search

asticode / go-asticoqui

Golang bindings for Coqui's speech-to-text library
MIT License
34 stars 4 forks source link

Segmentation Fault after Transcribing Some Audio #2

Open AndyEsser opened 2 years ago

AndyEsser commented 2 years ago

Good morning!

I recently switched from DeepSpeech bindings to Coqui bindings for Transcription. My Golang service has a websocket that receives WAV data that is Signed 16-bit PCM data at 16 kHz which matches the data the models I'm using was trained on.

I have witnessed some panics that has happened a couple times, but I can't find what causes it in a reproducible fashion yet. However, please see below Stack Trace - the culprit seems to be in the SpeechToText function and below:

https://pastebin.com/6DDpcj3Z

Is this a known bug/issue - or do I need to do something different - I never witnessed this issue when using the DeepSpeech bindings

asticode commented 2 years ago

Mmm this is not a known bug in the GO lib 🤔 The call to C.Model_STT seems identical between Coqui and DeepSpeech, I would go with an issue with Coqui itself. I've you tries with a more recent version of Coqui?

AndyEsser commented 2 years ago

Thanks for your response - I'm using the latest release (downloaded yesterday) for Coqui from here https://github.com/coqui-ai/STT/releases/tag/v1.0.0 - but have been unable to replicate the issue on my development Mac - it only seems to throw the problems on our Ubuntu servers - so perhaps an issue with the Linux version - I will dig a little deeper.

JRMeyer commented 2 years ago

@AndyEsser still running into this issue? If so, you should definitely post the issue and info on GitHub.com/coqui-ai/STT