Audio to text from scanner recording

[hotpocket]

Hey guys, I’ve got a question/idea that I didn’t know if it was possible or how hard it would be to do. And if this isn’t exactly the right section of someone knows where I should post it let me know and I’ll be happy to move it to the correct place

So what I’m wondering is if there is a way to sort of run speech-to-text on an audio recording or stream locally.

My thought process is for a recording after the fact to take the MP3 file from either the calls platform or a local recording, pop it into a speech to text service to get the output.

My other thought is relating to real time capture of a scanner using DSDplus. The flow I’m thinking might work is:
DSDplus -> VB-Cable A -> speech-to-text service -> text file.

Does anyone know if this would work, or even have any recommendations for a good speech-to-text service that is accurate?

(I know the audio would have to be really clear and what I’m wanting to convert would be the station alerting audio for my local fire department which usually doesn’t have any kind of distortion from what I’ve heard)

 

[jtwalker]

Pretty sure Azure (Microsoft) has a audio-to-text engine, but it won’t be free for much more than a little experimentation.

There are a lot of terms used in public safety radio that aren’t in the Webster dictionary, so success rate would be questionable for anything like this.

Might need a translation engine that learns with help from a human explaining some of these terms.

 

[mmckenna]

I believe Google has a tool for that.

Here's the problem though:

It's a 'best effort' thing. The software that does this doesn't handle back ground noise, poor audio quality, or narrow audio bandwidth well. We run it on our voice mail system at work, and we take voicemails to text and send it as e-mail. We include a .wav file so people can check the translation.

It really tends to hack things up when audio quality is a challenge. Even good clear audio from a scanner/receiver is going to have very limited bandwidth audio, back ground noise, etc. and you'll find it results in a lot of errors.

Try it out, but just don't expect it to be perfect.

 

[RobDLG]

I posted details of some audio transcription tests here:

Automatically transcribing DSD FME wav files

 

[footage]

For Mac users, MacWhisper is great. High accuracy for studio-recorded speech; medium-to-low accuracy for scanner recordings, unless you're transcribing a nearby repeater with excellent audio quality. It's very good for a rough transcript but really requires human editing for completeness and accuracy. I run all my recordings thru it.

 

[JustinWHT]

Text to speech fail

https://community.ui.com/questions/Speech-to-Text-FAIL/8af2baf4-536d-40ec-a1b0-cd3259eacfc6

 

[nickwilson159]

I posted details of some audio transcription tests here:

Automatically transcribing DSD FME wav files

Much appreciated, sir. I have been playing around with the Whisper library in Python, but hadn't gotten to evaluating the various models yet.