JockRusky
02/18/2026, 2:12 PMAI Evy
02/18/2026, 2:12 PMAI Evy
02/18/2026, 2:12 PMnormalize_for_speech setting, which converts text to its spoken form before audio generation. [(1) ]() This helps the TTS system handle punctuation more naturally. [(2) ]()
You can enable this via the API:
json
"normalize_for_speech": true
[(1) ]()
2. Add Prompt Instructions
Include guidance in your agent's prompt to prevent the LLM from generating problematic punctuation patterns: [(2) ]()
When responding, use natural sentence breaks. Avoid unusual punctuation that might be read aloud.
3. Use the Pronunciation Dictionary
You can add entries to guide how specific characters are handled: [(1) ]()
json
"pronunciation_dictionary": [
{
"word": ".",
"alphabet": "ipa",
"phoneme": ""
}
]
4. Try a Different Voice
Some voice models handle punctuation better than others. Consider testing different voice providers in your agent settings. [(2) ]()
The most effective approach is typically combining speech normalization with prompt engineering to prevent the issue at the source.AI Evy
02/18/2026, 2:12 PMAI Evy
02/18/2026, 2:12 PM