To address the issue of distorted voice at the end of sentences with the Deepgram text-to-speech model, here are some steps you can take: • Check TTS Configuration: Ensure the model, encoding, and sample rate are properly configured. For example:`tts = deepgram.TTS( model="aura-2-andromeda-en", encoding="linear16", sample_rate=24000 )` • Use SSML Tags: Add pauses or adjust prosody at sentence boundaries using SSML. For example:`text = "Hello, how are you?" await session.say(text)` • Handle Text Chunking: Ensure that text is processed correctly to avoid issues at sentence boundaries. Avoid unnecessary splitting of text into chunks. • Test with Custom Pronunciation: If specific words cause distortion, use SSML or phoneme tags to define pronunciation. • Verify Audio Handling: Ensure the audio stream is properly buffered and handled on the client side. If the issue persists, could you provide more details such as: • Are you using a specific Deepgram TTS model? • Does the distortion occur consistently with certain sentences? • Are you applying any custom text processing or SSML tags? • Have you tried different audio encodings or sample rates? This information will help narrow down the cause and suggest more targeted solutions. Sources: livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/tts.py | Agent speech and audio | LiveKit Docs | Agent speech and audio | LiveKit Docs | Turn detection and interruptions | LiveKit Docs