStr));
+}
+
+module.exports = {
+  parseJsonResult,
+  sanitizeJsonControlCharacters,
+};

extraResources/sherpa/vendor/non-streaming-asr.js

@@ -8,6 +8,7 @@
  */
 
 const addon = require('./addon.js');
+const { parseJsonResult } = require('./json-result.js');
 
 /**
  * Internal symbol to mark async-created recognizers.
@@ -127,7 +128,7 @@ class OfflineRecognizer {
       this.handle,
       stream.handle,
     );
-    return JSON.parse(jsonStr);
+    return parseJsonResult(jsonStr);
   }
 
   /**
@@ -137,7 +138,7 @@ class OfflineRecognizer {
    */
   getResult(stream) {
     const jsonStr = addon.getOfflineStreamResultAsJson(stream.handle);
-    return JSON.parse(jsonStr);
+    return parseJsonResult(jsonStr);
   }
 }
 

package.json

@@ -38,6 +38,7 @@
     "test:custom-languages": "tsc scripts/test-custom-languages.ts --outDir node_modules/.cache/custom-language-tests --module commonjs --moduleResolution node --target es2019 --esModuleInterop --skipLibCheck && node node_modules/.cache/custom-language-tests/scripts/test-custom-languages.js",
     "test:refine": "tsc scripts/test-refine-units.ts --outDir node_modules/.cache/refine-tests --module commonjs --moduleResolution node --target es2019 --esModuleInterop --skipLibCheck --resolveJsonModule && node node_modules/.cache/refine-tests/scripts/test-refine-units.js",
     "test:speaker-diarization": "node scripts/test-speaker-diarization-config.cjs && tsc scripts/test-speaker-diarization.ts --outDir node_modules/.cache/speaker-diarization-tests --module commonjs --moduleResolution node --target es2019 --esModuleInterop --skipLibCheck && node node_modules/.cache/speaker-diarization-tests/scripts/test-speaker-diarization.js",
+    "test:sherpa-json": "node scripts/test-sherpa-json-result.cjs",
     "test:proofread-speakers": "tsc scripts/test-proofread-speakers.ts --outDir node_modules/.cache/proofread-speaker-tests --module commonjs --moduleResolution node --target es2019 --esModuleInterop --skipLibCheck && node node_modules/.cache/proofread-speaker-tests/scripts/test-proofread-speakers.js",
     "test:manuscript": "tsc scripts/test-manuscript-matching.ts --outDir node_modules/.cache/manuscript-tests --module commonjs --moduleResolution node --target es2022 --esModuleInterop --skipLibCheck --resolveJsonModule && node node_modules/.cache/manuscript-tests/scripts/test-manuscript-matching.js",
     "longgap:gen": "tsc scripts/longgap/gen-audio.ts --outDir node_modules/.cache/longgap --module commonjs --moduleResolution node --target es2019 --esModuleInterop --skipLibCheck --resolveJsonModule && node node_modules/.cache/longgap/scripts/longgap/gen-audio.js",

scripts/test-sherpa-json-result.cjs

@@ -0,0 +1,26 @@
+'use strict';
+
+const assert = require('assert');
+const {
+  parseJsonResult,
+  sanitizeJsonControlCharacters,
+} = require('../extraResources/sherpa/vendor/json-result.js');
+
+const rawResult = '{"text":"first\nsecond\t\u0000end","tokens":["first"]}';
+assert.deepStrictEqual(parseJsonResult(rawResult), {
+  text: 'first\nsecond\t\u0000end',
+  tokens: ['first'],
+});
+
+const escapedResult = String.raw`{"text":"first\nsecond\t"}`;
+assert.deepStrictEqual(parseJsonResult(escapedResult), {
+  text: 'first\nsecond\t',
+});
+
+assert.strictEqual(parseJsonResult('{"text":"ok"}\u0000\n').text, 'ok');
+assert.strictEqual(
+  sanitizeJsonControlCharacters('{"text":"a\u000bb"}'),
+  '{"text":"a\\u000bb"}',
+);
+
+console.log('sherpa JSON result tests: 4 passed');