/docs

en.sent.recscore

Kernel Description

Pass multiple texts and keywords to the engine (does not affect the score), and the engine feeds back the branch read by the user, and gives the total score, fluency score, completeness score, accuracy score, and the hit keywords.

Support function: English and American pronunciation distinction.
Suitable for scenarios: man-machine dialogue.

Request Parameters

Parameters	Type	Required	Description
coreType	String	true	"en.sent.recscore" means AITalk Kernel.
refText	String	true	The answer branch list, multiple branch options are separated by \| . The number of branches allowed is 2-10. Each branch supports the entire sentence, with punctuation marks, and the number of words is 2-15. See more details about symbolsText Symbol Description
keyWords	String	true	Keywords list "#" is the first-level separator, and the parts separated by "#" represent various expressions, and the relationship between them is OR. "\|" is the second-level separator, and each key in the method is expressed with "\|" Words or phrases are separated, and the relationship is AND. Note: the content of keyWords must appear in the answer branch of the refText field above
accent	int	false	Specify British or American pronunciation. 1, 2 or 3, Default value is 3. 1: Accept British pronunciation. 2: Accept American pronunciation. 3: Accept both British and American pronunciation. Now about 4000 words have been marked British and American pronunciation in our dictionary.
attachAudioUrl	number	false	Enable to return audio url or not. 0: Disable 1: Enable
result	object	false
- use_details	Int	false	Set to 1 means that each word score will be displayed.

Sample Code

{
  "coreType": "en.sent.recscore", 
  "attachAudioUrl": 1, 
  "refText": " I go to school on foot. | I walk to school. | I go to school by bus. |I go to school. ", 
  "keyWords": "go to |school | on foot # walk to | school", 
  "accent": 2,
  "result": {
    "use_details": 1
  }
}

Example of Returned Results

The results are returned using the JSON structure, Example

Introduction to Returned Results

Key	Type	Description	Recommendations
uuid	string	Unique id of audio. Save uuid in database in case of need.	For reference
audioUrl	string	Audio url. Download audio in http protocol: Just add ".mp3" at the end of the URL. Download audio in https protocol: Remove the port number, add ".mp3" at the end, and add "https://" to the prefix. e.g.https://download.cloud.chivox.com/XXXXXXXX.mp3 Tips: Chivox save audio for one week. Please save audio to your own server for long-term storage. The domain name will be different for different access areas, and the actual return shall prevail.	For reference

result	object	Recognized result
- wavetime	number	Audio duration (Unit: milliseconds)	For reference
- conf	number	Confidence score. The matching degree between the recognized reading content and an answer branch in the text passed to the scoring engine. The threshold is set by the application layer to determine whether the result is correct, generally 75, which can be adjusted up or down according to the difficulty of the question.	Suggest to display
- rec	string	Recognized branch result.	Suggest to display
- overall	number	The overall score for evaluating the identified branch options.	Suggest to display
- fluency	object	Fluency score details(It has nothing to do with the pronunciation content).
- - overall	number	Overall score of fluency.	Suggest to display
- - pause	number	Number of pauses.	For reference
- - speed	number	Speech rate, which is the number of words read per minute.	For reference
- integrity	number	Score of integrity.	Suggest to display
- accuracy	number	Score of accuracy(pronunciation evaluation of the read part).	Suggest to display
- starttime	number	The start time of the sentence in audio (Unit: milliseconds).	For reference
- endtime	number	The end time of the sentence in audio (Unit: milliseconds).	For reference
- recscore	object	Need to configure the keyWords field, see the parameter description above for details.
- - status_code	number	The returned result and description are as follows: 20000: Pass 20001: The user requests to hear the question, and it is recommended that the application layer can play the question slowly. 20002: User voice content only contains part of keywords. 20003: The user's answer is completely irrelevant, and it is recommended that the application layer prompt the user to answer again. Notes：If the conf value is less than 55 points, the status_code here in recscore is set to 20003, and mathed field will be empty.	Suggest to display
- - matched	string	When the return code is 20000、20002, the matched keywords are also returned.	Suggest to display

- details	object
- - char	string	each word text	Suggest to display
- - score	number	each word score	Suggest to display
- - accent	int	0, 1, 2 or 3: 0: Its pronunciations are not marked with American or British accent. 1: British pronunciation. 2: American pronunciation. 3: English pronunciation is the same as American pronunciation.	Suggest to display
- - start	number	The start time of the word in audio (Unit: milliseconds).	For reference
- - end	number	The end time of the word in audio (Unit: milliseconds).	For reference

- info	object
- - tipId	number	Audio quality For details, please refer to TipId field description	For reference
- - tips	string	Error message prompt When prompted "post proc failed", it may be that the audio was not successfully recorded or the engine did not detect the user’s valid voice. Prompt the user to re-record	For reference

Note: The other returned parameters that are not shown above are internal ones, which you don't need to pay much atttention to.

Suggestions for handling scoring results

Firstly, it is judged whether the value of the conf field exceeds the set threshold. If it exceeds, the content of the rec field is extracted, that is, the branch text read by the user is recognized this time. Otherwise, the user is prompted to try to read aloud again or not hear your pronunciation clearly.