teh tts cannot say ids like this easily either breaks the audio or not speaking clearly # ex: agent_12345678-1234-1234-1234-123456789012 # ex: call_12345678-1234-1234-1234-123456789012

Thanks for reaching out, Nitish. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

they are in the text, s for the ocnversation some dynamic variable values are reaplced

I'm here to help you with the TTS issues you're experiencing with dynamic variable values like

agent_12345678-1234-1234-1234-123456789012

. A great way to improve how these are read aloud is by preprocessing the text before sending it to the TTS engine. Here are some friendly approaches you might find useful: 1. Use SSML for Character-by-Character Pronunciation: You can wrap the identifier in SSML tags like

<say-as interpret-as="characters">

so it’s read one character at a time. For example:

Your agent ID is <say-as interpret-as="characters">agent_12345678-1234-1234-1234-123456789012</say-as>.

2. Replace with Human-Readable Phrasing: Another friendly option is to use a callback function that swaps the identifier for a more natural, verbal-friendly phrase. For instance:

def before_tts_cb(agent, source):
    if isinstance(source, str):
        return source.replace("agent_12345678-1234-1234-1234-123456789012", "agent, followed by one two three four five six seven eight, dash, one two three four, dash...")
    return source

3. Use Phonetic Pronunciation Tags: If your TTS provider supports phonemes, you can gently guide how the text is pronounced by using phonetic spelling. 4. Leverage the

tts_node

in v1.0+: This feature lets you modify text or audio with more flexibility, which can really help with tricky pronunciations. These methods can really make a difference in clarity and help avoid those frustrating mispronunciations. Please feel free to reach out if you'd like me to walk you through implementing any of these solutions—I’m happy to assist! Sources: Agents v0.x migration guide - Node.js | LiveKit Docs | Agents 0.x migration guide | LiveKit Docs | VoicePipelineAgent | LiveKit Docs

def tts_node(self, text_stream: AsyncIterable[str], model_settings): """ Override TTS node to automatically transform ID-like tokens based on provider type. This method creates a copy of each text chunk and transforms only the copy, ensuring that transcripts remain clean while speech gets proper pronunciation. Args: text_stream: Async stream of text to be synthesized model_settings: TTS model settings Returns: Async stream of audio frames """ # Detect provider using utility function prov = detect_provider_heuristically(self, model_settings) async def transformed_stream() -> AsyncGenerator[str, None]: """Transform copies of text chunks to handle ID-like tokens appropriately.""" async for original_chunk in text_stream: if original_chunk: # Create a copy of the chunk to avoid modifying the original chunk_copy = str(original_chunk) # Transform the copy based on provider capabilities transformed_chunk = transform_chunk_for_provider(chunk_copy, prov) yield transformed_chunk else: yield original_chunk # Delegate to parent's default TTS node behavior with transformed stream return super().tts_node(transformed_stream(), model_settings) i tried this way, but the issue i faced in the transcription it's also vaible these tags

I don't have the answer you're looking for. You could also try asking your question: • in one of the other Slack channels or • to https://deepwiki.com/livekit/livekit_composite which is trained on all LiveKit source code If you find the answer, please post it here to help others!