The issue you're experiencing with ElevenLabs TTS multilingual v2 mispronouncing numbers in LiveKit agents v1 could be related to how the model processes numerical values. While the ElevenLabs turbo v2 and their website handle this correctly, the multilingual v2 model may have differences in text normalization or preprocessing that affect number pronunciation. To address this, you can try the following steps: 1. Enable SSML Parsing: Ensure that

enable_ssml_parsing

is set to

True

in your configuration. This allows you to use SSML (Speech Synthesis Markup Language) to customize pronunciation. 2. Use SSML Phoneme Tags: You can use SSML

phoneme

tags to specify the correct pronunciation for numbers or other terms. This is particularly useful for ensuring accuracy in financial values. 3. Preprocess Text: Convert numbers into their written-out forms (e.g., "118,152.31" to "one hundred eighteen thousand one hundred fifty-two point thirty-one") before sending the text to the TTS system. This can help improve pronunciation consistency. If these steps do not resolve the issue, it may be necessary to consult ElevenLabs' technical documentation or support team for model-specific details about why the multilingual v2 model behaves differently in this context. Sources: Speech & Audio Processing | livekit/livekit_composite | DeepWiki | How to Get Help from LiveKit | Rime integration guide | LiveKit Docs | ElevenLabs TTS integration guide | LiveKit Docs | tools/tts/tts.cpp

?