TTS voice model "az-AZ-BabekNeural (Male)" sometimes can't correctly pronounce "ş" and "ğ" letters in Azerbaijani

Hsn Vr 0 Reputation points
2026-08-03T17:51:39.6833333+00:00

TTS voice model "az-AZ-BabekNeural (Male)" sometimes can't correctly pronounce "ş" and "ğ" letters in Azerbaijani, he says "s" and "q" instead of it. Please fix it

Azure AI Immersive Reader
Azure AI Immersive Reader

An Azure Applied AI Service that embeds text reading and comprehension capabilities into your applications.

0 comments No comments

1 answer

Sort by: Most helpful
  1. Jerald Felix 18,760 Reputation points Volunteer Moderator
    2026-08-06T02:59:21.19+00:00

    Hello Hsn Vr,

    Greetings! Thanks for raising this question in Q&A forum.

    The behavior you described with az-AZ-BabekNeural mispronouncing Azerbaijani letters such as**"ş"** and**"ğ"** appears to be a voice model pronunciation issue rather than a problem with your code or Speech service configuration.

    1. Verify that your input text is encoded as UTF-8 and that the characters "ş" and "ğ" are being sent correctly to the Speech service.
    2. Test the same text with another Azerbaijani voice (if available) or with the Speech Studio samples to confirm whether the issue is specific to az-AZ-BabekNeural.
    3. If the pronunciation issue occurs consistently with the same words, collect a few sample phrases and the generated audio output.
    4. Open a support request through Azure Support and provide the sample text, audio output, region, and voice name so the Speech engineering team can review the voice model behavior.

    At this point, the next action is with the Microsoft Speech product team, as pronunciation quality issues in a specific neural voice typically require backend model review and improvement rather than customer-side changes.

    If this answer helps you kindly accept the answer which will help others who have similar questions Best

    Regards,

    Jerald Felix.

    Was this answer helpful?

    0 comments No comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.