search

Const-me / Whisper

High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model
Mozilla Public License 2.0
8.47k stars 725 forks source link

Whisper stuck #205

Open eagerly-learn opened 9 months ago

eagerly-learn commented 9 months ago

Hi All... I'm a total newbie re: github and I;m not a coder or programmer. I just teach basic 3D printing at a small semi-rural public library. I used Dragon Naturally Speaking back in 2009 and really liked it but the disc won't load on my Windows 10 laptop. MS wants $800 for DNS which I can't afford so I'm looking for a low cost to free voice to text software. Whisper was recommended by a colleague but after downloading and clciking on the .exe file it keeps gettign stuck after press the Capture tab. I'm unsure if i can post files or screenshots here as a newbie so I thought I'd ask respectfully. Thank you.

RickArcher108 commented 9 months ago

Hi "Eagerly". Your laptop may not have enough horsepower to run Whisper. How much material do you want to transcribe? If you like, I'll do it for you. No charge. No strings attached. Contact me at rick@batgap.com and we can discuss it.

eagerly-learn commented 8 months ago

Hi, I'm crazy busy at our small library handling a multitude of challenges so my focus has to change on the fly.

I wanted to use Whisper Whisper Stuck after Moel Request Run as Administrator.docx as my voice to text so I could us a non-cloud based (downloadable) software.

DNS was great but now too pricey at $800.

I've attached what I see when I try to use Whisper on my Dell Latitude E6540 laptop running W10 with 8GB RAM.

Would you please share your suggestions of how I can correct this?

How can I ascertain the min specs for Whisper?

If my laptop lacks the specs is there another free program besides Whisper for V to T?

Thank you.

tigros commented 8 months ago

i looked up ur laptop so it's AMD FirePro W4170M which has only 2GB, almost for medium but not quite, try base or small models.

eagerly-learn commented 8 months ago

Hi tigros, thank you so much for checking on my laptop specs! I'm not a programmer so please forgive my ignorance but I'm a little lost with your suggestion of ..."try base or small models." I'm trying to figure out exactly what my next step(s) should be to make Whisper work on this laptop? Or are you intimating Whisper can't really work on this laptop and I need an updated laptop? Thank you again for your kind help.

RickArcher108 commented 8 months ago

I'm not a programmer either and Tigros has been very patient and helpful with me. You can download smaller models here: https://huggingface.co/getwavery/ggml-models/tree/main. The result will not be as accurate, but at least your laptop may be able to run the software.

eagerly-learn commented 8 months ago

Hi Rick.... I'm so grateful for everyone's kind help here on this wonderful forum. Thank you for sharing that link. I visited that site but I'm not sure which of the ones on that list I should try installing and seeing it if works? Sorry for being such a Forest Gump. :)

RickArcher108 commented 8 months ago

Try a few of the medium ones, such as ggml-medium.bin and ggml-small-distil-en.bin. If neither of those work, try the small ones.

eagerly-learn commented 8 months ago

Hi Rick. Thank you so much! I was reading here https://www.wikihow.com/Open-BIN-Files you can't just open a .bin file in Windows 10. Have you found one method of opening better than another? e.g. creating a cue file -vs- burning an image to a DVD. So appreciate your help.

On Mon, Feb 26, 2024 at 5:52 PM RickArcher108 @.***> wrote:

Try a few of the medium ones, such as ggml-medium.bin https://huggingface.co/getwavery/ggml-models/blob/main/ggml-medium.bin and ggml-small-distil-en.bin https://huggingface.co/getwavery/ggml-models/blob/main/ggml-small-distil-en.bin. If neither of those work, try the small ones.

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-1965472763, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPN5654QFJ5IY6D3GHFTYVUGZ3AVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNRVGQ3TENZWGM . You are receiving this because you authored the thread.Message ID: @.***>

tigros commented 8 months ago

Hi,

you have 4 options on that page:

ggml-base-en.bin ggml-base.bin ggml-small-en.bin ggml-small.bin

don't try to open them in windows, that's useless.

point the desktop GUI to use them.

you're welcome to try Whisperer if you have many files to process.

eagerly-learn commented 8 months ago

I really don't have a grasp on this all. I have no idea how you open these .en files once downloaded on to my laptop's hard drive except to open File Explore and find them on my C drive and click on it. That's how I've installed all my .exe files to run all sorts of open source software. Thank you.

tigros commented 8 months ago

that's not a .exe nor a .en, it's a .bin, as in binary, it's not executable, just data.

you have file extensions hidden, google how to unhide file extensions, im sure there's videos about it too.

when you run WhisperDesktop.exe it asks for model path, that's where you put the the file you just downloaded, a .bin

RickArcher108 commented 8 months ago

Create a Whisper folder in the Program Files folder on your C drive and copy everything from the zip file into it. It should look something like this: image Then you should be able to click on WhisperDesktop.exe to open it.

RickArcher108 commented 8 months ago

Tigros, today I tried to run Whisper instead of Whisperer, and got an error message, so I wanted to download a fresh version of it, and couldn't find where to do that. Please remind me.

tigros commented 8 months ago

WhisperDesktop? it's here https://github.com/Const-me/Whisper/releases

no new version since July 22.

RickArcher108 commented 8 months ago

Eagerly, note that when you open Whisper Desktop you will see something like this:

image

Click the three horizontal dots to the right of the box beneath "Model Path" and choose the ggml file you want to use, either medium or if that doesn't work, small.

Then hit OK and that'll take you to the next dialog box where you can choose the mp3 file you want to transcribe, or you can use the Audio Capture feature:

image

eagerly-learn commented 8 months ago

I've never used github before so I have a learning curve.

I can see the .bin extension in File Explorer. I just don't know the specific steps to "activate" (proper verb?) it.

Ex. I visit my Whisper folder I created to download Whisper.exe and I clikon that .exe file and them Whisper loads but I get that error message I've shared.

I really need the step by step of after I've boot up my W10 laptop on how does one "implement" (correct verb?) to enable the .bin to work?

Thank you.

On Tue, Feb 27, 2024 at 10:00 PM tigros @.***> wrote:

that's not a .exe nor a .en, it's a .bin, as in binary, it's not executable, just data.

you have file extensions hidden, google how to unhide file extensions, im sure there's videos about it too.

when you run WhisperDesktop.exe it asks for model path, that's where you put the the file you just downloaded, a .bin

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-1968102173, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPNYCTXWH3URH4ZD2I23YV2MT7AVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNRYGEYDEMJXGM . You are receiving this because you authored the thread.Message ID: @.***>

eagerly-learn commented 8 months ago

Yes, that's how I've had Whisper set up and then I get that Whisper error message.

Thank you.

On Tue, Feb 27, 2024 at 10:05 PM RickArcher108 @.***> wrote:

Create a Whisper folder in the Program Files folder on your C drive and copy everything from the zip file into it. It should look something like this: image.png (view on web) https://github.com/Const-me/Whisper/assets/131281569/a7110262-835d-4a1b-9a7e-5bc12d1e312f Then you should be able to click on WhisperDesktop.exe to open it.

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-1968106805, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPNZVKWOAZBFRH3WCNSDYV2NGJAVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNRYGEYDMOBQGU . You are receiving this because you authored the thread.Message ID: @.***>

RickArcher108 commented 8 months ago

What error message are you getting? Keep in mind that you don't "open" the bin file. You link to it from within Whisper or Whisperer. You need to do that or they won't work.

eagerly-learn commented 4 months ago

Hi Rick,

Sorry for the late reply as I've had to resolve a plethora of tech issues.

II've attached the roor message I see when i tyr to run Whisper.

I'm just looking for a voice to text software I can talk into for notes, emails as I i did easily with Dragon Naturally Speaking ver 9.

Thank you.

On Thu, Feb 29, 2024 at 2:19 PM RickArcher108 @.***> wrote:

What error message are you getting?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

RickArcher108 commented 4 months ago

You might want to try Livewriter:

image

You can get it here: https://github.com/tigros/livewriter

eagerly-learn commented 4 months ago

I'm sorry as I'm still new with Github.

Would I download and install Whisper from here?

https://github.com/tigros/livewriter/blob/main/livewriter/Whisper.dll

Thank you kindly.

On Thu, Jul 4, 2024 at 6:08 PM RickArcher108 @.***> wrote:

You might want to try Livewriter:

image.png (view on web) https://github.com/Const-me/Whisper/assets/131281569/dd44c7ae-6dd6-453c-96b4-029c3d05e168

You can get it here: https://github.com/tigros/livewriter

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-2209599053, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPN5NPGCNDUHR3FCAV5DZKXBUJAVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMBZGU4TSMBVGM . You are receiving this because you authored the thread.Message ID: @.***>

RickArcher108 commented 4 months ago

You can get it from https://github.com/Const-me/Whisper/releases

I find GitHub confusing too. And Whisper can be tricky to install. If all you want to do is speech to text, I think you'll find Live Writer easier. You can get it here: https://github.com/tigros/livewriter

eagerly-learn commented 4 months ago

Hi Rick,

I'm so new to all this a sI used the DNS disc and would be up and running.

Shameful ow much M$$$$$ wants for it noe they they bought DNS.

Would you be so kind as to share with me which of these files in that Whisper folder I would download and install?

[image: image.png] Thank you ever so much.

Charles.

On Thu, Jul 4, 2024 at 6:43 PM RickArcher108 @.***> wrote:

I believe you can get it from https://github.com/openai/whisper.

I find GitHub confusing too. And Whisper can be tricky to install. If all you want to do is text to speech, I think you'll find Live Writer easier. You can get it here: https://github.com/tigros/livewriter

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-2209619274, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPNY47ZPPHPNJCASMWCLZKXF2HAVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMBZGYYTSMRXGQ . You are receiving this because you authored the thread.Message ID: @.***>

RickArcher108 commented 4 months ago

To me, DNS means Domain Name Server. Other than that I don't know what you mean. You would download WhisperDesktop.zip. You'll also need to download one of the ggml files here: https://huggingface.co/getwavery/ggml-models/tree/main - base, small, medium, or large, depending on what your computer can handle. The larger the file, the slower it runs, but the more accurate the results. You'll also need to install Python: https://www.python.org/downloads/.

There are people around here who are far more technically astute than I. I'll keep trying to help, but you may run into something that's beyond my minimal expertise.

eagerly-learn commented 4 months ago

Sorry. DNS is Dragon Naturally Speaking. It was developed by H/W team and worked great. You installed from a disk and a Dilbert cartoon spent 45 minutes having you say words to train it. $100 and worth it but I can't get Windows 10 to read the disk and now M$$$ wants $400-500 for it.

I don't have sophisticated needs - just transcribe my voice as I dictate into written text.

I was hoping if the Windows works didn't have one that Linux might. I looked at Ubuntu Studio but that's not a feature.

I'll try my best with your kind help.

Honestly, did everyone suddenly become touch typists?

Thank you!

On Fri, Jul 5, 2024 at 2:25 PM RickArcher108 @.***> wrote:

To me, DNS means Domain Name Server. Other than that I don't know what you mean. You would download WhisperDesktop.zip. You'll also need to download one of the ggml files here: https://huggingface.co/getwavery/ggml-models/tree/main - base, small, medium, or large, depending on what your computer can handle. The larger the file, the slower it runs, but the more accurate the results. You'll also need to install Python: https://www.python.org/downloads/.

There are people around here who are far more technically astute than I. I'll keep trying to help, but you may run into something that's beyond my minimal expertise.

— Reply to this email directly, view it on GitHub https://github.com/Const-me/Whisper/issues/205#issuecomment-2211259669, or unsubscribe https://github.com/notifications/unsubscribe-auth/BF6TPNY6CG5GHVLXJW5P34LZK3QH7AVCNFSM6AAAAABC45RDHWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMJRGI2TSNRWHE . You are receiving this because you authored the thread.Message ID: @.***>

RickArcher108 commented 4 months ago

As I said earlier, if all you want to do is speech to text, don't bother with Whisper. It's overkill. Install Live Writer. It's easier to install. You can get it here: https://github.com/tigros/livewriter. You'll still need to download one of the ggml files here: https://huggingface.co/getwavery/ggml-models/tree/main - base, small, medium, or large, depending on what your computer can handle.