fix: split ssml to correct chunks (#225)

* fix: split ssml to correct chunks

* fix: split ssml to correct chunks

* fixed: eslint

* fixed: eslint

* fixed: add comment to testcase

* fixed: review comments

* fixed: review comments

* fixed: review comments

* fixed: review comments

* fixed: review comments

Co-authored-by: Quan HL <quanluuhoang8@gmail.com>

lib/tasks/say.js

@@ -1,96 +1,26 @@
 const Task = require('./task');
 const {TaskName, TaskPreconditions} = require('../utils/constants');
+const pollySSMLSplit = require('polly-ssml-split');
 
 const breakLengthyTextIfNeeded = (logger,  text) => {
   const chunkSize = 1000;
-  if (text.length <= chunkSize) return [text];
-
-  const result = [];
   const isSSML = text.startsWith('<speak>');
-  let startPos = 0;
+  if (text.length <= chunkSize || !isSSML) return [text];
-  let charPos = isSSML ? 7 : 0;  // skip <speak>
+  const options = {
-  let tag;
+    // MIN length
-  //logger.debug({isSSML}, `breakLengthyTextIfNeeded: handling text of length ${text.length}`);
+    softLimit: 100,
-  while (startPos + charPos < text.length) {
+    // MAX length, exclude 15 characters <speak></speak>
-    if (isSSML && !tag && text[startPos + charPos] === '<') {
+    hardLimit: chunkSize - 15,
-      const tagStartPos = ++charPos;
+    // Set of extra split characters (Optional property)
-      while (startPos + charPos < text.length) {
+    extraSplitChars: ',;!?',
-        if (text[startPos + charPos] === '>') {
+  };
-          if (text[startPos + charPos - 1] === '\\') tag = null;
+  pollySSMLSplit.configure(options);
-          else if (!tag) tag = text.substring(startPos + tagStartPos, startPos + charPos - 1);
+  try {
-          break;
+    return pollySSMLSplit.split(text);
-        }
+  } catch (err) {
-        if (!tag) {
+    logger.info({err}, 'Error spliting SSML long text');
-          const c = text[startPos + charPos];
+    return [text];
-          if (c === ' ') {
-            tag = text.substring(startPos + tagStartPos, startPos + charPos);
-            //logger.debug(`breakLengthyTextIfNeeded: enter tag ${tag} (space)`);
-            break;
-          }
-        }
-        charPos++;
-      }
-      if (tag) {
-        //search for end of tag
-        //logger.debug(`breakLengthyTextIfNeeded: searching forward for </${tag}>`);
-        const e1 = text.indexOf(`</${tag}>`, startPos + charPos);
-        const e2 = text.indexOf('/>', startPos + charPos);
-        const tagEndPos = e1 === -1 ? e2 : e2 === -1 ? e1 : Math.min(e1, e2);
-        if (tagEndPos === -1) {
-          //logger.debug(`breakLengthyTextIfNeeded: exit tag ${tag} not found, exiting`);
-        } else {
-          //logger.debug(`breakLengthyTextIfNeeded: exit tag ${tag} found at ${tagEndPos}`);
-          charPos = tagEndPos + 1;
-        }
-        tag = null;
-      }
-      continue;
-    }
-
-    if (charPos < chunkSize) {
-      charPos++;
-      continue;
-    }
-
-    // start looking for a good break point
-    let chunkIt = false;
-    const a = text[startPos + charPos];
-    const b = text[startPos + charPos + 1];
-    if (/[\.!\?]/.test(a) && /\s/.test(b)) {
-      //logger.debug('breakLengthyTextIfNeeded: breaking at sentence end');
-      chunkIt = true;
-    }
-    if (chunkIt) {
-      charPos++;
-      const chunk = text.substr(startPos, charPos);
-      if (isSSML) {
-        result.push(0 === startPos ? `${chunk}</speak>` : `<speak>${chunk}</speak>`);
-      }
-      else result.push(chunk);
-      charPos = 0;
-      startPos += chunk.length;
-
-      //logger.debug({chunk: result[result.length - 1]},
-      //  `breakLengthyTextIfNeeded: chunked; new starting pos ${startPos}`);
-
-    }
-    else charPos++;
   }
-
-  // final chunk
-  if (startPos < text.length) {
-    const chunk = text.substr(startPos);
-    if (isSSML) {
-      result.push(0 === startPos ? `${chunk}</speak>` : `<speak>${chunk}`);
-    }
-    else result.push(chunk);
-
-    //logger.debug({chunk: result[result.length - 1]},
-    //  `breakLengthyTextIfNeeded: final chunk; starting pos ${startPos} length ${chunk.length}`);
-
-  }
-
-  return result;
 };
 
 class TaskSay extends Task {

package-lock.json

@@ -34,6 +34,7 @@
         "moment": "^2.29.4",
         "parse-url": "^8.1.0",
         "pino": "^6.14.0",
+        "polly-ssml-split": "^0.1.0",
         "sdp-transform": "^2.14.1",
         "short-uuid": "^4.2.0",
         "to-snake-case": "^1.0.0",
@@ -5956,6 +5957,19 @@
         "node": ">=8"
       }
     },
+    "node_modules/polly-ssml-split": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/polly-ssml-split/-/polly-ssml-split-0.1.0.tgz",
+      "integrity": "sha512-vweYqyWC4WwUZPh8cywLeYpj5IswgAXhc+twq8Y6inqFo32JU8YlAZtFmHPhdI456gh3bSwupLaL+6WV9CQuUw==",
+      "dependencies": {
+        "polly-text-split": "^0.1.4"
+      }
+    },
+    "node_modules/polly-text-split": {
+      "version": "0.1.4",
+      "resolved": "https://registry.npmjs.org/polly-text-split/-/polly-text-split-0.1.4.tgz",
+      "integrity": "sha512-WhYm13sQyPxdn5yWpGi45WFWZOruKBqs+y0iXWVz16y+yV612WjOwqvh4s1j7CgWbid+8rbjiHjxvZwJE1zVFw=="
+    },
     "node_modules/prelude-ls": {
       "version": "1.2.1",
       "resolved": "https://registry.npmjs.org/prelude-ls/-/prelude-ls-1.2.1.tgz",
@@ -12345,6 +12359,19 @@
         "find-up": "^4.0.0"
       }
     },
+    "polly-ssml-split": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/polly-ssml-split/-/polly-ssml-split-0.1.0.tgz",
+      "integrity": "sha512-vweYqyWC4WwUZPh8cywLeYpj5IswgAXhc+twq8Y6inqFo32JU8YlAZtFmHPhdI456gh3bSwupLaL+6WV9CQuUw==",
+      "requires": {
+        "polly-text-split": "^0.1.4"
+      }
+    },
+    "polly-text-split": {
+      "version": "0.1.4",
+      "resolved": "https://registry.npmjs.org/polly-text-split/-/polly-text-split-0.1.4.tgz",
+      "integrity": "sha512-WhYm13sQyPxdn5yWpGi45WFWZOruKBqs+y0iXWVz16y+yV612WjOwqvh4s1j7CgWbid+8rbjiHjxvZwJE1zVFw=="
+    },
     "prelude-ls": {
       "version": "1.2.1",
       "resolved": "https://registry.npmjs.org/prelude-ls/-/prelude-ls-1.2.1.tgz",

package.json

@@ -56,7 +56,8 @@
     "uuid-random": "^1.3.2",
     "verify-aws-sns-signature": "^0.1.0",
     "ws": "^8.9.0",
-    "xml2js": "^0.4.23"
+    "xml2js": "^0.4.23",
+    "polly-ssml-split": "^0.1.0"
   },
   "devDependencies": {
     "clear-module": "^4.1.2",

test/data/bad/bad-say-ssml.json

@@ -0,0 +1,9 @@
+{
+  "say": {
+    "text": "<speak>I already told you <emphasis level=\"strong\">I already told you I already told you I already told you I already told you! I already told you I already told you I already told you I already told you? I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you told I already told you I already told you told I already told you I already told you. I already told you <break time=\"3s\"/> I really like that person!</emphasis> this is another long text.</speak>",
+    "synthesizer": {
+      "vendor": "google",
+      "language": "en-US"
+    }
+  }
+}

test/data/good/say-ssml.json

@@ -0,0 +1,9 @@
+{
+  "say": {
+    "text": "<speak>I already told you I already told you I already told you I already told you I already told you! I already told you I already told you I already told you I already told you? I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you I already told you told I already told you I already told you told I already told you I already told you. I already told you <break time=\"3s\"/> I <emphasis level=\"strong\">really like that person!</emphasis> this is another long text.</speak>",
+    "synthesizer": {
+      "vendor": "google",
+      "language": "en-US"
+    }
+  }
+}

test/unit-tests.js

@@ -44,7 +44,22 @@ test('unit tests', (t) => {
 
   task = makeTask(logger, require('./data/good/say-text-array'));
   t.ok(task.name === 'say', 'parsed say with multiple segments');
+  
+  task = makeTask(logger, require('./data/good/say-ssml'));
+  // the ssml is more than 1000 chars, 
+  // expecting first chunk is length > 100, stop at ? instead of first .
+  // 2nd chunk is long text < 1000 char, stop at .
+  // 3rd chunk is the rest.
+  t.ok(task.text.length === 3 &&
+    task.text[0].length === 187 &&
+    task.text[1].length === 882 &&
+    task.text[2].length === 123, 'parsed say');
 
+  task = makeTask(logger, require('./data/bad/bad-say-ssml'));
+  t.ok(task.text.length === 1 &&
+    task.text[0].length === 1162, 'parsed bad say');
+
+  
   const alt = require('./data/good/alternate-syntax');
   const normalize = require('../lib/utils/normalize-jambones');
   normalize(logger, alt).forEach((t) => {