search

AIFanatic / google-offline-speech-recognition

This project aims to research google's offline speech recognition, from several android apps and ideally make them interoperable by replicating it on any system that supports tensorflow.
59 stars 11 forks source link

suggestion #6

Closed 0wwafa closed 1 day ago

0wwafa commented 4 days ago

if you do find / 2>/dev/null -name "*.tflite" >tflite.txt on any android phone you find TONS of models.

The most interesting ar the offline TTS and STT from google... bot in searchbox app and gboard. There is also an offline transcription mopdel in the latest pixel 8/samsung s24recorder app but it's only for english.

I think it would br great to run the on low end PCs or raspberry pi using tensorflow lite and some python magic...

biemster commented 3 days ago

The models from the recorder app should be compatible with libsoda from chrome, but this requires a not too old CPU (not more than about 10 years or so), and x86. I still did not find a soda lib that will work on rpi.

0wwafa commented 3 days ago

The models from the recorder app should be compatible with libsoda from chrome, but this requires a not too old CPU (not more than about 10 years or so), and x86. I still did not find a soda lib that will work on rpi.

I just wish there was a small and fast utility like llama.ccp to run those tflite models from a PC.

biemster commented 3 days ago

Since you are here I guess you know about https://github.com/biemster/gasr, what is blocking you from using that for your use case?

0wwafa commented 1 day ago

Since you are here I guess you know about https://github.com/biemster/gasr, what is blocking you from using that for your use case?

nothing.. I just was thinking at something more straightforward.