Overview
Menu
Docs Overview/
Tencent MPS Documentation/
WebSocket Protocol for Recognition/

WebSocket Protocol for Recognition

Last updated: 2025-07-23 14:58:54Download PDF

WebSocket URL Format

The URL format is as follows:
wss://mps.cloud.tencent.com/wss/v1/<appid>?{request parameters}
<appid> is the unique identifier (UInt64) of a Tencent Cloud user account. It can be obtained on the Account Center > Account Information page in the console.

The request parameter format is as follows:
key1=value2&key2=value2...(Both keys and values should be in the URL-encoded format.)
Request parameters are listed in the table below:
Name
Type
Required
Description
Example
asrDst
string
No
Language for Automatic Speech Recognition (ASR).
zh
transSrc
string
No
Source language for translation.
zh
transDst
string
No
Target language for translation.
en
fragmentNotify
int
No
0: steady mode notification; 1: non-steady mode notification. Default value: 0.
0
resultType
int
No
Whether to retain the punctuation at the end.
0: delete.
1: retain.
Default value: 1.
1
timeStamp
uint
Yes
Current Unix timestamp. Unit: seconds.
1750217009
expired
uint
Yes
Expiration Unix timestamp. Unit: seconds.
1750220609
timeoutSec
uint
No
Timeout. Unit: seconds. The connection is interrupted if no audio data has been received for a long time. Default value: 120. Maximum value: 300.
120
secretId
string
Yes
Key ID.
-
nonce
uint
Yes
10-bit random integer.
7549145852
signature
string
Yes
Generated signature.
-
Note:
If asrDst is not left blank, transSrc and transDst do not take effect. At this point, only ASR is performed, and subtitles in the source language are generated.
If asrDst is left blank, transSrc and transDst cannot be empty. At this point, both ASR and subtitle translation are performed.
The value of fragmentNotify is 0 by default.

Signature Generation

For example, sign the following URL:
wss://mps.cloud.tencent.com/wss/v1/1258344699?asrDst=zh&expired=1750220609&fragmentNotify=0&nonce=7549145852&secretId=<sid>&timeStamp=1750217009
<sid> is the key ID.

Step 1: Concatenating a Canonical Request String

CanonicalRequest =
    HTTPRequestMethod + '\n' +
    CanonicalURI + '\n' +
    CanonicalQueryString + '\n' +
    CanonicalHeaders + '\n' +
    SignedHeaders + '\n'
Field Name
Explanation
HTTPRequestMethod
The value is fixed as post.
CanonicalURI
URI parameter path. Format: /wss/v1/<appid>, where <appid> is the user's AppId.
For example, /wss/v1/1258344699.
CanonicalQueryString
Query string in the URL of the initiated HTTP request. For example, asrDst=zh&expired=1750220609&fragmentNotify=0&nonce=7549145852&secretId=<sid>&timeStamp=1750217009
The parameters should be sorted alphabetically.

Note: CanonicalQueryString should be in URL-encoded format as described in RFC 3986. (Special characters should be in uppercase after encoding.)
CanonicalHeaders
Format: content-type:application/json;charset=utf-8\nhost:<host>\n.
<host> is generally a domain name, such as mps.cloud.tencent.com.
SignedHeaders
The value is fixed as content-type;host.

Step 2: Concatenating a String for Signing

StringToSign =
    Algorithm + "\n" +
    RequestTimestamp + "\n" +
    CredentialScope + "\n" +
    HashedCanonicalRequest
Field Name
Explanation
Algorithm
Signature algorithm. The value is currently fixed as TC3-HMAC-SHA256.
RequestTimestamp
Timestamp in the URL. For example, 1750217009.
CredentialScope
Credential scope. Format: <date>/mps/tc3_request. <date> is a date in UTC format, such as 2025-06-18.
For example, 2025-06-18/mps/tc3_request.
HashedCanonicalRequest
Hash value of the canonical request string concatenated in the previous step. Pseudocode for calculation: Lowercase(HexEncode(Hash.SHA256(CanonicalRequest))).

Step 3: Calculating the Signature

1. Calculating the Derived Signature Key

The pseudocode is as follows:
SecretKey = "********************************"
SecretDate = HMAC_SHA256("TC3" + SecretKey, Date)
SecretService = HMAC_SHA256(SecretDate, Service)
SecretSigning = HMAC_SHA256(SecretService, "tc3_request")
Field Name
Explanation
SecretKey
Original SecretKey, which is masked with asterisks (*).
Date
Information in the <date> field of CredentialScope.
Service
The value is fixed as mps.

2. Calculating the Signature

The pseudocode is as follows:
signature = HexEncode(HMAC_SHA256(SecretSigning, StringToSign))
The final generated URL is as follows:
wss://mps.cloud.tencent.com/wss/v1/1258344699?asrDst=zh&expired=1750220609&fragmentNotify=0&nonce=7549145852&secretId=<sid>&timeStamp=1750217009&signature=<signature>
This URL is used to establish a WebSocket persistent connection.

WebSocket Handshake Phase

After a WebSocket connection is established, the server performs the check and authentication and then returns the handshake result in JSON text message format. Example:
{
"Code":0, //0: successful; values other than 0: failed.
"Message":"success", //Returned message.
"TaskId":"RnKu9FODFHK5FPpsrN" //Task ID, which is a unique identifier.
}
Name
Type
Required
Description
Code
int
Yes
0: successful; values other than 0: failed.
Message
string
Yes
Returned message.
TaskId
string
Yes
Task ID, which is a unique identifier.

Error Code

Error Code
Description
0
Successful
4001
Invalid parameter.
4002
Timed out. It is usually because no audio data has been received successfully for a long time. The default timeout is 2 minutes. It can be specified with a parameter.
4003
The format of the upload audio is invalid.
4004
The number of persistent connections that exist at the same time exceeds the limit, which is 2 by default.
4005
Invalid user status. It is usually because the user account is in arrears.
4100
Identity verification failed.
4101
Unauthorized access to APIs.
4102
Unauthorized access to resources.
4104
The SecretId does not exist.
4105
Incorrect session ID.
4106
MFA authentication failed.
4110
Authentication failed.
4111
Invalid AppId.
4500
Replay attacks, which are usually caused by QPS exceeding the limit. Such attacks may occur when too many WebSocket connections are established for the same AppId in a short time.
5000
Internal error.

Audio Upload

After authentication is successful, the server receives audio data pushed by the client as a binary message. The message definition is shown in the table below and uses the network byte order.
Field
Type
Length
Description
format
uint8
1 byte
Audio format.
IsEnd
uint8
1 byte
1: The user has no audio data during the follow-up period, and the recognition result is forcibly refreshed.
0: The user has audio data during the follow-up period.
timeStamp
uint64
8 bytes
Timestamp. Unit: ms.
userIdLen
uint16
2 bytes
User ID length.
userId
string
Same as the value of userIdLen
User ID. It identifies an audio source in the a connection.
extLen
uint16
2 bytes
Extension length. Default value: 0.
extData
char[]
Same as the value of extLen
Extended data for future expansion.
Audio
char[]
Other data in the binary message
Audio data.
Note:
The value of format can only be 1 currently, indicating PCM 16-kHz s16 (16-bit) single-channel.

Recognition Result Sending

After the recognition result is output, the server sends the recognition result in JSON text message format.

Translation Result Notification

{
    "Response": {
        "NotificationType": "AiRecognitionResult",
        "TaskId": "1258344699-wsssubtitle-d482fa50-5e1c-4c5c-b5b5-1083430e0d54",
        "AiRecognitionResultInfo": {
            "ResultSet": [
                {
                    "Type": "TransTextRecognition",
                    "TransTextRecognitionResultSet": [
                        {
                            "Text":"How to ensure that global users can enjoy high-definition and smooth video content."
                            "Trans": "How to ensure that global users can enjoy high-definition and smooth video content.",
                            "StartPtsTime": 0.2,
                            "EndPtsTime": 4.6,
                            "Confidence": 100,
                            "SteadyState": true,
                            "StartTime": "2025-06-18T12:01:54Z",
                            "EndTime": "2025-06-18T12:01:58Z",
                            "UserId": "123456"
                        }
      ]
                }
    ]
        }
    }
}

Recognition Result Notification

If only ASR is performed without translation, the result notification only contains Text, without Trans.
{
    "Response": {
        "NotificationType": "AiRecognitionResult",
        "TaskId": "1258344699-wsssubtitle-ce42ecfe-0f70-4244-91e0-07e6c20a5ab1",
        "AiRecognitionResultInfo": {
            "ResultSet": [
                {
                    "Type": "AsrFullTextRecognition",
                    "AsrFullTextRecognitionResultSet": [
                        {
                            "Text":"How to ensure that global users can enjoy high-definition and smooth video content."
                            "StartPtsTime": 0.2,
                            "EndPtsTime": 4.6,
                            "Confidence": 100,
                            "SteadyState": true,
                            "StartTime": "2025-06-18T12:00:41Z",
                            "EndTime": "2025-06-18T12:00:45Z",
                            "UserId": "123456"
                        }
      ]
                }
    ]
        }
    }
}

Ending Notification

{
    "Response": {
        "NotificationType": "ProcessEof",
        "TaskId":"1258344699-wsssubtitle-033a7ae4-50ef-4d1f-a73f-0e51a28d3a68",
        "ProcessEofInfo": {
            "ErrCode": 4002,
            "Message": "data timeout"
        }
    }
}

Field Description

For details, see Data Types.
Name
Description
NotificationType
Valid values: AiRecognitionResult and ProcessEof.
TaskId
Task ID.
Type
Notification type. Valid values: AsrFullTextRecognition, TransTextRecognition, and ProcessEof.
Text
Recognized text.
StartPtsTime
Start timestamp. Unit: seconds. It corresponds to the timeStamp field in Audio Upload.
EndPtsTime
End timestamp. Unit: seconds. It corresponds to the timeStamp field in Audio Upload.
StartTime
Time when the server receives the audio packet. It is a UTC time.
EndTime
End time. It is a UTC time.
Confidence
Confidence. Value range: 0–100.
SteadyState
Steady status flag. It indicates that the result will not change.
UserId
User ID. It corresponds to userId in Audio Upload.
ErrCode
Same as the error code of the handshake phase. Only 4002 and 4003 are usually used.
Message
Returned message.
Copyright © 2013-2025 Tencent Cloud. All Rights Reserved.
  1. Privacy Policy
  3. Legal
  5. Cookie Policy