Update text-to-speech.md

Author: sally-baolian

articles/ai-services/speech-service/text-to-speech.md

@@ -91,24 +91,24 @@ For detailed information, see [Speech service pricing](https://azure.microsoft.c
 > [!IMPORTANT]
 > Each Chinese character is counted as two characters for billing, including kanji used in Japanese, hanja used in Korean, or hanzi used in other languages.  
 
-### Personal voice
+### Model training and hosting time for custom neural voice
 
-When you use the personal voice feature, you're billed for both profile storage and synthesis.
+Custom neural voice training and hosting are both calculated by hour and billed per second. For the billing unit price, see [Speech service pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services/).
 
-*  **Profile storage**: Billed per voice per day. Storing a voice for less than a day will be rounded up to a full day for billing.
-*  **Synthesis**: Billed per character. For details on billable characters, see the above [billable characters](#billable-characters).
+Custom neural voice (CNV) training time is measured by ‘compute hour’ (a unit to measure machine running time). Typically, when training a voice model, two computing tasks are running in parallel. So, the calculated compute hours are longer than the actual training time. On average, it takes less than one compute hour to train a CNV Lite voice; while for CNV Pro, it usually takes 20 to 40 compute hours to train a single-style voice, and around 90 compute hours to train a multi-style voice. The CNV training time is billed with a cap of 96 compute hours. So in the case that a voice model is trained in 98 compute hours, you'll only be charged with 96 compute hours. 
 
-### Text to speech avatar
+Custom neural voice (CNV) endpoint hosting is measured by the actual time (hour). The hosting time (hours) for each endpoint is calculated at 00:00 UTC every day for the previous 24 hours. For example, if the endpoint has been active for 24 hours on day one, it's billed for 24 hours at 00:00 UTC the second day. If the endpoint is newly created or suspended during the day, it's billed for its accumulated running time until 00:00 UTC the second day. If the endpoint isn't currently hosted, it isn't billed. In addition to the daily calculation at 00:00 UTC each day, the billing is also triggered immediately when an endpoint is deleted or suspended. For example, for an endpoint created at 08:00 UTC on December 1, the hosting hour will be calculated to 16 hours at 00:00 UTC on December 2 and 24 hours at 00:00 UTC on December 3. If the user suspends hosting the endpoint at 16:30 UTC on December 3, the duration (16.5 hours) from 00:00 to 16:30 UTC on December 3 will be calculated for billing.
 
-When using the text to speech avatar feature, charges will be incurred based on the minutes of video output. However, with the real-time avatar, charges are based on the minutes of avatar activation, irrespective of whether the avatar is actively speaking or remaining silent. To optimize costs for real-time avatar usage, refer to the provided tips in the [sample code](https://github.com/Azure-Samples/cognitive-services-speech-sdk/tree/master/samples/js/browser/avatar#chat-sample) (search "Use Local Video for Idle").
+### Personal voice
 
-### Model training and hosting time for custom neural voice
+When you use the personal voice feature, you're billed for both profile storage and synthesis.
 
-Custom neural voice training and hosting are both calculated by hour and billed per second. For the billing unit price, see [Speech service pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services/).
+*  **Profile storage**: After a personal voice profile is created, it will be billed until it is removed from the system. The billing unit is per voice per day. If voice storage lasts for a period of less than 24 hours, it will be billed as one full day.
+*  **Synthesis**: Billed per character. For details on billable characters, see the above [billable characters](#billable-characters).
 
-Custom neural voice (CNV) training time is measured by ‘compute hour’ (a unit to measure machine running time). Typically, when training a voice model, two computing tasks are running in parallel. So, the calculated compute hours are longer than the actual training time. On average, it takes less than one compute hour to train a CNV Lite voice; while for CNV Pro, it usually takes 20 to 40 compute hours to train a single-style voice, and around 90 compute hours to train a multi-style voice. The CNV training time is billed with a cap of 96 compute hours. So in the case that a voice model is trained in 98 compute hours, you'll only be charged with 96 compute hours. 
+### Text to speech avatar
 
-Custom neural voice (CNV) endpoint hosting is measured by the actual time (hour). The hosting time (hours) for each endpoint is calculated at 00:00 UTC every day for the previous 24 hours. For example, if the endpoint has been active for 24 hours on day one, it's billed for 24 hours at 00:00 UTC the second day. If the endpoint is newly created or suspended during the day, it's billed for its accumulated running time until 00:00 UTC the second day. If the endpoint isn't currently hosted, it isn't billed. In addition to the daily calculation at 00:00 UTC each day, the billing is also triggered immediately when an endpoint is deleted or suspended. For example, for an endpoint created at 08:00 UTC on December 1, the hosting hour will be calculated to 16 hours at 00:00 UTC on December 2 and 24 hours at 00:00 UTC on December 3. If the user suspends hosting the endpoint at 16:30 UTC on December 3, the duration (16.5 hours) from 00:00 to 16:30 UTC on December 3 will be calculated for billing.
+When using the text-to-speech avatar feature, charges will be incurred based on the length of video output and will be billed per second. However, for the real-time avatar, charges are based on the time when the avatar is active, regardless of whether it is speaking or remaining silent, and will also be billed per second. To optimize costs for real-time avatar usage, refer to the tips provided in the [sample code](https://github.com/Azure-Samples/cognitive-services-speech-sdk/tree/master/samples/js/browser/avatar#chat-sample) (search "Use Local Video for Idle"). Avatar hosting is billed per second per endpoint. You can suspend your endpoint to save costs. If you want to suspend your endpoint, you can delete it directly. To use it again, simply redeploy the endpoint.
 
 ## Reference docs