Change in TTS voice

Question

As of today, the voice of "pt-BR-ThalitaMultilingualNeural" has completely changed. Is there a reason for that? When I listen to the voice in the Voice Gallery (https://speech.microsoft.com/portal/voicegallery), it doesn’t sound the same.

Will this new voice remain?

Answer 1

Hi Leandro Camparoni

The voice pt-BR‑ThalitaMultilingualNeural sounds different because Microsoft updated the underlying multilingual neural text‑to‑speech model behind the voice. The voice name did not change, but the backend model that generates the audio was replaced or upgraded by Microsoft to improve naturalness, expressiveness, or multilingual consistency. This kind of change has been reported before for other multilingual voices (for example, AvaMultilingualNeural) and can happen without any SDK, API, or configuration change on your side.

The Speech Studio Voice Gallery always plays audio from the currently active production model. When Microsoft updates the backend model, the gallery preview immediately reflects the new sound. As a result, when you listen to Thalita today in the Voice Gallery, you are hearing the new version of the voice, not the previous one. This is expected behavior for managed Azure Speech voices.

If your application still uses pt-BR-ThalitaMultilingualNeural, it is now automatically generating speech with the new model. Azure Speech does not provide version pinning for standard neural or multilingual voices, so applications transparently move to the updated voice model when Microsoft deploys it. There is no configuration option to keep using the old sound

Microsoft does not allow rollback to previous versions of standard multilingual neural voices. Once the backend model is replaced, the older version is retired and becomes inaccessible. This limitation is acknowledged in community discussions, and the Azure Speech release notes do not guarantee advance notice for such voice changes

The new Thalita voice will remain active until Microsoft performs another backend update. However, multilingual voices are more likely to evolve over time compared to single‑language voices, because they are periodically retrained or migrated to newer TTS or HD voice architectures. Future changes can therefore happen again without prior notice. [learn.microsoft.com]

If long‑term voice stability is important for production use, Microsoft guidance suggests alternatives:

• Prefer single‑language neural voices, which historically change less often.
• Use Custom Voice, where you control the voice lifecycle and updates (subject to eligibility and approval).
• Pre‑generate and store audio for fixed content instead of generating it dynamically. These are currently the only supported ways to mitigate unexpected voice changes in Azure Speech. [learn.microsoft.com]

Note:Thalita voice sounds different because Microsoft updated the underlying multilingual neural model. The Voice Gallery correctly reflects this new production voice, there is no rollback option, and future changes are possible. If consistency is required, switching strategy (single‑language voice, custom voice, or pre‑generated audio) is recommended.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thankyou!

Answer 2

The change in the voice of "pt-BR-ThalitaMultilingualNeural" could be due to updates or improvements made to the Azure Speech service, which may include enhancements to voice quality or characteristics. Such updates are often part of ongoing efforts to improve the naturalness and accuracy of the text-to-speech (TTS) voices.

As for whether this new voice will remain, typically, changes to voices in Azure Speech can be permanent, especially if they are part of a broader update or enhancement. However, specific details about the permanence of a voice change would typically be communicated through official release notes or announcements from Microsoft.

For the most accurate and up-to-date information, it would be best to refer to the official Azure Speech release notes or announcements.