Hi ElevenLabs Support Team,I’m currently working on automating voice generation using your API in

[stray forum]

AI-generated Answer for rasul_13412_31838

It seems like you're encountering a few challenges with using the pronunciation dictionary and the ElevenLabs API. Here's some guidance based on your questions:

1. Pronunciation Dictionary Support in the /text-to-speech/{voice_id}/stream Endpoint:
   While the context does not explicitly confirm whether the pronunciation dictionary is supported in this specific endpoint, issues with pronunciation dictionaries have been noted in other scenarios. You may want to verify the dictionary's integration by using the GET Agent API endpoint to confirm if the dictionary is correctly added. If you're not using the Agent API, you might need to test alternative approaches, such as using the System Prompt as a makeshift pronunciation dictionary.

2. Limitations for Russian Language Pronunciation with eleven_turbo_v2:
   The eleven_turbo_v2 model is primarily optimized for English, and there may be limitations when working with non-English languages like Russian. For better results with Russian, you might consider switching to the eleven_multilingual_v2 model, which is designed to support multiple languages, including Russian.

3. Correct Way to Include a Pronunciation Dictionary ID:
   Ensure that the pronunciation_dictionary_locators parameter is correctly formatted in your request body. The dictionary ID should be included under this parameter, and you can confirm its presence by checking the API response when retrieving agent details. If you're still encountering issues, double-check the JSON structure for any syntax errors or misplaced data.

4. Alias-Based Pronunciation for Non-English Text:
   Alias-based pronunciation may not work as expected for non-English text, especially with models like eleven_turbo_v2. If you're facing issues, you could try rewriting the pronunciation using standard IPA characters or testing the System Prompt to guide pronunciation.

If the error persists, it might be related to the JSON structure in your request. The error message indicates a "JSON decode error" due to "Extra data," which suggests there might be an issue with how the JSON is formatted. Double-check for any trailing commas, misplaced brackets, or other structural issues in your request body.

If you'd like, you can share the specific JSON structure you're using (excluding sensitive information like API keys), and I can help review it further.

[foggy gazelle]

Hi ElevenLabs Support Team,

I’ve been actively using your platform and experimenting with pronunciation dictionaries, agents, and the TTS API for generating audio. My goal is to automate realistic voiceovers for Russian-language content, with support for custom pronunciation using agents and dictionaries.

I’ve already created an AI Agent in the ElevenLabs interface and uploaded a pronunciation dictionary linked to that agent. However, I discovered that the /v1/voice-agents/{agent_id}/generate-audio endpoint seems to be unavailable — I consistently receive a 404 Not Found error when attempting to use it.

I’ve also seen that the Agent API endpoints such as GET /v1/convai/agents/{agent_id} work correctly — but they don’t allow me to trigger audio generation.

Could you please clarify:

Is there currently any way to generate speech directly from an Agent (via API)?

Is there a way to use a pronunciation dictionary (attached to an agent) when calling /text-to-speech/{voice_id}?

Can I request early access to the full Agent API for voice generation?

What is the recommended way to integrate agents into automated voice workflows using Make or external platforms?

I’d love to fully integrate ElevenLabs into my workflow and make use of the Agent system via API.

Thanks a lot in advance!