Issue with Ryan Multilingual Voice – Accent Changing from American to Australian

Question

Experiencing an issue where the Ryan Multilingual Voice unexpectedly changes its accent from American to Australian during audio generation. Our bot language is set to English - US, and am are consistently using Ryan Multilingual Voice.

We have tested audio generation using the Microsoft Speech Studio, and the output sounds correct. However, during live calls on our Yellow server, we intermittently observe a shift in accent.

Issue Details:

• Voice Used: Ryan (Multilingual)
• Expected Accent: American (English - US)
• Observed Issue: Intermittent shift to an Australian accent during live call audio generation
• Speech Studio Output: Correct American accent
• Live Call Output: Accent changes to Australian at times

Answer 1

Hello vijo.jose,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having issues with Ryan Multilingual Voice – Accent Changing from American to Australian.

Most of all, you will need to understand that multilingual voices dynamically adjust accents based on input/text unless explicitly locked to a locale. The issue likely stems from missing SSML restrictions or regional misconfigurations, not updates/network issues. Follow the below steps to address this and ensure consistent accent output.

1. Ensure live calls use SSML with strict locale enforcement:

      <voice name='en-US-RyanMultilingualNeural'>
        <lang xml:lang='en-US'>[Your text here]</lang>
      </voice>

This prevents automatic accent detection based on input text.

1. Confirm the exact voice name in live calls matches the Speech Studio version. Multilingual voices have region-specific identifiers (e.g., en-US-RyanMultilingualNeural vs. a generic "Ryan" alias).
2. Check if the Yellow server uses the same Azure region as Speech Studio (e.g., eastus vs. australiaeast). Accent shifts may occur if the server defaults to an Australian endpoint.
3. Audit live call inputs for phrases/spellings that might trigger Australian English (e.g., "colour" vs. "color"). Multilingual models auto-adapt to dialects unless explicitly restricted.
4. Capture raw SSML/logs from live calls where the accent shifted. Compare against Speech Studio requests to identify discrepancies in voice parameters or locales.
5. If the above doesn't work as expected contact Microsoft Support via your Azure Portal to escalate with the following details:
   1. SSML snippets from failed/successful calls.
   2. Azure region and voice model name used.
   3. Timestamps/logs of accent shifts.

I hope this is helpful! Do not hesitate to let me know if you have any other questions.

---

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.