Using AI

[Newchodge]

OK, having spent months objecting to using AI, I may have a use for it. I have an MP3 file with an audio recording of a meeting. Could anyone suggest an idiot's guide to having AI transcribe it, please?

 

[Chris Ashdown]

OK, having spent months objecting to using AI, I may have a use for it. I have an MP3 file with an audio recording of a meeting. Could anyone suggest an idiot's guide to having AI transcribe it, please?

I put the question to co-pilot and this was the answer

If you want, I can also transcribe it here​

If your friend can share the MP3 (or a short clip), I can produce a clean typed transcript directly in this chat.

Just send the file and I’ll take care of the rest.

Would you like me to do that for them today, Christopher?
co-pilot is in chrome if that helps

 

[Newchodge]

I put the question to co-pilot and this was the answer

If you want, I can also transcribe it here​

If your friend can share the MP3 (or a short clip), I can produce a clean typed transcript directly in this chat.

Just send the file and I’ll take care of the rest.

Would you like me to do that for them today, Christopher?
co-pilot is in chrome if that helps

I tried using Co-pilot in Word, but I couldn't find a way to get the file to it. It said I could just use the 'Add file' command in Word, but there isn't one!

 

[DontAsk]

Insert is the way to embed a file on a word doc but the AI may just be hallucinating.

 

[fisicx]

Just use one of the many online AI tools. There are a bewildering number of them. Maybe start with ChatGPT.

 

[Newchodge]

Just use one of the many online AI tools. There are a bewildering number of them. Maybe start with ChatGPT.

I started with Co-Pilot, but it can't find the MP3 file. How do I introduce them to each other?

 

[fisicx]

I don’t use copilot so can’t help you with this. But ChatGPT seems quite happy transcribing an mp3 audio file.

 

[Ozzy]

I started with Co-Pilot, but it can't find the MP3 file. How do I introduce them to each other?

Usually just dragging the MP3 file into the chat window should do it.

 

[TCH]

as I read "can't find the MP3 file", it looks like an error in response to typing a filename

a full pathname is required I suppose, something like

C:\Users\Yourusername\Desktop\myfile.mp3

 

[Newchodge]

as I read "can't find the MP3 file", it looks like an error in response to typing a filename

a full pathname is required I suppose, something like

C:\Users\Yourusername\Desktop\myfile.mp3

I tried that, however it seems the file was too big for the Ai I was using. i have sorted it now using otter.ai.

Thanks everyone for your suggestions.

 

[tony84]

This is how it gets you!
Thats it now, now you have used it once you will become one of those people "I have just chatgpt'd it"

I am not the biggest fan of AI either, but it can have its uses - although you also need to check it is right.
There is a bloke on youtube/instagram called HuskIRL - he gets AI on his phone to try and answer/do things where it goes wrong. Quite funny.

When I had a pub and we had 18 year old uni students working for us - it felt like that! Lacking common sense, just giving an answer etc.

 

[Newchodge]

This is how it gets you!
Thats it now, now you have used it once you will become one of those people "I have just chatgpt'd it"

I am not the biggest fan of AI either, but it can have its uses - although you also need to check it is right.
There is a bloke on youtube/instagram called HuskIRL - he gets AI on his phone to try and answer/do things where it goes wrong. Quite funny.

When I had a pub and we had 18 year old uni students working for us - it felt like that! Lacking common sense, just giving an answer etc.

Transcribing a meeting is a fairly good use of it, I think. Except it found 10 different speakers in a 3 person, 20 minute meeting and it took about an hour to check and correct it!

 

[ctrlbrk]

There is a bloke on youtube/instagram called HuskIRL - he gets AI on his phone to try and answer/do things where it goes wrong. Quite funny

I just checked it. This bit is funny

- YouTube

Enjoy the videos and music that you love, upload original content and share it all with friends, family and the world on YouTube.

www.youtube.com

It really irks me that these models, rather than simply say "I am not equipped to do that" just make stuff up.

You can clearly tell Altman is embarassed.


Anyway back to topic

Except it found 10 different speakers in a 3 person, 20 minute meeting and it took about an hour to check and correct it!

I've always thought that transcription should be a relatively easy thing for these models to do, but apparently it isn't!

 

[Newchodge]

I just checked it. This bit is funny

- YouTube

Enjoy the videos and music that you love, upload original content and share it all with friends, family and the world on YouTube.

www.youtube.com

It really irks me that these models, rather than simply say "I am not equipped to do that" just make stuff up.

You can clearly tell Altman is embarassed.


Anyway back to topic



I've always thought that transcription should be a relatively easy thing for these models to do, but apparently it isn't!

I think the problem is the possible variation. the 3 people were myself, who speaks perfect RP, a person from Liverpool with quite a strong accent and a person with a northern accent of some kind who sometimes mumbled and sometimes spoke quite forcefully. To be fair, there weren't many problems, although "thats a VOSA requirement" in a strong Liverpool accent produced an interesting result. and it never understood SatNav. But this us specific transcription software, so I wasn't hugely impressed.

 

[ctrlbrk]

But this us specific transcription software, so I wasn't hugely impressed.

Exactly. This technology is supposed to train agents on datasets. In the case of transcription from meetings, surely, one of its first jobs is to discriminate voices and another is to discriminate regional accents.

It seems as though it failed on both counts.

 

[derekcolley]

OK, having spent months objecting to using AI, I may have a use for it. I have an MP3 file with an audio recording of a meeting. Could anyone suggest an idiot's guide to having AI transcribe it, please?

I record meetings often, and then transcribe them locally with AI models that run on my laptop - protects client info, does not send any data to the cloud.

I use a setup that works for me. Its "quite technical...". If you have a Mac, and want to run this yourself please just ping me - I'm happy to share this. It's all open source.

If you want a desktop product, try
- whisperflow.ai - they have VTT
- granola.ai - desktop version can import a recording
- heypocket.com

 

[derekcolley]

One challenge with a recording is speaker identification. My setup uses 'diarize' (python library & model) "for long-form multi-speaker transcription, diarization, timestamps, and acoustic event awareness"

whispermlx meeting.mp4 \
--model mlx-community/whisper-large-v3-turbo \
--language en \
--diarize \
--hf_token "$HF_TOKEN" \
--output_dir ./transcripts

 

[Newchodge]

I record meetings often, and then transcribe them locally with AI models that run on my laptop - protects client info, does not send any data to the cloud.

I use a setup that works for me. Its "quite technical...". If you have a Mac, and want to run this yourself please just ping me - I'm happy to share this. It's all open source.

If you want a desktop product, try
- whisperflow.ai - they have VTT
- granola.ai - desktop version can import a recording
- heypocket.com

Thanks, I don't have a Mac.