diff --git a/packages/text-to-speech/src/htmlToSsml.ts b/packages/text-to-speech/src/htmlToSsml.ts
index 2a1417d34..ef55b09b1 100644
--- a/packages/text-to-speech/src/htmlToSsml.ts
+++ b/packages/text-to-speech/src/htmlToSsml.ts
@@ -55,8 +55,7 @@ const ANCHOR_ELEMENTS_BLOCKED_ATTRIBUTES = [
 
 function ssmlTagsForTopLevelElement() {
   return {
-    opening: `<p>`,
-    closing: `</p>`,
+    opening: `<break />`,
   }
 }
 
@@ -180,10 +179,6 @@ function emitElement(
     }
   }
 
-  if (isTopLevel) {
-    emit(textItems, topLevelTags.closing)
-  }
-
   return Number(maxVisitedIdx)
 }
 
diff --git a/packages/text-to-speech/src/textToSpeech.ts b/packages/text-to-speech/src/textToSpeech.ts
index b60fdfeda..5b35aae48 100644
--- a/packages/text-to-speech/src/textToSpeech.ts
+++ b/packages/text-to-speech/src/textToSpeech.ts
@@ -138,7 +138,7 @@ export const synthesizeTextToSpeech = async (
       }
     }
     // for ssml
-    const audioData: Buffer = Buffer.from([])
+    let audioData: Buffer = Buffer.from([])
     // split ssml into chunks of 200 characters to stream faster
     const ssmlChunks = input.text.match(/.{1,200}/g)
     // TODO: not split on words and bookmarks
@@ -151,7 +151,7 @@ export const synthesizeTextToSpeech = async (
         }
         timeOffset = timeOffset + result.audioDuration
         wordOffset = wordOffset + ssmlChunk.length
-        Buffer.concat([audioData, Buffer.from(result.audioData)])
+        audioData = Buffer.concat([audioData, Buffer.from(result.audioData)])
       }
     }