AI Guardrails:Large model speech moderation

1. Service

Voice Moderation (Large Model Edition) is powered by the Qwen Moderation Large Model, trained specifically to identify content risks in audio. It provides the following voice moderation services:

• Audio and Video Media Detection_LLM Edition:

  This service moderates audio files. It uses the Tongyi Large Model to deliver enhanced speech recognition and improved risk detection. It supports 26 languages, including Chinese, English, Japanese, German, Korean, Russian, French, Portuguese, Arabic, Italian, Spanish, Hindi, Indonesian, Thai, Turkish, Ukrainian, Vietnamese, Czech, Danish, Filipino, Finnish, Icelandic, Malay, Norwegian, Polish, and Swedish.

• Social & Entertainment Live Stream Detection_LLM Version:

  This service moderates live audio streams. It uses the Tongyi Large Model to deliver enhanced speech recognition and improved risk detection. It supports 26 languages, including Chinese, English, Japanese, German, Korean, Russian, French, Portuguese, Arabic, Italian, Spanish, Hindi, Indonesian, Thai, Turkish, Ukrainian, Vietnamese, Czech, Danish, Filipino, Finnish, Icelandic, Malay, Norwegian, Polish, and Swedish.

Service selection

Billing

The large model services for Voice Moderation V2.0 support two billing methods: pay-as-you-go and .

Pay-as-you-go

After you activate the Voice Moderation V2.0 service, pay-as-you-go is the default billing method. You are billed daily for your actual usage. You incur no fees if you do not call the service.

Get started

Step 1: Activate the service

Go to the activation page to activate the Voice Moderation 2.0 service.

After you activate the Voice Moderation 2.0 service, the default billing method is pay-as-you-go. You are billed based on your actual usage, and you incur no fees if you do not use the service. After you integrate and use the API, the system automatically generates bills based on your usage. For more information, see Billing. You can also purchase a resource package. Resource packages offer tiered discounts compared with the pay-as-you-go method and are suitable for users with predictable or high usage.

Step 2: Grant permissions to a RAM user

Before you access the SDK or API, you must grant permissions to a RAM user. To authenticate API calls, use an AccessKey from your Alibaba Cloud account or a RAM user. For more information about how to obtain an AccessKey, see Obtain an AccessKey.

1. Log in to the RAM console with your Alibaba Cloud account or as a RAM administrator.

2. Create a RAM user. For more information, see Create a RAM user.

3. Grant the AliyunYundunGreenWebFullAccess system policy to the RAM user. For more information, see Grant permissions to a RAM user. You can then use the RAM user to call Content Moderation APIs.

Step 3: Install and integrate the SDK

See the Voice Moderation 2.0 SDK and Integration Guide for installation and integration instructions. The service is available in the following regions:

Step 4: Adjust moderation rules (Optional)

In the Content Moderation console, you can adjust detection rules for the Voice Moderation large model, manage text moderation rules, replicate a service, configure a custom dictionary, query detection records, and check usage.

API reference

Usage notes

You can call this API to create audio content moderation tasks. For instructions on how to construct an HTTP request, see Making native HTTP calls. You can also use an SDK to call the API. For more information, see the SDK and integration guide for audio moderation (enhanced edition) V2.0.

• API operations:

  • Submit a moderation task: VoiceModeration

  • Query a moderation task: VoiceModerationResult

  • Cancel a moderation task: VoiceModerationCancel

• Billing:

  This is a paid API. You are charged only for requests that return an HTTP status code of 200. You are not charged for requests that result in other error codes. For more information, see Billing.

• Service performance:

  Service performance	Description
  Audio file size	The audio moderation (enhanced edition) supports audio files up to 500 MB.
  Audio and video file formats	Supported audio file formats: MP3, WAV, AAC, WMA, OGG, M4A, and AMR. Supported video file formats: AVI, FLV, MP4, MPG, ASF, WMV, MOV, RMVB, and RM.
  Live audio stream	Supported protocols: RTMP, HLS, HTTP-FLV, and RTSP.
  QPS	The QPS limit for task submission is 100.
  Concurrent streams	The enhanced edition supports 50 concurrent streams by default.

Submit a moderation task

Request parameters

Response parameters

Examples

• Request example

{
    "Service": "audio_detection_byllm_global",
    "ServiceParameters": {
        "cryptType": "SHA256",
        "seed": "abc***123",
        "callback": "https://aliyun.com/callback",
        "url": "http://aliyundoc.com/test.mp3"
    }
}

• Successful response example

{
    "Code": 200,
    "Data": {
        "TaskId": "AAAAA-BBBBB",
        "DataId": "voice20240307***"
    },
    "Message": "SUCCESS",
    "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****"
}

Query task results

When a live stream moderation task is in progress, a query returns the N most recent audio slices. After the task is complete, the query returns all audio slices.

• To query a moderation task, call the VoiceModerationResult operation.

• Billing: This API operation is not billed.

• Query timeout: Query the results 30 seconds after submitting an asynchronous moderation task. The service stores results for up to 24 hours before automatically deleting them.

Response parameters

Examples

Request example

{
    "Service": "audio_detection_byllm",
    "ServiceParameters": {
        "taskId": "AAAAA-BBBBB"
    }
}

Successful response example

{
    "Code": 200,
    "Data": {
        "DataId": "voice20240307***",
        "LiveId": "liveId1****",
        "RiskLevel": "high",
        "SliceDetails": [
            {
                "StartTime": 0,
                "EndTime": 4065,
                "RiskLevel": "high",
                "Result": [
                    {
                        "Label": "political_entity",
                        "Description": "Suspected political entity",
                        "Confidence": 100.0,
                        "RiskLevel": "high",
                        "RiskWords": "WordA,WordB",
                        "RiskPositions": [
                            {
                                "EndPos": 14,
                                "RiskWord": "WordA",
                                "StartPos": 16
                            }
                        ]
                    }
                ],
                "Text": "Content Moderation product test case",
                "Url": "https://aliyundoc.com"
            }
        ]
    },
    "Message": "OK",
    "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****"
}

Cancel a moderation task

You can only cancel moderation tasks for live streams. Tasks for file types cannot be canceled.

• To cancel a moderation task, call the VoiceModerationCancel operation.

• Billing: This operation is free of charge.

Request parameters

Response parameters

Examples

• Request example

{
    "Service": "live_detection_byllm_global",
    "ServiceParameters": {
        "taskId": "AAAAA-BBBBB"
    }
}

• Successful response example

{
    "Code": 200,
    "Message": "OK",
    "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****"
}

Callback message format

Codes

This section describes the codes that the API returns. You are only billed for successful requests, which are indicated by a code of 200.