Merge pull request #6786 from PatrickFarley/speech-updates

move and update best practices

Author: prmerger-automator[bot]

articles/ai-services/speech-service/batch-transcription-create.md

@@ -372,14 +372,6 @@ You can store the results of a batch transcription to a writable Azure Blob stor
 
 If you want to store the transcription results in an Azure Blob storage container by using the [Trusted Azure services security mechanism](batch-transcription-audio-data.md#trusted-azure-services-security-mechanism), consider using [Bring-your-own-storage (BYOS)](bring-your-own-storage-speech-resource.md). For more information, see [Use the Bring your own storage (BYOS) Azure AI Foundry resource for speech to text](bring-your-own-storage-speech-resource-speech-to-text.md).
 
-## Bulk submissions and polling
-
-Batch transcription is asynchronous, and requests are processed one at a time in each region. Submitting jobs at a higher rate does not speed up processing. For example, sending 600 or 6,000 requests per minute has no effect on throughput.
-
-[When monitoring job status](./batch-transcription-get.md), polling every few seconds is unnecessary. If you submit multiple jobs, only the first job will be processed initially; subsequent jobs will wait until the first job completes. Polling all jobs frequently increases system load without benefit. Checking status every ten minutes is sufficient, and polling more often than once per minute is not recommended. 
-
-To optimize throughput for large-scale batch transcription, consider distributing your jobs across multiple supported Azure regions. This approach can help balance load and reduce overall processing time, provided your data and compliance requirements allow for multi-region usage. Review [region availability](./regions.md) and ensure your storage and resources are accessible from each region you plan to use.
-
 ## Related content
 
 - [Learn more about batch transcription](batch-transcription.md)

articles/ai-services/speech-service/batch-transcription.md

@@ -34,6 +34,21 @@ To use the batch transcription REST API:
 > [!IMPORTANT]
 > Batch transcription jobs are scheduled on a best-effort basis. At peak hours it might take up to 30 minutes or longer for a transcription job to start processing. See how to check the current status of a batch transcription job in [this section](batch-transcription-get.md#get-transcription-status).
 
+## Best practices for improving performance
+
+
+**Request size**: Batch transcription is asynchronous, and requests are processed one at a time in each region. Submitting jobs at a higher rate does not speed up processing. For example, sending 600 or 6,000 requests per minute has no effect on throughput. We recommend submitting ~1000 files in a single `Transcription_Create` request. This way, you send fewer requests overall.
+
+**Time distribution**: Distribute your requests over time: Submit them across several hours rather than sending them all within a few minutes. Backend processing maintains a stable performance level due to fixed bandwidth, so sending requests too quickly doesn’t improve performance.
+
+**Job monitoring**: [When monitoring job status](./batch-transcription-get.md), polling every few seconds is unnecessary. If you submit multiple jobs, only the first job will be processed initially; subsequent jobs will wait until the first job completes. Polling all jobs frequently increases system load without benefit. Checking the status every 10 minutes is sufficient, and polling more often than once per minute is not recommended. 
+- Because of the sequential processing, you can get job status by checking only a subset of the files: check the first 100 files, and if they're not completed, later batches are likely not competed either. We recommend waiting at least one minute (ideally five minutes) before checking again. 
+
+**Avoid peak traffic for API calls**: The `ListFiles`, `Update`, and `Get` API calls behave similarly to the `Create` call and should be minimized during peak traffic times. 
+
+**Load balancing**: To optimize throughput for large-scale batch transcription, consider distributing your jobs across multiple supported Azure regions. This approach can help balance load and reduce overall processing time, provided your data and compliance requirements allow for multi-region usage. Review [region availability](./regions.md) and ensure your storage and resources are accessible from each region you plan to use.
+
+
 ## Related content
 
 - [Locate audio files for batch transcription](batch-transcription-audio-data.md)

articles/ai-services/speech-service/speech-services-quotas-and-limits.md

@@ -169,7 +169,12 @@ The following quotas are adjustable for Standard (S0) resources. The Free (F0) r
 - Text to speech [maximum number of transactions per time period](#text-to-speech-quotas-and-limits-per-resource) for standard voices and custom voices
 - Speech translation [concurrent request limit](#real-time-speech-to-text-and-speech-translation)
 
-Before requesting a quota increase (where applicable), check your current TPS (transactions per second) and ensure that it's necessary to increase the quota. Speech service uses autoscaling technologies to bring the required computational resources in on-demand mode. At the same time, Speech service tries to keep your costs low by not maintaining an excessive amount of hardware capacity.
+Before requesting a quota increase (where applicable), check your current TPS (transactions per second) and ensure that you need to increase the quota.
+
+> [!NOTE]
+> Batch transcription is an asynchronous process, and jobs are processed one-by-one in a queue. So, increasing the quota won't improve transcription performance. For performance improvements, see [Batch transcription best practices](./batch-transcription.md#best-practices-for-improving-performance).
+
+Speech service uses autoscaling technologies to bring the required computational resources in on-demand mode. At the same time, Speech service tries to keep your costs low by not maintaining an excessive amount of hardware capacity.
 
 Let's look at an example. Suppose that your application receives response code 429, which indicates that there are too many requests. Your application receives this response even though your workload is within the limits defined by the [Quotas and limits reference](#quotas-and-limits-reference). The most likely explanation is that Speech service is scaling up to your demand and didn't reach the required scale yet. Therefore the service doesn't immediately have enough resources to serve the request. In such cases, increasing the quota won’t help. In most cases, the Speech service will scale up soon, and the issue causing response code 429 will be resolved.