Describe Image
Last updated
Last updated
Use Score Task/Task Achievement function to score describe image style questions.
In order to score describe image style questions, developers can pass a description of the image along with an audio file response from a user and the Score Task/Task Achievement function will automatically assess how closely the audio describes description of the image.
The image description provided to the API is used for guidance and not for pattern matching and instead, the task achievement AI looks for semantic similarity between the user's response and the image description. Semantic matching allows the user to get a high score even if they speak in their own words and their answer does not exactly resemble the image description string.
Let us review the below example to illustrate scoring a user's response to a prompt wherein the user is asked to describe an image:
Image description: “The image shows a busy street in a city. There are people walking on the sidewalk, and some are waiting at a bus stop. In the background, I can see tall buildings, and there’s a bright blue sky. A few cars are driving by, and it looks like a sunny day.”
User's Response:
If the user utters a relevant response as above, the developers can create a detailed report that provides the user a positive score as below:
If the user utters an irrelevant response, then based on the API, zero score can be assigned to the user and developer's can create a detailed report as shown below to communicate the user's failure. Please note that while the language scores can be non-zero, the task achievement score can be zero.
Such functionality can be built by utilizing the Score Task/Task Achievement function of the Speechace API.