Share via

Why have they changed Vivienne Multilingual's voice?

Albert MG 0 Reputation points
2026-03-18T11:26:22.8266667+00:00

Since March 17th, Vivienne's multilingual voice has changed; it's no longer the same. It sounds deeper, and it's not the same in Spanish (Spain) or English (US). The voice the model has isn't the multilingual one; it's the default one, and it sounds terrible. It's as if something has been misconfigured. This isn't just a personal issue; I've tried it in Azure, and even the same voice that appears in Clipchamp, and it's definitely not the same.

Azure Language in Foundry Tools
Azure Language in Foundry Tools

An Azure service that provides natural language capabilities including sentiment analysis, entity extraction, and automated question answering.

0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Q&A Assist
    2026-03-18T11:26:43.6266667+00:00
    AI generated content. This question contains an answer created with AI from Q&A Assist. Learn more

    VivienneMultilingualNeural is part of the Azure Speech multilingual and HD voice family, which is actively being updated and improved. These voices are based on large language models and are explicitly documented as evolving over time, including changes to quality, prosody, and sometimes timbre.

    Key points relevant to the change being observed:

    1. Multilingual and HD voices are versioned and updated
      • Azure Speech release notes show ongoing updates to multilingual and Dragon HD voices (for example, multiple rounds of “quality improved and bugs fixed,” new HD versions, and Omni variants).
      • HD and “Latest” model suffixes (such as DragonHDLatestNeural, DragonHDOmniLatestNeural, and other Latest voices) are specifically called out as models that can be updated over time for better quality and features. This can result in audible changes in pitch, tone, and expressiveness while keeping the same voice name.
    2. Multilingual voices share a unified model across many locales
      • Multilingual voices (including VivienneMultilingualNeural) are backed by a single multilingual model that supports many locales from the same base voice.
      • When that underlying multilingual model is updated, the change affects all supported languages at once (for example, Spanish and English), which matches the behavior being described.
    3. Voice behavior can differ between “standard” and HD/Omni variants
      • Release notes show that some voices now have Dragon HD or Dragon HD Omni variants (for example, …:DragonHDLatestNeural or …:DragonHDOmniLatestNeural) that maintain the same persona but with different acoustic behavior.
      • If an integration (such as Clipchamp or an Azure region/endpoint) has been switched to a newer “Latest” or HD/Omni-backed implementation behind the same display name, the audible result can sound deeper or more expressive even though the name has not changed.
    4. “Latest” models are explicitly not fixed
      • Documentation for personal/custom voices notes that …LatestNeural models are updated from time to time and that behavior can change with those updates; a fixed version requires using a versioned model name instead of Latest.
      • The same principle applies to base multilingual/HD voices: when using a Latest-style or non-versioned multilingual voice, changes like the one being observed are expected over time as the service is improved.

    Given the above, the change in VivienneMultilingualNeural’s sound since mid‑March is consistent with a backend model update rather than a misconfiguration on a single tenant. The same updated behavior would appear across Azure Speech and any integrated products (such as Clipchamp) that call the same voice name.

    To restore a previous sound, the only supported path would be to:

    • Use a versioned voice model if/when Microsoft exposes one for that persona, or
    • Create a custom/personal voice based on a base model whose version is controlled, then keep that version fixed instead of relying on a Latest/evolving model.

    For an issue that feels like a regression (for example, “sounds terrible” or clearly worse quality), the recommended next step is to open a support ticket with Azure Support, referencing the specific voice name, region, and approximate date when the change was noticed, so the product group can review the recent deployment for that voice.

    References:

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.