Hi,
The Speech SDK exposes semantic segmentation via:
speechConfig.setProperty(sdk.PropertyId.Speech_SegmentationStrategy, "Semantic");
The documentation says:
Semantic segmentation isn't available for all languages and locales.
Source: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-recognize-speech
However, I could not find either:
A list of locales that support semantic segmentation.
A documented fallback behavior when semantic segmentation is requested for an unsupported locale.
A documented error/cancellation behavior that clients can rely on when semantic segmentation is unsupported.
A result/property indicating which segmentation strategy was actually applied.
This makes semantic segmentation difficult to enable safely in production. If a client cannot know whether a locale supports it, and cannot rely on a documented fallback or explicit failure, then the feature has to remain disabled even when it could improve continuous recognition UX.
Could the SDK/service documentation provide one of the following?
A semantic-segmentation support column/list by locale.
A clear guarantee that unsupported locales fall back to default segmentation.
A clear guarantee that unsupported locales produce a specific cancellation/error signal.
A way to inspect the effective segmentation strategy for a recognition session.
Thanks.
Version
1.48.0 (Latest, Default)
Hi,
The Speech SDK exposes semantic segmentation via:
speechConfig.setProperty(sdk.PropertyId.Speech_SegmentationStrategy, "Semantic");
The documentation says:
Semantic segmentation isn't available for all languages and locales.
Source: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-recognize-speech
However, I could not find either:
A list of locales that support semantic segmentation.
A documented fallback behavior when semantic segmentation is requested for an unsupported locale.
A documented error/cancellation behavior that clients can rely on when semantic segmentation is unsupported.
A result/property indicating which segmentation strategy was actually applied.
This makes semantic segmentation difficult to enable safely in production. If a client cannot know whether a locale supports it, and cannot rely on a documented fallback or explicit failure, then the feature has to remain disabled even when it could improve continuous recognition UX.
Could the SDK/service documentation provide one of the following?
A semantic-segmentation support column/list by locale.
A clear guarantee that unsupported locales fall back to default segmentation.
A clear guarantee that unsupported locales produce a specific cancellation/error signal.
A way to inspect the effective segmentation strategy for a recognition session.
Thanks.
Version
1.48.0 (Latest, Default)