services(azure): chunk audio so interruptions work better

CHANGELOG.md

@@ -5,6 +5,15 @@ All notable changes to **pipecat** will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Fixed
+
+- Fixed an issue in `AzureTTSService` that was causing interruptions to take
+  longer. Azure Speech Synthesizer returns big chunks of audio. Currently, we
+  don't have a way to stop a long `AudioRawFrame`, so we have to chunk longer
+  audio into small chunks.
+
 ## [0.0.29] - 2024-06-07
 
 ### Added

src/pipecat/services/azure.py

@@ -67,7 +67,13 @@ async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
         if result.reason == ResultReason.SynthesizingAudioCompleted:
             await self.stop_ttfb_metrics()
             # Azure always sends a 44-byte header. Strip it off.
-            yield AudioRawFrame(audio=result.audio_data[44:], sample_rate=16000, num_channels=1)
+            audio = result.audio_data[44:]
+            # Chunk it to 20ms so interruptions work properly
+            bytes_frames = 640
+            while len(audio) > 0:
+                chunk = audio[:bytes_frames]
+                yield AudioRawFrame(audio=chunk, sample_rate=16000, num_channels=1)
+                audio = audio[bytes_frames:]
         elif result.reason == ResultReason.Canceled:
             cancellation_details = result.cancellation_details
             logger.warning(f"Speech synthesis canceled: {cancellation_details.reason}")