Customize a language model with Azure AI Video Indexer

Applies to: Cloud-based Azure AI Video Indexer

Azure AI Video Indexer supports automatic speech recognition through integration with Microsoft Custom Speech Service. You can customize a language model by uploading adaptation text from your domain. After you train your model, the engine recognizes new words that appear in the adaptation text by assuming default pronunciation, and the language model learns more probable word sequences. For the list of supported languages, see Supported languages.

For example, Kubernetes (in the context of Azure Kubernetes Service) is a domain-specific word. If the word is new to Azure AI Video Indexer, it might be recognized as communities. Train the model to recognize it as Kubernetes. In other cases, the words exist, but the language model doesn't expect them to appear in a specific context. For example, container service isn't a two-word sequence that a non-specialized language model would recognize as a specific phrase.

There are two ways to customize a language model:

Option 1: Edit the transcript that Azure AI Video Indexer generated. By editing and correcting the transcript, you train a language model to provide improved results in the future.
Option 2: Upload text files to train the language model. The file can contain a list of words as you want them to appear in the Azure AI Video Indexer transcript. Or, it can include the relevant words naturally in sentences and paragraphs. Because this approach usually gives better results, include full sentences or paragraphs related to your content in the upload file.

Important

Don't include words or sentences as they're currently transcribed incorrectly (for example, communities) in the upload file. Doing so negates the intended effect. Only include the words as you want them to appear (for example, Kubernetes).

Prerequisites

An Azure account
An Azure AI Video Indexer account

Optimize your custom language model

Azure AI Video Indexer learns based on probabilities of word combinations, so to learn best:

Provide enough real examples of sentences as they would be spoken.
Put only one sentence per line. Otherwise, the system learns probabilities across sentence boundaries.
Use a single-word sentence to boost that word against others, but the system learns best from full sentences.
When introducing new words or acronyms, provide as many examples as possible in full sentences to give the system more context.
Try several adaptation options and evaluate which works best.
Avoid repetition of the exact same sentence multiple times. It might create bias against the rest of the input.
Avoid including uncommon symbols (~, # @ % &) because the system discards them. The sentences in which they appear also get discarded.
Avoid very large inputs, such as hundreds of thousands of sentences, because doing so dilutes the boosting effect.

Web portal
API

Create a language model

Go to the Azure AI Video Indexer website and sign in.
To customize a model in your account, select the Content model customization button on the left of the page.
Select the Language tab. You see a list of supported languages.
Under the language that you want, select Add model.
Enter a name for the language model, and then press Enter. This step creates the model and gives you the option to upload text files.
To add a text file, select Add file. Your File Explorer opens.
Navigate to and select the text file. You can add multiple text files to a language model. You can also add a text file by selecting the ... button on the right side of the language model and selecting Add file.
When you're done uploading the text files, select Train.

The training process can take a few minutes. When the training is done, Trained appears next to the model. You can preview, download, and delete the file from the model.

Use a language model on a new video

To use your language model on a new video, complete one of the following actions:

Select Upload at the top of the page.
Drop your audio or video file or browse for your file.
Select a language model you created from the Video source language list.
Select the Upload option at the bottom of the page. Your new video gets indexed by using your language model.

Use a language model to reindex

Sign in to the Azure AI Video Indexer home page.
Select the ellipsis (...) on the video and then select Re-index.
From the Video source language list, select a language model that you created.
Select Re-index. Your video is reindexed by using your language model.

Edit a language model

You can edit a language model by changing its name, adding files to it, and deleting files from it. If you add or delete files from the language model, you need to train the model again by selecting Train.

Rename the language model

You can change the name of the language model by selecting the ellipsis (...) on the right side of the language model and selecting Rename. Enter the new name.

Add files

Select Add file. Your File Explorer opens.
Navigate to and select the text file. You can add multiple text files to a language model.

You can also add a text file by selecting the ellipsis (...) button on the right side of the language model and selecting Add file.

Delete files

This action removes the file completely from the language model.

Select the ellipsis (...) button on the right side of the text file.
Select Delete. A new window appears, telling you that the deletion can't be undone.
Select the Delete option in the new window.

Delete a language model

This action removes the language model from your account. Any video that uses the deleted language model keeps the same index until you reindex the video. If you reindex the video, you can assign a new language model to the video. Otherwise, Azure AI Video Indexer uses its default model to reindex the video.

Select the ellipsis (...) button on the right side of the language model.
Select Delete. A new window appears, telling you that the deletion can't be undone.
Select the Delete option in the new window.

Customize language models by correcting transcripts

Azure AI Video Indexer customizes language models based on the actual corrections users make to the transcriptions of their videos. It captures all lines that you corrected in the transcription of your video and adds them to a text file called From transcript edits. These edits are used to retrain the language model that was used to index the video.

Edits that you make in the widget's timeline are also included.

If you don't specify a language model when indexing this video, Azure AI Video Indexer stores all edits for this video in a default language model called Account adaptations within the detected language of the video.

If you make multiple edits to the same line, Azure AI Video Indexer uses only the last version of the corrected line for updating the language model.

Note

Only textual corrections are used for the customization. Corrections that don't involve actual words (for example, punctuation marks or spaces) aren't included.

Select the video that you want to edit from your library.
Select the Timeline tab.
Select the pencil icon to edit the transcript.
Transcript corrections appear in the Language tab of the Content model customization page. To view the "From transcript edits" file for each of your language models, select the model to open it.

Create a language model

Make a Create Language Model API request to create a new custom language model for the specified account. You can upload files for the language model with this request. Alternatively, you can create the language model here and upload files for the model later by updating the language model.

You must upload files in the request body by using FormData and provide values for the required parameters. You can define the key-value pair for this task in two ways:

The key is the file name, and the value is the .txt file.
The key is the file name, and the value is a URL to a .txt file.

Note

You must still train the model with its enabled files so that the model learns each file's contents.

Example response

{
    "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
    "name": "TestModel",
    "language": "En-US",
    "state": "None",
    "languageModelId": "00000000-0000-0000-0000-000000000000",
    "files": [
    {
        "id": "cccccccc-3333-4444-5555-dddddddddddd",
        "name": "hellofile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-28T11:55:34.6733333"
    },
    {
        "id": "eeeeeeee-6666-7777-8888-ffffffffffff",
        "name": "worldfile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-28T11:55:34.86"
    }
    ]
}

Train a language model

The Train Language Model request trains a custom language model for the specified account with the contents of the uploaded and enabled files in the language model.

Note

You must first create the language model and upload its files. You can upload files when creating the language model or by updating the language model.

Example response

{
    "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
    "name": "TestModel",
    "language": "En-US",
    "state": "Waiting",
    "languageModelId": "a0a0a0a0-bbbb-cccc-dddd-e1e1e1e1e1e1",
    "files": [
    {
        "id": "cccccccc-3333-4444-5555-dddddddddddd",
        "name": "RenamedFile",
        "enable": false,
        "creator": "John Doe",
        "creationTime": "2018-04-27T20:10:10.5233333"
    },
    {
        "id": "eeeeeeee-6666-7777-8888-ffffffffffff",
        "name": "hellofile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-27T20:10:10.68"
    }
    ]
}

The id is a unique identifier used to distinguish language models. However, languageModelId is used for uploading a video to index and reindexing a video requests. Azure AI Video Indexer upload and reindex requests also refer to it as linguisticModelId.

Delete a language model

Use the Delete Language Model API request to delete a custom language model from the specified account. Any video that uses the deleted language model keeps the same index until you reindex the video. If you reindex the video, you can assign a new language model to the video. Otherwise, Azure AI Video Indexer uses its default model to reindex the video.

Example response

There's no returned content when the language model is deleted successfully.

Update a language model

The Update Language Model request updates a custom language model in the specified account.

Note

Make sure you created the language model. Use this call to: enable or disable all files under the model, update the name of the language model, and upload files to add to the language model.

To upload files to add to the language model, upload files in the request body by using FormData. You must also provide values for the required parameters mentioned earlier. There are two ways to do this:

The key is the file name, and the value is the .txt file.
The key is the file name, and the value is a URL to a .txt file.

Example response

{
    "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
    "name": "TestModel",
    "language": "En-US",
    "state": "Waiting",
    "languageModelId": "a0a0a0a0-bbbb-cccc-dddd-e1e1e1e1e1e1",
    "files": [
    {
        "id": "cccccccc-3333-4444-5555-dddddddddddd",
        "name": "RenamedFile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-27T20:10:10.5233333"
    },
    {
        "id": "eeeeeeee-6666-7777-8888-ffffffffffff",
        "name": "hellofile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-27T20:10:10.68"
    }
    ]
}

To download the contents of the file, use the id of the files returned in the response.

Update a file from a language model

The Update Language Model File request updates the name and enable state of a file in a custom language model in the specified account.

Example response

{
    "id": "cccccccc-3333-4444-5555-dddddddddddd",
  "name": "RenamedFile",
  "enable": false,
  "creator": "John Doe",
  "creationTime": "2018-04-27T20:10:10.5233333"
}

To download the contents of the file, use the id of the file returned in the response.

Get a specific language model

Make a Get Language Model API request to return information on the specified language model in the specified account, such as language and the files that are in the language model.

Example response

{
    "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
    "name": "TestModel",
    "language": "En-US",
    "state": "None",
    "languageModelId": "00000000-0000-0000-0000-000000000000",
    "files": [
    {
        "id": "cccccccc-3333-4444-5555-dddddddddddd",
        "name": "hellofile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-28T11:55:34.6733333"
    },
    {
        "id": "eeeeeeee-6666-7777-8888-ffffffffffff",
        "name": "worldfile",
        "enable": true,
        "creator": "John Doe",
        "creationTime": "2018-04-28T11:55:34.86"
    }
    ]
}

To download the contents of the file, use the id of the file returned in the response.

Get all language models

Use a Get Language Models API request to return a list of all custom Azure AI language models in the specified account.

Example response

[
    {
        "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
        "name": "TestModel",
        "language": "En-US",
        "state": "None",
        "languageModelId": "00000000-0000-0000-0000-000000000000",
        "files": [
        {
            "id": "cccccccc-3333-4444-5555-dddddddddddd",
            "name": "hellofile",
            "enable": true,
            "creator": "John Doe",
            "creationTime": "2018-04-28T11:55:34.6733333"
        },
        {
            "id": "eeeeeeee-6666-7777-8888-ffffffffffff",
            "name": "worldfile",
            "enable": true,
            "creator": "John Doe",
            "creationTime": "2018-04-28T11:55:34.86"
        }
        ]
    },
    {
        "id": "aaaaaaaa-0000-1111-2222-bbbbbbbbbbbb",
        "name": "AnotherTestModel",
        "language": "En-US",
        "state": "None",
        "languageModelId": "a0a0a0a0-bbbb-cccc-dddd-e1e1e1e1e1e1",
        "files": []
    }
]

Delete a file from a language model

The Delete Language Model File request deletes the specified file from the specified Azure AI language model in the specified account.

Example response

There's no returned content when the file is deleted from the language model successfully.

Get metadata on a file from an Azure AI language model

The Get Azure AI Language Model File Data request returns the contents and metadata of the specified file from the chosen Azure AI language model in your account.

Example response

{
    "content": "hello\r\nworld",
    "id": "cccccccc-3333-4444-5555-dddddddddddd",
    "name": "Hello",
    "enable": true,
    "creator": "John Doe",
    "creationTime": "2018-04-27T20:10:10.5233333"
}

Note

The contents of this example file are the words "hello" and "world" in two separate lines.

Download a file from an Azure AI language model

The Download Azure AI Language Model File Content request downloads a text file that contains the contents of the specified file from the specified Azure AI language model in the specified account. This text file matches the contents of the file that you originally uploaded.

Example response

The response is the download of a text file with the contents of the file in the JSON format.

Feedback

Was this page helpful?

Last updated on 2026-07-14

Customize a language model with Azure AI Video Indexer

Prerequisites

Optimize your custom language model

Create a language model

Use a language model on a new video

Use a language model to reindex

Edit a language model

Rename the language model

Add files

Delete files

Delete a language model

Customize language models by correcting transcripts

Feedback

Additional resources