Share via

Azure Fast Transcription – Inability to enforce British spelling (en‑GB) without post‑processing

Julien S 106 Reputation points
2026-03-19T16:56:55.44+00:00

Hello,

I am currently using Azure Speech – Fast Transcription with the locale explicitly set to en-GB, on a Speech resource deployed in UK South.

In most cases, the recognition quality is good. However, I consistently observe that some words are transcribed using American spelling (e.g. “analyzed” instead of “analysed”), even though:

  • the spoken audio is clearly British English,
  • the locale is forced to en-GB,
  • and the service region is UK South.

After reviewing the Azure documentation, I understand that:

  • Fast Transcription outputs display text only (no lexical form),
  • it does not support Custom Speech, phrase lists, or output-style control,
  • and it relies on the default Speech‑to‑Text base model.

From the Custom Speech overview documentation, Microsoft explains that speech recognition uses a Universal Language Model trained across dialects and domains, and that all standard Speech‑to‑Text scenarios use this model by default.

This seems to indicate that, even when en-GB is specified, orthographic variants (en‑GB vs en‑US spelling) are not strictly enforced, and American spellings may still be selected during output normalization.

My question

Is there any supported way, within Azure Speech itself, to strongly enforce British spelling conventions when using Fast Transcription, without relying on external post‑processing?

More specifically:

  • Is there any roadmap, hidden configuration, or upcoming feature that would allow controlling orthographic preferences for English variants?
  • Or is the current behavior expected by design due to the shared / universal language model and the use of display-text normalization?

I am not looking for a workaround via regex or external spell-checkers (which we already considered), but rather to understand whether Azure Speech is expected to support locale‑specific orthography guarantees in the future, especially for Fast Transcription.

Thank you in advance for any clarification.

Kind regards,Hello,

I am currently using Azure Speech – Fast Transcription with the locale explicitly set to en-GB, on a Speech resource deployed in UK South.

In most cases, the recognition quality is good. However, I consistently observe that some words are transcribed using American spelling (e.g. “analyzed” instead of “analysed”), even though:

  • the spoken audio is clearly British English,
  • the locale is forced to en-GB,
  • and the service region is UK South.

After reviewing the Azure documentation, I understand that:

  • Fast Transcription outputs display text only (no lexical form),
  • it does not support Custom Speech, or output-style control,
  • and it relies on the default Speech‑to‑Text base model.

From the Custom Speech overview documentation, Microsoft explains that speech recognition uses a Universal Language Model trained across dialects and domains, and that all standard Speech‑to‑Text scenarios use this model by default.

This seems to indicate that, even when en-GB is specified, orthographic variants (en‑GB vs en‑US spelling) are not strictly enforced, and American spellings may still be selected during output normalization.

My question

Is there any supported way, within Azure Speech itself, to strongly enforce British spelling conventions when using Fast Transcription, without relying on external post‑processing?

More specifically:

  • Is there any roadmap, hidden configuration, or upcoming feature that would allow controlling orthographic preferences for English variants?
  • Or is the current behavior expected by design due to the shared / universal language model and the use of display-text normalization?

I am not looking for a workaround via regex or external spell-checkers (which we already considered), but rather to understand whether Azure Speech is expected to support locale‑specific orthography guarantees in the future, especially for Fast Transcription.

Thank you in advance for any clarification.

Kind regards,

Julien

Azure AI Speech
Azure AI Speech

An Azure service that integrates speech processing into apps and services.

0 comments
Answer accepted by question author
  1. Anshika Varshney 9,335 Reputation points Microsoft External Staff Moderator
    2026-04-01T16:24:20.1133333+00:00

    Hi Julien S,

    This behavior is expected when using Azure Speech Fast Transcription and is not caused by a misconfiguration or service issue.

    Here is how Fast Transcription works and why enforcing strict British spelling is limited.

    First, understand how Fast Transcription handles language and spelling. Fast Transcription always returns display text only. Display text is optimized to be readable for humans and goes through a normalization step. During this step, the speech service uses a shared English language model. Even if you set the locale to en‑GB and use a UK region, the model may still output American spellings for some words. This is by design and not considered an error. https://learn.microsoft.com/en-us/azure/ai-services/speech-service/fast-transcription-create

    Second, know the current limitations of Fast Transcription. Fast Transcription does not support Custom Speech, phrase lists, spelling style controls, or orthographic enforcement. Because of this, you cannot force British spelling rules within the Fast Transcription API itself. The service relies on a universal English language model that blends dialects, and spelling variations can appear even when the locale is explicitly set.

    Third, confirm that locale settings are being applied correctly. Setting the locale to en‑GB improves recognition accuracy for British English audio, but it does not guarantee British spelling for every word. Locale support indicates what language or accent is recognized, not how spelling variants are enforced in the final text output. https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support

    Fourth, understand why this is different from other Speech features. Features like Custom Speech or batch transcription allow more control over recognition behavior, but Fast Transcription is designed for speed and low latency. To achieve that performance, it intentionally limits customization, including spelling control and output formatting.

    Finally, what this means in practice. When using Fast Transcription, mixing of en‑GB and en‑US spellings is expected and currently unavoidable inside the service. If exact British spelling is a hard requirement, spelling normalization must be handled after transcription. There is no supported configuration today that enforces British orthography directly within Fast Transcription.

    In short, Fast Transcription correctly follows the specified locale for recognition, but spelling differences are a known product behavior due to the underlying universal language model.

    I Hope this helps. Do let me know if you have any further queries.

    Thankyou!

    1 person found this answer helpful.
    0 comments
Answer accepted by question author
  1. Q&A Assist
    2026-03-19T16:57:38.69+00:00
    AI generated content. This question contains an answer created with AI from Q&A Assist. Learn more

    The current behavior is expected by design, and the provided documentation does not describe any supported way in Azure Speech – Fast Transcription to strictly enforce British spelling (en‑GB) or to control orthographic variants between English dialects without external post‑processing.

    From the available documentation:

    • Fast transcription uses the Speech service’s unified speech recognition model and requires specifying a locale for each audio input to improve accuracy, but it does not describe any orthography‑control or dialect‑preference settings beyond the locale itself.
    • The locales parameter is documented as affecting language/locale identification and accuracy/latency, not spelling conventions. When a single locale is specified, that locale is used for transcription, but no guarantees are stated about variant spellings within that locale.
    • Fast transcription exposes only a limited set of configuration options (locales, phraseList, profanityFilterMode) and does not list any option for display‑text style, spelling variant, or orthography control.
    • Limitations sections emphasize that accuracy depends on correct locale selection, audio quality, overlapping speech, and vocabulary coverage, but do not mention any mechanism to tune orthography within a locale.

    The context also does not mention any roadmap item, hidden configuration, or upcoming feature related to enforcing locale‑specific spelling variants (such as en‑GB vs en‑US) for Fast Transcription or for Azure Speech more broadly.

    Based on the documentation provided, Azure Speech Fast Transcription currently:

    • Supports choosing the locale (en-GB vs other English locales) via locales.
    • Does not expose controls for orthographic preferences within a locale.
    • Does not document any planned feature or configuration to guarantee locale‑specific spelling beyond the general locale setting.

    Therefore, within the scope of the documented Fast Transcription capabilities, there is no supported way to enforce British spelling conventions without external post‑processing.

    References:

    0 comments

0 additional answers

Sort by: Most helpful
Most helpful Newest Oldest

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.