Speech Recognition Polyfill (Voice-to-Text)

❗ Test the extension by using speech color changer and don't forget to leave a review!

The extension's code is open source and viewable on GitHub.

If you don't care about privacy just set the provider to Google and you'll have the best overall experience (equivalent to the Speech Recognition API used in Google Chrome).

⭐ Key Features:

• Works out of the box if your first language is English, one option away for other languages.
• Extension uses WASM, no external client or app needed.
• Privacy first, vosk (local) is the default model.
• Decent Web Speech API Support through Polyfill (obviously)
• Customizable Keybind for Speech to Text anywhere, alongside the polyfill
• Many Local Models to choose from (Vosk/Whisper/Nvidia)
• Realtime Streaming/Word-to-Word Continuous Speech like Chrome (Vosk/Nvidia/AssemblyAI/Google)
• Local & Offline Transcription Support (Vosk/Nvidia/Whisper..see info part)
• Cloud/Server Transcription Support
• Extensive Per-Site and Default Customization Options
• Icon Status and Notification toasts to guide you. Toasts aren't enabled by default
• Exporting and Importing Settings
• Auto Language Detection based on data from site's and some speech models.
• Many languages to choose from when using their respected country code (support depends on the engine used).

• On first install/every update, this extension will open the options page, the default model language is English but this is easily changeable. This extension allows per-site customization and a multitude of different models to decipher language. It also has a key-bind for Speech to Text (default: Alt + A). Keep in mind that this is not a complete solution and the API doesn't have full support but it's pretty close. Speech Detection is nearly as instantaneous (depending on the model), it is now has Google Chrome's Cloud API. The extension icon color/indicator changes depending on the process so pin it to your menu to verify the extension is working as intended. A red mic/error icon does not necessarily mean your mic isn't working but rather the speech may have been cancelled by user input, missing cloud API key, or that it is unintelligible (usually its the latter).
• Some obscure websites with non-standard text input fields may not work properly with the keybind (ex: https://onlinenotepad.org/notepad), this is not a complete limitation as some may think BUT rather an annoyance that will take me several hours to make it works consistently across most sites without text repetition, deletion, replacement or other dumb input glitches. 99% of websites will work just fine right now. That's pretty much the last barrier in this extension that likely won't be fixed or atleast by me due to time constraints.
• Make sure you are using the correct mic and speak loud, slow, and clear otherwise your voice may not be detected or unintelligible. Change the default model to the cloud or slightly larger or different local ones if you experience problems with voice recognition (this may impair performance). You can also try enabling "boost microphone gain" if you are a soft speaker. If you are constantly processing audio over 1-2 mins, disable the ultimatum processing timeout feature in the settings.
• If you're using Duolingo or similar and are trying to do the speaking practice of the language that you are learning, it is recommended to set the language in the extension to the one you are learning (navigate to the site -> click extension icon -> set language then click "save for site"). This isn't required but it will significantly improve the accuracy of your speech since the model now knows the exact language you are trying to speak. (This isn't exactly necessary for every site, one example is google translate which tells us the exact language that is being used through the input box's data so auto-detect works fine). Look at the images for more help.
• If you do care about privacy and want the overall fastest and most accurate local model you'll need to use Nvidia Parakeet. Be aware that these models are >1GB. Though, I haven't tested larger Vosk Models yet so those may be good too. Nvidia Parakeet will never have model-based site overrides.
• You can also use my Speech Recognition Polyfill Userscript as an alternative for simplicity/accuracy, or cross-browser compatibility purposes, but the extension is overall better and isn't that complicated to use anyways. The userscript only supports Google's APIs.
• Android is not officially supported, it is only made available for those who want to test it.
• Userscript-based Chrome Extension

• 8GB of RAM is a minimum requirement for local models since they could easily take up a decent chuck when utilizing larger models. DO NOT make an issue about RAM usage unless there is a memory leak.
• A modern CPU/GPU is recommended, mainly for local models. Emphasis on the GPU part.
• A decent internet connection (50mbps UP/DOWN) is recommended, but especially before using offline mode since the model needs to cache itself first.
• Not being illiterate and reading the FAQ before making a review.
• Understanding that implementing the Web Speech API perfectly isn't feasible and that this extension has more support than anything else available.
• Understanding that anything can break and that I have no obligation to continuously update the extension.
• Understanding that you need a beefy PC if you want the best performance out of local models. DO NOT complain that the model you want to use is slow. Vosk is the default for a reason.

RAM is not an issue until you use local models. The extension will take ~1GB of ram on default settings and up to ~7GB if you use some of the biggest models (you don't need to use the biggest models lol). I've implemented extremely decent memory management to compensate.