Accessibility platform

1. A method comprising:

receiving, by one or more processors, a request for a markup language document;

obtaining, by the one or more processors, the markup language document;

processing a first portion of the markup language document with a machine learning technique to generate a first dictation corresponding to the first portion of the markup language document, the machine learning technique being trained to establish a relationship between a plurality of training markup language documents and training dictations corresponding to the training markup language documents, the machine learning technique being trained by:

obtaining a first batch of training data comprising a first set of the plurality of training markup language documents and a first set of the corresponding training dictations;

processing the first set of training markup language documents with the machine learning technique to generate estimated dictations for the training markup language documents;

computing a loss function based on a deviation between the estimated dictations for the training markup language documents and the corresponding training dictations; and

updating parameters of the machine learning technique based on the computed loss function; and

replacing the first portion of the markup language document with the first dictation.

2. The method of claim 1 , wherein the markup language document comprises a webpage that is requested by a web browser application implemented on a client device.

3. The method of claim 1 , wherein replacing the first portion comprises replacing the first portion of the markup language document with markup language corresponding to the first dictation.

4. The method of claim 1 , further comprising:

transcribing a second portion of the markup language document to modify the markup language document to include the first dictation generated using the machine learning technique and the transcribed second portion of the markup language document.

5. The method of claim 1 , wherein the machine learning technique comprises a neural network.

6. The method of claim 5 , wherein the neural network comprises at least one of a recurrent neural network or a bi-directional long short-term memory (LTSM) neural network.

7. The method of claim 1 , wherein replacing the first portion comprises:

modifying the first portion to hide content in the first portion from being displayed; and

adding markup language corresponding to the first dictation.

8. The method of claim 1 , further comprising generating a first of the dictations corresponding to a first of the plurality of training markup language documents by:

displaying a training webpage based on the first training markup language document;

receiving input selecting a portion of the displayed training webpage;

identifying a portion of the first training markup language document corresponding to the selected portion;

recording speech that reads out the selected portion of the displayed training webpage;

transcribing the recorded speech to generate the first dictation corresponding to the first training markup language document; and

associating the first dictation with the identified portion of the first training markup language document.

9. The method of claim 1 , further comprising:

computing a confidence score for the machine learning technique to generate a first dictation corresponding to the first portion of the markup language document;

comparing the confidence score to a threshold; and

replacing the first portion of the markup language document with the first dictation in response to determining that the confidence score exceeds the threshold.

10. The method of claim 9 , further comprising:

computing a second confidence score for the machine learning technique to generate a second dictation corresponding to a second portion of the markup language document;

comparing the second confidence score to the threshold; and

preventing replacing the second portion with the second dictation in response to determining that the second confidence score fails to exceed the threshold.

11. A method comprising:

receiving, by one or more processors, a request for a markup language document

obtaining, by the one or more processors, the markup language document; processing a first portion of the markup language document with a machine learning technique to generate a first dictation corresponding to the first portion of the markup language document, the machine learning technique being trained to establish a relationship between a plurality of training markup language documents and training dictations corresponding to the training markup language documents;

replacing the first portion of the markup language document with the first dictation, wherein replacing the first portion of the markup language document with the first dictation generates a first version of the markup language document, and further comprising:

transcribing the first portion of the markup language document to generate a second version of the markup language document;

applying a Monte Carlo simulation to randomly present the first version of the markup language document to a first group of users and randomly presenting the second version of the markup language document to a second group of users; and

generating a survey indicating task completion of the first group of users and the second group of users.

12. The method of claim 11 , further comprising selecting a dictation for the first portion of the markup language document from the first and second versions of the markup language document based on the survey.

13. A system comprising:

one or more processors coupled to a memory comprising non-transitory computer instructions that when executed by the one or more processors perform operations comprising:

receiving a request for a markup language document;

obtaining the markup language document;

processing a first portion of the markup language document with a machine learning technique to generate a first dictation corresponding to the first portion of the markup language document, the machine learning technique being trained to establish a relationship between a plurality of training markup language documents and training dictations corresponding to the training markup language documents, the machine learning technique being trained by:

obtaining a first batch of training data comprising a first set of the plurality of training markup language documents and a first set of the corresponding training dictations;

processing the first set of training markup language documents with the machine learning technique to generate estimated dictations for the training markup language documents;

computing a loss function based on a deviation between the estimated dictations for the training markup language documents and the corresponding training dictations; and

updating parameters of the machine learning technique based on the computed loss function; and

replacing the first portion of the markup language document with the first dictation.

14. The system of claim 13 , wherein the markup language document comprises a webpage that is requested by a web browser application implemented on a client device.

15. The system of claim 13 , wherein replacing the first portion comprises replacing the first portion of the markup language document with markup language corresponding to the first dictation.

16. The system of claim 13 , further comprising operations for:

transcribing a second portion of the markup language document to modify the markup language document to include the first dictation generated using the machine learning technique and the transcribed second portion of the markup language document.

17. A non-transitory computer readable medium comprising non-transitory computer-readable instructions for performing operations comprising:

receiving a request for a markup language document;

obtaining the markup language document;

processing a first portion of the markup language document with a machine learning technique to generate a first dictation corresponding to the first portion of the markup language document, the machine learning technique being trained to establish a relationship between a plurality of training markup language documents and training dictations corresponding to the training markup language documents, the machine learning technique being trained by:

obtaining a first batch of training data comprising a first set of the plurality of training markup language documents and a first set of the corresponding training dictations;

processing the first set of training markup language documents with the machine learning technique to generate estimated dictations for the training markup language documents;

computing a loss function based on a deviation between the estimated dictations for the training markup language documents and the corresponding training dictations; and

updating parameters of the machine learning technique based on the computed loss function; and

replacing the first portion of the markup language document with the first dictation.

18. The non-transitory computer readable medium of claim 17 , wherein the markup language document comprises a webpage that is requested by a web browser application implemented on a client device.

19. The non-transitory computer readable medium of claim 17 , wherein replacing the first portion comprises replacing the first portion of the markup language document with markup language corresponding to the first dictation.