tencent cloud

APIs

문서APIsAutomatic Speech RecognitionReal-Time Speech Translation APIsReal-Time Speech Translation (WebSocket)

Real-Time Speech Translation (WebSocket)

다운로드
포커스 모드
폰트 크기
마지막 업데이트 시간: 2026-10-14 10:45:12
AI 번역

API Description

This API service uses the WebSocket protocol to recognize and translate real-time audio streams, synchronously returning the recognized and translated results to achieve real-time speech translation.
Before using this API, activate the ASR service, go to the API key management page to create a key, and generate an AppID, SecretID, and SecretKey. These credentials are used to generate a signature for API calls, and the signature is used for API authentication.

API Requirements

When integrating the real-time speech translation API, follow the requirements below.
Content
Description
Language type
Supports translation between Chinese, English, and other languages. You can set the corresponding language types through the interface parameters source and target.
audio attribute
Sample rate: 16000 Hz
Sample precision: 16 bits
Channel: mono
Audio format
pcm,wav
request protocol
wss protocol
Request URL
wss://asr.cloud.tencent.com/asr/speech_translate/?{Request parameters}
API authentication
Signature authentication mechanism. For details, see Signature Generation.
response format
Uniformly uses JSON format.
data transmission
It is recommended to send a data packet of 200 ms duration (that is, a 1:1 real-time rate) every 200 ms. The corresponding PCM size is 6400 bytes at a 16 kHz sample rate.
If the audio is sent at a rate exceeding the 1:1 real-time rate, or if the interval between audio data packets exceeds 6 seconds, the engine may encounter an error. In this case, the backend will return an error and proactively disconnect the connection.
concurrency limit
By default, the concurrency limit per account is 5. If you need a higher concurrency limit, submit a ticket for consultation.

API Call Process

The API call process consists of two stages: the handshake stage and the recognition and translation stage. In both stages, the backend returns a text message containing a JSON-serialized string. The format is described as follows:
Field Name
Type
Description
code
Integer
Status code. 0 indicates normal, and a non-zero value indicates that an error occurred.
message
String
Error description. When an error occurs, display the specific reason for the error occurrence. As the business develops or experience optimization progresses, this text may be frequently changed or updated.
voice_id
String
Globally unique identifier of the audio stream. One identifier corresponds to one WebSocket connection. User-generated (UUID recommended), with a maximum length of 128 characters.
Note: A new voice_id must be generated each time a WebSocket connection is established. If the translation or WebSocket connection is interrupted for any reason (such as connection failure, normal termination, or a translation error code), the voice_id becomes invalid, and a new voice_id must be used to restart the translation.
sentence_id
String
Unique ID of the current sentence.
result
Result
Latest speech translation result.
final
Integer
When this field returns 1, it indicates that the translation of the audio stream is complete. When it returns 2, it indicates that the synthesis of the translated text is complete.
The format of the translation result Result structure is as follows:
Field Name
Type
Description
source
String
Source language type, such as "zh", "en", "zh_en", and so on.
zh: Chinese
en: English
zh_en: Chinese-English mixed
ja: Japanese
ko: Korean
yue: Cantonese
id: Indonesian
th: Thai
ru: Russian
Example value: zh.
target
String
Target language type, such as "zh", "en", "zh_en", and so on.
zh: Chinese
en: English
zh_en: Chinese-English mixed
ja: Japanese
ko: Korean
yue: Cantonese
id: Indonesian
th: Thai
ru: Russian
Example value: en.
source_text
String
Recognized source language text content, encoded in UTF-8.
Example value: ASR speech recognition result.
target_text
String
Translated target language text content, encoded in UTF-8.
Example value: ASR speech recognition results.
start_time
Integer
Start time of the current sentence in the entire audio stream, in milliseconds.
Example value: 1000.
end_time
Integer
End time of the current sentence in the entire audio stream, in milliseconds.
Example value: 3000.
sentence_end
Boolean
Whether the current sentence ends. true indicates that the sentence ends, and false indicates that the sentence has not ended.
Example value: true.

Handshake Phase

Request Format

During the handshake stage, the client initiates a WebSocket connection request. The request URL format is as follows:
wss://asr.cloud.tencent.com/asr/speech_translate/<appid>?{Request parameters}
Here, <appid> must be replaced with the AppID of your Tencent Cloud account, which can be obtained on the API Key Management page. The format of {Request parameters} is as follows:
key1=value2&key2=value2...
Parameter description:
Parameter Name
Required
Type
Description
secretid
Yes
String
The SecretID of the Tencent Cloud registered account can be obtained on the API Key Management page.
Example value: *****Qq1zhZMN8dv0******
signature
Yes
String
API signature parameters
Example value: *****g1JfeBi%2FYnTjyjekxfDA%3D
timestamp
Yes
Integer
Current UNIX timestamp in seconds. If the value differs too much from the current time, a signature expiration error will occur.
Example value: 1745932688
expired
Yes
Integer
UNIX timestamp of the signature expiration time, in seconds. expired must be greater than timestamp, and expired - timestamp must be less than 90 days.
Example value: 1746019088
nonce
Yes
Integer
A random positive integer. It must be generated by the user and can contain up to 10 digits.
Example value: 8743357
voice_id
Yes
String
Globally unique identifier of the audio stream. One identifier corresponds to one WebSocket connection. User-generated (UUID recommended), with a maximum length of 128 characters.
Note: A new voice_id must be generated each time a WebSocket connection is established. If the translation or WebSocket connection is interrupted for any reason (such as connection failure, normal termination, or a translation error code), the voice_id becomes invalid, and a new voice_id must be used to restart the translation.
Example value: TBgVFCsxdn3i2DGt
voice_format
Yes
Integer
Speech encoding format. 1: pcm; 12: wav
Example value: 1
source
Yes
String
Source language type. Supported language types:
zh: Chinese
en: English
zh_en: Chinese-English mixed
ja: Japanese
ko: Korean
yue: Cantonese
id: Indonesian
th: Thai
ru: Russian
Example value: zh.
target
Yes
String
Target language type. The supported target languages for each source language are listed below:
zh (Chinese): zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese), id (Indonesian), th (Thai)
en (English): zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese), id (Indonesian), th (Thai)
zh_en (Chinese-English mixed): zh_en (Chinese-English mixed), zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese), id (Indonesian), th (Thai)
ja (Japanese): zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese)
ko (Korean): zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese)
yue (Cantonese): zh (Chinese), en (English), ja (Japanese), ko (Korean), yue (Cantonese)
id (Indonesian): zh (Chinese), en (English), id (Indonesian)
th (Thai): zh (Chinese), en (English), th (Thai)
ru (Russian): zh (Chinese), en (English), ru (Russian)
Note: If both source and target are set to zh_en, Chinese is automatically translated into English, and English is automatically translated into Chinese.
Example value: en.
trans_model
Yes
String
Translation model type. Supported models:
hunyuan-translation-lite: Lite edition
hunyuan-translation: Standard edition
Example value: hunyuan-translation-lite
hotword_list
No
String
Temporary hotword list: This parameter is used to improve recognition accuracy.
● Single hotword limit: "hotword|weight". A single hotword cannot exceed 30 characters (up to 10 Chinese characters). The weight ranges from 1 to 11 or is 100, for example, "Tencent Cloud|5" or "ASR|11".
● Temporary hotword list limit: Multiple hotwords are separated by commas. A maximum of 128 hotwords is supported, for example, "Tencent Cloud|10,ASR|5,ASR|11".
Note:
● When the hotword weight is set to 11, the current hotword is upgraded to a super hotword. It is recommended to set only important and mandatory hotwords to 11. Setting too many hotwords with a weight of 11 will affect the overall character accuracy rate.
● When the hotword weight is set to 100, the current hotword enables the enhanced homophone and same-tone replacement feature (only supported for 8k_zh and 16k_zh). For example, when the hotword is configured as "mizhi|100", the recognition result of "mizhi", which shares the same pinyin (mizhi) as "mizhi", will be forcibly replaced with "mizhi". Therefore, it is recommended that customers enable this feature based on their actual needs. It is recommended to set only important and mandatory hotwords to 100. Setting too many hotwords with a weight of 100 will affect the overall character accuracy rate.
● Hotwords cannot contain spaces, for example, ASR Tencent Cloud.
Example value: Voice Assistant|10
filter_dirty
No
Integer
Whether to filter out dirty words (currently effective only when the source language is zh or zh_en). The default value is 0.
0: Do not filter out dirty words; 1: Filter out dirty words; 2: Replace dirty words with "*".
Example value: 0.
filter_modal
No
Integer
Whether to filter out filler words (currently effective only when the source language is zh or zh_en). The default value is 0.
0: Do not filter out modal particles; 1: Partially filter out; 2: Strictly filter out.
Example value: 0.
filter_punc
No
Integer
Whether to filter out periods at the end of sentences (currently effective only when the source language is zh or zh_en). The default value is 0.
0: Do not filter out periods at the end of sentences; 1: Filter out periods at the end of sentences.
Example value: 0.
convert_num_mode
No
Integer
Whether to perform intelligent conversion of Arabic numerals (currently effective only when the source language is zh or zh_en). The default value is 1.
0: Do not convert, but directly output Chinese numerals; 1: Intelligently convert to Arabic numerals based on the scenario; 3: Enable math-related numeral conversion.
Example value: 1.
noise_threshold
No
Float
The noise parameter threshold defaults to 0 and ranges from -2 to 2. For some audio segments, a larger value indicates a higher likelihood of being classified as noise, while a smaller value indicates a higher likelihood of being classified as speech.
Example value: 0.
domain
No
Integer
Domain-specific name, 1: Technology; 2: Movie; 3: Song
vad_silence_time
No
int
Voice segmentation detection threshold. Silence duration exceeding the threshold is considered as sentence segmentation. Value range: 500-2000 (default: 1000), in ms. Currently only supports source language types: zh, en, and zh_en.
Example value: 1000.
max_speak_time
No
int
Forced sentence segmentation feature. Value range: 5000-90000 (unit: ms), default value: 10000. In continuous speaking without interruption, this parameter implements forced sentence segmentation (at this point the result becomes steady state, sentence_end=true). (Currently only supports source language types: zh, en, and zh_en)

signature generation

1. Sort all parameters except signature in lexicographical order, concatenate the request URL (excluding the protocol part: wss://) as the original text for signing. Here, taking appid=125922*** and secretid=*****Qq1zhZMN8dv0****** as an example to concatenate the original text for signing, the concatenated original text is:
asr.cloud.tencent.com/asr/speech_translate/125922***?expired=1673494772&nonce=1673408372&secretid=*****Qq1zhZMN8dv0******&source=zh&target=en&timestamp=1673408372&trans_model=hunyuan-translation-lite&voice_format=1&voice_id=c64385ee-3e5c-4fc5-bbfd-7c71addb35b0
2. Encrypt the original text for signing using HMAC-SHA1 with the SecretKey, and then encode it with base64. For example, for the original text from the previous step, secretkey=*****SkqpeHgqmSz*****, encrypt it using the HMAC-SHA1 algorithm and then encode it with base64:
Base64Encode(HmacSha1("asr.cloud.tencent.com/asr/speech_translate/125922***?expired=1673494772&nonce=1673408372&secretid=*****Qq1zhZMN8dv0******&source=zh&target=en&timestamp=1673408372&trans_model=hunyuan-translation-lite&voice_format=1&voice_id=c64385ee-3e5c-4fc5-bbfd-7c71addb35b0", "*****SkqpeHgqmSz*****"))
The obtained signature value is:
G8jDQBRg1JfeBi/YnTjyjekxfDA=
3. urlencode the signature value (must urlencode, URL encoding must be performed and special characters such as "+" and "=" must be encoded,otherwise occasional authentication failures may occur), then concatenate to get the final request URL:
wss://asr.cloud.tencent.com/asr/speech_translate/125922****?expired=1673494772&nonce=1673408372&secretid=A*********************************0r&source=zh&target=en&timestamp=1673408372&trans_model=hunyuan-translation-lite&voice_format=1&voice_id=c64385ee-3e5c-4fc5-bbfd-7c71addb35b0&signature=G8jDQBRg1JfeBi%2FYnTjyjekxfDA%3D

Request response

After the client initiates a connection request, the backend establishes the connection and performs signature verification. If the verification succeeds, the backend returns a confirmation message with a code value of 0 to indicate a successful handshake. If the verification fails, the backend returns a message with a non-zero code value and disconnects.
{"code":0,"message":"success","voice_id":"RnKu9FODFHK5FPpsrN"}

Translation Phase

After a successful handshake, the system enters the translation stage, where the client uploads audio data and receives translation result messages. If TTS is enabled, the client also needs to receive binary audio data.

Uploading Data

During the translation process, the client continuously uploads binary messages to the backend, with the content being binary audio stream data. It is recommended to send a data packet of 200 ms duration (that is, a 1:1 real-time rate) every 200 ms. The corresponding PCM size is 6400 bytes at a 16 kHz sample rate. If the audio is sent too fast and exceeds the 1:1 real-time rate, or if the interval between audio data packets exceeds 6 seconds, the engine may encounter errors, and the backend will return an error and actively disconnect.
After the audio stream upload is complete, the client needs to send a text message with the following content to notify the backend to end the translation.
{"type": "end"}

Receives a message

While the client is uploading data, it needs to simultaneously receive real-time translation results returned by the backend. Example of the result:
{"code":0,"message":"success","voice_id":"RnKu9FODFHK5FPpsrN","sentence_id":"0669d099-ffdf-404e-ad91-5eaff18b20df","result":{"source":"zh","target":"en","source_text":"Real time","target_text":"Real time","start_time":0,"end_time":1240,"sentence_end":false}}
{"code":0,"message":"success","voice_id":"RnKu9FODFHK5FPpsrN","sentence_id":"0669d099-ffdf-404e-ad91-5eaff18b20df","result":{"source":"zh","target":"en","source_text":"Real time voice translation","target_text":"Real time voice translation","start_time":0,"end_time":2840,"sentence_end":true}}
After the backend finishes translating all uploaded audio data, it finally returns a message with the final value of 1. After the client receives it, the client must actively disconnect the WebSocket connection.
{"code":0,"message":"success","voice_id":"CzhjnqBkv8lk5pRUxhpX","final":1}
If an error occurs during the translation process, the backend returns a message with a non-zero code value and disconnects.
{"code":6008,"message":"The client has not sent audio data for more than 15 seconds.","voice_id":"CzhjnqBkv8lk5pRUxhpX"}

Developer Resources

SDK

Tencent Cloud Speech SDK for Go:GitHub
Tencent Cloud Speech SDK for Python:GitHub
Tencent Cloud Speech SDK for Java:GitHub

SDK Call Example

Golang Example
Python Example
Java Example

Error Codes

Value
Description
6000
Excessive audio data is sent. Send at most 3 seconds of audio data within 1 second.
Note: The effect of real-time translation is to output recognition and translation results while speaking. The total duration of audio data sent within 1 second should be 1 second.
6001
The parameter is invalid.
6002
Authentication failed.
6003
The account has not enabled this service. Enable the service in the console.
6004
The resource package is exhausted. Enable postpaid billing or purchase a resource package.
6005
The service is suspended due to overdue payment. Recharge your account promptly.
6006
The account's current concurrent calls exceed the limit.
6007
Audio decoding failed. Check that the uploaded audio data format is consistent with the call parameters.
6008
The client has not sent audio data for more than 15 seconds.
6009
The client connection is disconnected.
6010
The client uploaded an unknown text message.
6011
The audio data shard is too large. Adjust the shard size according to the documentation.
6012
The request failed due to high machine load or network jitter. Initiate the request again.
Note: This issue is usually occasional. A small number of occurrences can be ignored. Initiate a new request.
6013
The request failed due to high machine load or network jitter. Initiate the request again.
Note: This issue is usually occasional. A small number of occurrences can be ignored. Initiate a new request.
7000
The request failed due to high machine load or network jitter. Initiate the request again.
Note: This issue is usually occasional. A small number of occurrences can be ignored. Initiate a new request.
이전 주제: Real-Time Speech Recognition (WebSocket)다음 주제: Data Types

도움말 및 지원

문제 해결에 도움이 되었나요?

피드백

키워드로 검색
    Enter keywords to search documentation