jack3898 / fakeyou.ts Goto Github PK

Ttsmodels have nice methods like fetchTtsModelByName I should make sure the equivalent is available for V2v models.

Today fakeyous servers have been very slow for anonymous users and the 120 attempt limit sometimes is not enough.

It should still be 120, but the user should be able to decide the intervals and attempt counts to suit their application.

The great thing about a tsconfig is that is can be extended.

I should be able to provide a tsconfig so that users of the project have the recommended config:

E.g. in a user's project this is what their tsconfig could look like:

{
    "extends": "fakeyou.ts/tsconfig.recommended.json"
}

This tsconfig will be really minimal. It'll just make sure their module resolution is set and that they have strict null checks set to true.

The rest will be up to the user.

For the sake of getting fakeyou.ts developed fast, I started without having any git hooks.

However, now the repository has matured, I think it is time to add pre-commit and pre-push hooks.

Pre-commit will simply run a lint and type check, whereas pre-push will run all of pre-commit + unit tests.

All commits should now follow a given structure for consistency.

You should be able to fetch comments on a user's profile.

You should be able to fetch user profiles

The data property is there, but all properties should be migrated to the class and I should look at removing all of the getters (which is what it currently does)

The Product class is simply unnecessary, as the product only consists of a namespace and slug. This could be simplified into a Map. Delete the Product class.

The fakeyouPlan key in the sessionUser instance is wrong, as FakeYou return false information. This should be removed and instead the user should use fetchSubscription() instead for correct plan details.

A recent bug could have been identified if the e2e tests used the api closer to how an actual user would.

I should update the tests to use the client instance.

Static methods have caused an organisation nightmare, and should be migrated to the parent class.

More information in the v3 announcement.

#58

The dataloader seemed like a great idea at first, but is very hard to read.

More information in the v3 announcement.

#58

The explicit URL requirements was added to enforce a url check. But this could just be moved and handled by the fetch class.

Would be useful for dependabot to automatically create PR's for security updates and NPM package updates.

All classes need access to the client as their first argument. This interface will enforce this as a policy on service classes that need it.

FakeYou has rate limiting, so it is not possible to hit their api with a bunch of requests at once.

It would be really handy to be able to do the following, so that the user can leave the responsibility of getting around the rate limiter to the package:

const [audio1, audio2, audio3] = Promise.all(
    model.infer("Speech 1"),
    model.infer("Speech 2"),
    model.infer("Speech 3"),
);

This can be handled with a dataloader.

The online documentation leaves a lot to be desired!
Even though the API is simple, I think for a library project it's worth including more jsdoc to help people out

Provide the ability to fetch TTS models from a user profile

There was some duplication around the audio file to get the code done quick, but logic is repeated and could be abstracted into one class.

Must also support the cursor pagination api

To stop repeated logins, and to speed up e2e tests, the package should be able to read FAKEYOU_SESSION to auto-authenticate the user without the need to use the login() function.

The login function will still be available, and useage of it will override the cookie in the environment.

Should update the workflows to use this environment variable and run e2e tests before deploy.

When downloading the TSS model result, the error is in the title.

This relates to a recent refactor in 2.1.1 and is not a bug in 2.1.0.

This needs fixing as the hishest priority.

Currently the bot only support TTS, but V2V should be supported too!

The schemas all live in a monolithic file, they should be broken out and tidied up as it's starting to get hard to follow!

Fetch categories and provide tooling to pick models from them

Workflows should use npm ci and have more explicit build steps

This package has gone a long way without them! I should add a test runner into the build pipelines and make sure things are green before publish.

At the moment wav is the only option. Users should be able to download other formats other than just wav.

I have submitted a PR to js.org, so get an official fakeyouts.js.org domain for documentation.

I have also installed typedoc to generate the site with documentation.

The documentation though still does not contain much useful information, and I want the documentation to be A*, so I should add more information and fix the issue where some classes are not being included in the generated documentation. This now includes information about api data units and classes as a whole.

I should also refactor to named exports as I will need to provide exports in the root index file to allow the documentation to generate them in.

For queues, fakeyou.ts uses https://api.fakeyou.com/tts/queue_length however there is https://api.fakeyou.com/v1/stats/queues which contains a bit more info (including v2v queue).

I am not sure why FakeYou decided to split these into two separate endpoints! But maybe the latter one is worth implementing.

Inspired by Rust, it would be very powerful to define reusable functions that can be implemented on any class using bind().

So long as the class meets the contract of the binding, the function will work. These functions can then be reused on any class with no inheritance.

E.g. if a class has a username property, it should be easy to define an implementation function that can fetch the user so long as the class holds a username.

Dependabots PRs all failed the pipeline, and the error logs suggest the authentication cookie is not working or there might be a bug. This should be looked into before accepting any of those PRs.

Allow the ability to edit the user profile details

The Leaderboard class has getters, which creates new LeaderboardUser instances every time. This is memory inefficient and wasteful.

The getters should be removed, and replaced with values calculated when the Leaderboard class is instantiated.

Add a utility to simply return the queue length

Instead of manually implementing caching for each new feature, just cache it at the request level.

The npm package is multiple kilobytes because it's not only generating js source code but declarations and sourcemaps.

One tool I could look into is esbuild for minification, which has the handy cherry on top of being faster than TSC.

Provide the ability to fetch logged in user subscription details

Should allow for community api requests, like leaderboard, user fetching users!

Add debugging logs, like when the cache has been hit and poll statuses (and more).
Should be off by default, and opt-in.

Utils is a bit mish-mash, could do with some TLC

Test all functionality manually, update README's and publish version 3!

This package should support a modern browser environment!

• Remove toDisk
• Disable CORS checks in Rest class. Browser environments have it enabled blocking fetches to FakeYou API.
• Remove use of Buffers and use ArrayBuffer
• Change imports with node prefix to standard import e.g. node:crypto to crypto
• Change dependencies to use unpkg
• Change logging util to conditionally use chalk

CDN: unpkg.com/fakeyou.ts@latest will be where the magic happens! 🥳

The app has decent e2e coverage but some parts could be additionally unit tested!