Cross-browser seems not work as expected

Riley-Brown / react-speech-to-text

React Hooks for Speech to Text

MIT License

118 stars 31 forks source link

Cross-browser seems not work as expected #4

Closed ilhamwahabi closed 3 years ago

ilhamwahabi commented 3 years ago

If we enable crossBrowser and provide googleApiKey seems it's still not using Google Cloud for the transcription, so:

we can see it through network tab that there is no request happened. Which is supposed to happen
if we change googleCloudRecognitionConfig there is nothing changed

Reproduction (I'm just using example code): https://codesandbox.io/s/nifty-raman-thfsj?file=/src/App.js

Riley-Brown commented 3 years ago

@iwgx Are you testing on chrome? If so, chrome will still use the speech recognition API since it's available, the cross-browser is intended for other browsers such as Firefox, Edge, Brave, etc.

Riley-Brown commented 3 years ago

I also added a speechRecognitionProperties param you can pass to the hook the customize the settings for chrome speech recognition API such as language code https://github.com/Riley-Brown/react-speech-to-text#speechrecognitionproperties

ilhamwahabi commented 3 years ago

I see, thanks for the confirmation

Previously I think if we pass the crossBrowser params, it will override the SpeechRecognition so It will use GCP in all browser. Since I think GCP will more reliable

Riley-Brown commented 3 years ago

The reasoning for this is because Chrome is the most popular browser so more users would use the built-in Chrome speech recognition API which is free and provides faster speech-to-text than using the GCP method.

I can add a useOnlyGoogleCloud param to override this if you think it's needed

ilhamwahabi commented 3 years ago

I thinks that's needed, since we can control if we want:

only use GCP
only use Speech Recognition (and handle error in unsupported browser)
use Speech Recognition with GCP as fallback

BTW the reliability that I mean is in terms of transcription. I think since GCP process it in their server, it will have higher accuracy than processed the browser (although of course it will be faster)

Riley-Brown commented 3 years ago

Just published version 0.5.0 with useOnlyGoogleCloud param to force GCP on chrome browsers

Also, this is from the speech recognition docs

Note: On some browsers, like Chrome, using Speech Recognition on a web page involves a server-based recognition engine. Your audio is sent to a web service for recognition processing, so it won't work offline.

So I would think Chrome built in speech API would use the same engine that google cloud speech to text uses so the results should be the same

ilhamwahabi commented 3 years ago

Thanks for the update

And I just learn about that, will investigate further if there is different between them