Merge pull request #7770 from wangkenpu/wake/pronscore

[PronScore] Add FAQ, update doc for EnableMiscue configuration and remove price for content score

Author: ttorble

articles/ai-services/speech-service/faq-stt.yml

@@ -12,7 +12,7 @@ metadata:
 title: Speech to text FAQ
 summary: |
   This article answers commonly asked questions about the speech to text capability. If you can't find answers to your questions here, check out [other support options](../cognitive-services-support-options.md?context=%2fazure%2fcognitive-services%2fspeech-service%2fcontext%2fcontext%253fcontext%253d%2fazure%2fcognitive-services%2fspeech-service%2fcontext%2fcontext).
-  
+
 
 sections:
   - name: General
@@ -26,14 +26,14 @@ sections:
           Where do I start if I want to use a base model?
         answer: |
           First, get an API key and region in the [Azure portal](https://portal.azure.com). If you want to make REST calls to a predeployed base model, see the [REST APIs](rest-speech-to-text.md) documentation. If you want to use WebSockets, [download the Speech SDK](speech-sdk.md).
-          
+
       - question: |
           Do I always need to build a custom speech model?
         answer: |
           No. If your application uses generic, day-to-day language, you don't need to customize a model. If your application is used in an environment where there's little or no background noise, you don't need to customize a model.
-          
+
           You can deploy baseline and customized models in the portal and then run accuracy tests against them. You can use this feature to measure the accuracy of a base model versus a custom model.
-          
+
       - question: |
           How do I know when the processing for my dataset or model is complete?
         answer: |
@@ -53,62 +53,62 @@ sections:
           I get several results for each phrase with the detailed output format. Which one should I use?
         answer: |
           Always take the first result, even if another result ("N-Best") might have a higher confidence value. Speech service considers the first result to be the best. The result can also be an empty string if no speech was recognized.
-          
+
           The other results are likely worse and might not have full capitalization and punctuation applied. These results are most useful in special scenarios, such as giving users the option to pick corrections from a list or handling incorrectly recognized commands.
-          
+
       - question: |
           Why are there multiple base models?
         answer: |
           You can choose from more than one base model in Speech service. Each model name contains the date when it was added. When you start training a custom model, use the most recent model to get the best accuracy. Older base models are still available for some time after a new model is made available. You can continue using the model that you worked with until it's retired (see [Model and endpoint lifecycle](./how-to-custom-speech-model-and-endpoint-lifecycle.md)). We still recommend that you switch to the latest base model for better accuracy.
-          
+
       - question: |
           Can I update my existing model (model stacking)?
         answer: |
           You can't update an existing model. As a solution, combine the old dataset with the new dataset and readapt.
-          
+
           The old dataset and the new dataset must be combined in a single .zip file (for acoustic data) or in a .txt file (for language data). When the adaptation is finished, redeploy the new, updated model to obtain a new endpoint.
-          
+
       - question: |
           When a new version of a base model is available, is my deployment automatically updated?
         answer: |
           Deployments are *not* automatically updated.
-          
+
           If you adapted and deployed a model, the existing deployment remains as is. You can decommission the deployed model, readapt it by using the newer version of the base model, and redeploy it for better accuracy.
-          
+
           Both base models and custom models are retired after some time (see [Model and endpoint lifecycle](./how-to-custom-speech-model-and-endpoint-lifecycle.md)).
-          
+
       - question: |
           Can I download my model and run it locally?
         answer: |
           You can run a custom model locally in a [Docker container](speech-container-howto.md?tabs=cstt).
-          
+
       - question: |
           Can I copy or move my datasets, models, and deployments to another region or subscription?
         answer: |
           You can use the [Models_Copy REST API](/rest/api/speechtotext/models/copy) to copy a custom model to another region or subscription. Datasets and deployments can't be copied. You can import a dataset again in another subscription and create endpoints there by using the model copies.
-          
+
       - question: |
           Are my requests logged?
         answer: |
           By default, requests aren't logged (neither audio nor transcription). If necessary, you can select the **Log content from this endpoint** option when you [create a custom endpoint](how-to-custom-speech-deploy-model.md#add-a-deployment-endpoint). You can also enable audio logging in the [Speech SDK](how-to-use-logging.md) on a per-request basis, without having to create a custom endpoint. In both cases, audio and recognition results of requests will be stored in secure storage. Subscriptions that use Microsoft-owned storage are available for 30 days.
-          
+
           You can export the logged files on the deployment page in Speech Studio if you use a custom endpoint with **Log content from this endpoint** enabled. If audio logging is enabled via the SDK, call the API to access the files. You can also use API to [delete the logs](/rest/api/speechtotext/endpoints/delete-base-model-log) anytime.
-          
+
       - question: |
           Are my requests throttled?
         answer: |
           For information, see [Speech service quotas and limits](speech-services-quotas-and-limits.md).
-          
+
       - question: |
           How am I charged for dual channel audio?
         answer: |
           If you submit each channel separately in their own file, you're charged for the audio duration of each file. If you submit a single file with the channels multiplexed together, you're charged for the duration of the single file. For more information about pricing, see the [Azure AI services pricing page](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services/).
-          
+
           > [!IMPORTANT]
           > If you have further privacy concerns that prevent you from using the custom speech service, contact one of the support channels.
 
           ## Increasing concurrency
-          
+
           For information, see [Speech service quotas and limits](speech-services-quotas-and-limits.md).
 
   - name: Importing data
@@ -117,7 +117,7 @@ sections:
           What is the limit to the size of a dataset, and why is it the limit?
         answer: |
           The limit is because of the restriction on the size of files for HTTP upload. For the actual limit, see [Speech service quotas and limits](speech-services-quotas-and-limits.md). You can split your data into multiple datasets and select all of them to train the model.
-          
+
       - question: |
           Can I zip (compress) my text files so that I can upload a larger text file?
         answer: |
@@ -154,18 +154,18 @@ sections:
           How long does it take to train a custom model with audio data?
         answer: |
           Training a model with audio data can be a lengthy process. Depending on the amount of data, it can take several days to create a custom model. If it can't be finished within one week, the service might abort the training operation and report the model as failed.
-          
+
           In general, Speech service processes approximately 10 hours of audio data per day in regions that have dedicated hardware. Training with text only is faster and ordinarily finishes within minutes.
-          
-          Use one of the regions where dedicated hardware is available for training. The Speech service uses up to 100 hours of audio for training in these regions. 
-          
+
+          Use one of the regions where dedicated hardware is available for training. The Speech service uses up to 100 hours of audio for training in these regions.
+
   - name: Accuracy testing
     questions:
       - question: |
           What is word error rate (WER), and how is it computed?
         answer: |
           WER is the evaluation metric for speech recognition. WER is calculated as the total number of errors (insertions, deletions, and substitutions), divided by the total number of words in the reference transcription. For more information, see [Test model quantitatively](how-to-custom-speech-evaluate-data.md#evaluate-word-error-rate-wer).
-          
+
       - question: |
           How do I determine whether the results of an accuracy test are good?
         answer: |
@@ -188,9 +188,20 @@ sections:
         answer: |
           Uploading a list of words adds them to the vocabulary, but it doesn't teach the system how the words are ordinarily used. By providing full or partial utterances (sentences or phrases of things that users are likely to say), the language model can learn the new words and how they're used. The custom language model is good not only for adding new words to the system, but also for adjusting the likelihood of known words for your application. Providing full utterances helps the system learn better.
 
+  - name: Pronunciation assessment
+    questions:
+      - question: |
+          Why does the recognized text differ from the reference text?
+        answer: |
+          The recognized text is generated based on the audio input, the reference text, the `EnableMiscue` configuration and the assessment mode.
+
+          In **scripted assessment**, there are two modes, single-shot and continuous, and the behavior differs slightly. In single-shot mode, if `EnableMiscue` is set to `false`, the system forces the recognized text to match the reference text. When `EnableMiscue` is `true`, only the words present in the reference text are considered as recognized results from the audio input. Continuous mode does not support the `EnableMiscue` option and behaves similarly to single-shot mode with `EnableMiscue` set to `true`. Differences between recognized and reference text might occur due to factors such as pronunciation variations, background noise, or limitations in the speech recognition model.
+
+          In **unscripted assessment**, the recognized text is generated solely from the audio input without any reference text, which can lead to discrepancies between the recognized text and the intended content. In these cases, the recognized text reflects what the system interprets from the audio and may not always align with the expected message. If you notice significant differences, review the audio quality and speaker clarity, or consider using Azure Speech-to-Text to transcribe the audio first. You can then use that transcription as the reference text for a more accurate assessment.
+
 additionalContent: |
 
   ## Next steps
-  
+
   - [Speech to text quickstart](get-started-speech-to-text.md)
   - [What's new](releasenotes.md)

articles/ai-services/speech-service/how-to-pronunciation-assessment.md

@@ -84,7 +84,7 @@ For how to use Pronunciation Assessment in streaming mode in your own applicatio
 
 ::: zone pivot="programming-language-csharp"
 
-If your audio file exceeds 30 seconds, use continuous mode for processing. The sample code for continuous mode can be found on [GitHub](https://github.com/Azure-Samples/cognitive-services-speech-sdk/blob/master/scenarios/csharp/sharedcontent/console/speech_recognition_samples.cs) under the function `PronunciationAssessmentContinuousWithFile`.
+If your audio file exceeds 30 seconds, use continuous mode for processing. In continuous mode, the `EnableMiscue` option is not supported. To obtain `Omission` and `Insertion` tags, you need to compare the recognized results with the reference text. You can find a sample implementation for continuous mode on [GitHub](https://github.com/Azure-Samples/cognitive-services-speech-sdk/blob/master/scenarios/csharp/sharedcontent/console/speech_recognition_samples.cs) under the function `PronunciationAssessmentContinuousWithFile`.
 
 ::: zone-end
 
@@ -519,7 +519,7 @@ This table lists some of the key pronunciation assessment results for the script
 This table lists some of the key pronunciation assessment results for the unscripted assessment, or speaking scenario.
 
 > [!NOTE]
-> Prosody assessment is only available in the [en-US](./language-support.md?tabs=pronunciation-assessment) locale.
+> Prosody assessment is only available in the [en-US](./language-support.md?tabs=pronunciation-assessment) locale. For unscripted assessment, the speech-to-text (STT) model used is different from Azure STT. If you need assessment based on highly accurate recognized text, we recommend first calling Azure STT to obtain the reference text, and then performing scripted assessment.
 
 | Response parameter | Description | Granularity |
 |:-------------------|:------------|:------------|

articles/ai-services/speech-service/pronunciation-assessment-tool.md

@@ -311,9 +311,9 @@ After you press the stop button, you can see **Pronunciation score**, **Accuracy
 
 ## Pricing
 
-As a baseline, usage of pronunciation assessment costs the same as speech to text for Standard or commitment tier [pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services). If you [purchase a commitment tier](../commitment-tier.md) for speech to text, the spend for pronunciation assessment goes towards meeting the commitment. 
+As a baseline, usage of pronunciation assessment costs the same as speech to text for Standard or commitment tier [pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services). If you [purchase a commitment tier](../commitment-tier.md) for speech to text, the spend for pronunciation assessment goes towards meeting the commitment.
 
-The pronunciation assessment feature also offers other scores that aren't included in the baseline speech to text price: prosody, grammar, topic, and vocabulary. These scores are available as an add-on charge above the baseline speech to text price. For information about pricing, see [speech to text pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services).
+The pronunciation assessment feature also offers other scores that aren't included in the baseline speech to text price, such as the prosody score. These scores are available as an add-on charge above the baseline speech to text price. For information about pricing, see [speech to text pricing](https://azure.microsoft.com/pricing/details/cognitive-services/speech-services).
 
 Here's a table of available pronunciation assessment scores, whether it's available in the [scripted](#conduct-a-reading-assessment) or [unscripted](#conduct-a-speaking-assessment) assessments, and whether it's included in the baseline speech to text price or the add-on price.
 
@@ -324,9 +324,6 @@ Here's a table of available pronunciation assessment scores, whether it's availa
 | Completeness | Scripted | Yes |
 | Miscue | Scripted and unscripted | Yes |
 | Prosody | Scripted and unscripted | No |
-| Grammar | Unscripted only | No |
-| Topic | Unscripted only | No |
-| Vocabulary | Unscripted only | No |
 
 ## Responsible AI