How to Install & Use Whisper AI Voice to Text
How to Install & Use Whisper AI Voice to Text
In this step-by-step tutorial, learn how to transcribe speech into text using OpenAI’s Whisper AI. Whisper AI is an AI speech recognition system that can transcribe and translate audio files in approximately 100 different languages.
📚 RESOURCES
- Install Python: https://www.python.org/
- Install PyTorch: https://pytorch.org/get-started/locally/
- Install Chocolatey: https://chocolatey.org/
⌚ TIMESTAMPS
00:00 Introduction
00:40 Install overview
01:00 Install Python
02:31 Install PyTorch
03:55 Install Chocolatey package manager
04:53 Install ffmpeg
05:28 Install Whisper AI
05:59 Transcribe one file
07:18 Output files
07:58 Transcribe multiple files
08:39 Available models
09:51 Transcribe in other languages
10:31 Translate to English
11:06 Help
11:40 Quality
12:04 Uninstall
12:14 Wrap up
📺 RELATED VIDEOS
- Run Whisper AI in the cloud for free using Google Colab: • Best FREE Speech to Text AI - Whisper AI
😢 Uninstall instructions:
- Uninstall Whisper AI
In command prompt, enter:
pip uninstall openai-whisper
- Uninstall ffmpeg
In command prompt, enter:
choco uninstall ffmpeg
- Uninstall Chocolatey
In File Explorer, delete the folder:
“C:\ProgramData\chocolatey”
- Uninstall PyTorch
In Command Prompt, enter:
Pip3 uninstall torch torchvision torchaudio
- Uninstall Python
Go to Installed Apps in Windows Settings, search for Python and Python Launcher, click the three dots, and then uninstall.
📩 NEWSLETTER
- Get the latest high-quality tutorial and tips and tricks videos emailed to your inbox each week: https://kevinstratvert.com/newsletter/
🔽 CONNECT WITH ME
- Official web site: http://www.kevinstratvert.com
- LinkedIn: / kevinstratvert
- Discord: https://bit.ly/KevinStratvertDiscord
- Twitter: / kevstrat
- Facebook: / kevin-stratvert-101912218227818
- TikTok: / kevinstratvert
- Instagram: / kevinstratvert
🎒 MY COURSES
- Go from Excel novice to data analysis ninja in just 2 hours: https://kevinstratvert.thinkific.com/
🙏 REQUEST VIDEOS
https://forms.gle/BDrTNUoxheEoMLGt5
🔔 SUBSCRIBE ON YOUTUBE
https://www.youtube.com/user/kevlers?…
🙌 SUPPORT THE CHANNEL
- Hit the THANKS button in any video!
- Amazon affiliate link: https://amzn.to/3kCP2yz (Purchasing through this link gives me a small commission to support videos on this channel — the price to you is the same)
#stratvert #whisperai #openai
Content
0 -> Hi everyone, Kevin here. Today, we're going to
look at how you can both install and also use
5.16 -> OpenAI's Whisper AI. With Whisper,
you can transcribe speech to text.
12.12 -> It has extremely high quality. In fact, you
can click on the caption icon down below on
17.22 -> this video to see captions generated by Whisper.
It works with over 96 different languages and my
24.54 -> favorite part. It's completely free to use. If you
would prefer not to install anything on your PC,
30.48 -> check out the video right up above, and that
shows you how you can use Whisper AI entirely
34.8 -> in the cloud. In this video, we're going to
install it on your PC. Let's check this out.
40.08 -> To get Whisper AI working on your computer,
we need to install five different items,
45.36 -> and I know that sounds like a lot, but we'll walk
through step by step how you install all of them.
51.36 -> Also at the very end, if you no longer have a
need for a Whisper AI for transcribing audio,
56.58 -> I’ll also walk you through how you can uninstall
all of this. First, we need to download something
62.46 -> called Python. You can click on the card up
above or the link down below in the description.
67.8 -> Python is the programming language that
Whisper AI uses. On the Python homepage,
73.14 -> click on the text that says download and on the
download page, you have a few different versions.
78.48 -> Whisper AI works from version 3.7 all the way
up to 3.10. It currently does not work on 3.11.
87.06 -> If I scroll down a little bit, here we see
all of the different release versions. I'll
90.66 -> click on 3.10.10, and on this page, if we scroll
all the way to the bottom, here you can choose
96.42 -> your operating system. I'm running a Windows
machine, so I'll select the Windows installer
100.68 -> 64 bit. Once you finish downloading Python, in
your downloads folder, click on the exe file.
106.98 -> This kicks off the installation process, and
there's one thing you have to make sure to check.
111.48 -> Down at the very bottom, you'll see this checkbox
that says add python.exe to path. Check this box.
117.84 -> This allows us to run Python directly from
the command prompt, and we're going to do that
123.3 -> later. So make sure to check that. Next, click on
install now and run through the install. And just
128.88 -> like that, it looks like the setup was successful.
To confirm the installation, go down to the search
134.52 -> icon down below on the taskbar and type in CMD for
command prompt. This opens up the command prompt
140.58 -> and you can type in python -V, V for version, and
when you hit enter, here it tells me that I have
147.42 -> Python 3.10.10 installed, and that's exactly what
I expected. Next, we need to install something
153.78 -> called PyTorch. You can click on the card up above
or the link down below in the description. PyTorch
160.38 -> is a machine learning library. Here on the
homepage, if we scroll down just a little bit,
165 -> we see a section that says start locally.
Basically, we want to run this on our computer.
170.58 -> Right down below, we have to configure a few
different settings. Right here we want to install
174.66 -> the current stable version, so I'll make sure
this is selected. Right here, you can choose your
180 -> operating system. It works on Linux, Mac, and also
Windows. I'll select Windows. Right down here,
185.1 -> we have to choose the package type and I'll
select PIP since we just installed Python.
190.08 -> For the language, we'll use Python. And right
down here, we can choose the compute platform.
194.76 -> If you have a high-end graphics card in your
computer, like let's say an Nvidia graphics card,
200.04 -> I would recommend choosing CUDA 11.8. That's the
most recent version. Over on the right-hand side,
205.5 -> if you don't have a high-powered GPU in
your computer, then you could select CPU,
209.82 -> but this doesn't go as quickly as a dedicated
graphics card. So ideally you could select this
214.74 -> option. Once you make all these selections here,
let's copy this command down below. I’ll press
220.38 -> control C. Back in command prompt, you can press
control V or your right mouse button, and that
226.62 -> will paste the command that we just copied. To
install PyTorch, simply press enter now. And it
232.62 -> looks like it has now successfully completed
installing. We're on number three now. See,
237.72 -> we're making some really good progress. Here
we need to download a package manager called
243.24 -> Chocolatey, and this will work on a Windows. If
you're running Mac, I recommend downloading and
248.34 -> installing something called Homebrew. In the top
right-hand corner, let's click on the text that
253.14 -> says install. This drops us onto the install
page, and right down here, we need to choose
258.66 -> how to install Chocolatey, Here, I'll select
individual. If we go down a little bit more,
263.94 -> you'll see a text box. Let's click into this
and then select copy. On your Windows desktop,
269.76 -> go down to the search icon and type in
PowerShell. Here, we see PowerShell as
275.94 -> the best match. Right click on that and then
select run as administrator. This now opens up
282.54 -> PowerShell. You can press control V or your right
mouse button, and that pastes in the command that
287.82 -> we just copied from Chocolatey. Here, press enter,
and this will go through and install Chocolatey.
293.76 -> Now that we've finished installing Chocolatey,
we're going to use the Chocolatey package manager
298.56 -> to install something called FFMPEG, and we're
going to use FFMPEG to read the different audio
306.18 -> files, so whether it's a WAV file or whether
it's an MP3. Down below within PowerShell,
311.52 -> type in choco, this is using Chocolatey, then
type in install, and we want to install FFMPEG,
319.14 -> then hit enter. This will now install the package.
Here I'll click on yes. And here it looks like
325.8 -> FFMPEG was successfully installed. Here I am
now in command prompt in administrator mode,
332.16 -> and this brings us to the fifth and final
item to install. And that's Whisper AI.
339.18 -> To install it, type in pip install, and here I'll
type in a -U. That way, if for whatever reason
346.86 -> you already have Whisper on your computer,
that will upgrade it to the latest version.
351.06 -> Next type in OpenAI-Whisper, and then hit enter.
This will now go through and install Whisper AI.
359.04 -> Congratulations. We have now finished installing
all of the prerequisites to run Whisper AI.
365.82 -> Next, navigate to the folder that has all of
your different audio files. This will work with
371.16 -> WAV files, MP3, MP4, all types of audio and also
video files. Within File Explorer, click into the
378.72 -> address field right up here and then type in CMD
and press enter. This opens up command prompt,
384.12 -> and we're currently in the same directory that
all of our audio files are in. So that's perfect,
389.28 -> and we're now ready to finally run Whisper.
To run Whisper, simply type in whisper space,
395.52 -> and then type in the file name. Here I'll type
in sampleaudio1.wav. If your file say has spaces
402.54 -> in it over, here you can put quotes around it.
So here I could also type in sampleaudio1.wav,
407.4 -> close my quotes, and that will also work. And
that's all you need to do. Let's now hit enter,
413.34 -> and this will start running Whisper AI. By
default, this will use the small model and later
420.3 -> on, I'll show you how you can use other models.
One of the really neat things is here you see that
424.74 -> it automatically detects the language used in the
file and here it's successfully identified that
429.84 -> I used English and right down below, I can see
all the different text that makes up this file,
435.42 -> so it looks like it has successfully transcribed
the file. Let's now minimize command prompt,
440.58 -> and this brings me back into File Explorer and
you'll probably notice that we have several new
446.1 -> files here and they're all different file formats,
but they all include the transcript. Here for
451.5 -> instance, I can click into the JSON file and here
I see a JSON file, and here we have all of the
456.9 -> transcribed text. Especially if you want to pull
your text in paragraph format, this is a really
461.58 -> good way to do that here. I can click into the SRT
file, and this is a caption file that includes a
467.46 -> transcript of everything that was said, along
with time stamps up above. Here you have some
472.44 -> additional caption formats and here you also have
a TXT file, and here we just see the pure text
477.48 -> without any timestamps at all. Let's now go back
into command prompt to see how we can transcribe
483.42 -> multiple files at once. Just like we did before,
let's type in Whisper and I'll type in one of my
489.18 -> file names, SampleAudio1.wav here by simply insert
the space. And then I can type in another file
495.72 -> name, here SampleAudio2.wav. And now I could press
enter and it'll go through and transcribe both
502.56 -> audio files. This works especially well if let's
say you have a number of different files that you
507.18 -> need to transcribe. And just like that, it has
now finished transcribing both of my files. Here,
512.88 -> if I minimize command prompt, here I can see that
I have all of these files for each one of my audio
517.8 -> files. That's pretty quick and easy. By default,
Whisper AI uses the small model, but you have five
524.64 -> different models that you can choose from. In
general, the larger the model, the better the
529.5 -> quality that you'll get, but you do need to have
a GPU that's capable of running that. Also, you'll
536.7 -> find that the larger the model, it also tends to
take a longer time to process. And at least from
541.92 -> what I found is there are diminishing returns as
you go larger. Next, let's look at how you can use
547.74 -> one of these different models when you run your
transcript. Back in command prompt, to use another
552.9 -> model, simply type in whisper, your file name dot
wave. And next let's type in a dash dash and type
559.32 -> in model. And here you can specify the model that
you would like to use. I'll type in medium and now
565.32 -> simply press enter and it'll use that other model.
If you haven't used that model before, first it
571.62 -> will need to download it. And it's now finished
transcribing all of my audio. And the main thing
577.2 -> that stands out to me is it looks like it included
some additional punctuation, here attention comma,
582.36 -> over here it included a comma, and I didn't
get that when I just used the small model. So,
587.22 -> you do get slightly better quality, but again, it
will take a little bit longer. Back within File
592.92 -> Explorer, here I have a file titled German.wav,
and this is audio in German, and I would like to
599.58 -> transcribe this. Right up above, let's again
launch command prompt. Within command prompt,
605.16 -> just like we've been doing all along, type in
whisper and then the file name, Here I'll type in
610.44 -> German.wav. Now I could just press enter and it'll
auto detect the language, but I can also specify
616.68 -> the language. Here I'll type in language dash dash
language space, and then here I could specify that
622.38 -> it's German. So, it doesn't have to auto detect
and then I could press enter. And just like that,
627.78 -> here I have a transcript in German of all of the
audio in this file. Along with transcribing audio
634.02 -> in different languages, you can also translate
the audio into English. Unfortunately, you cannot
640.5 -> currently translate into any other language.
Here I simply entered in the same command that
645.72 -> I entered in previously. Here I'll enter a dash
dash and then task and currently by default,
652.32 -> it's set to transcribe, but here I could also
set it to translate and then I'll press enter
657.3 -> and there I can see a translation of all of this
German text. Now it's not perfect and I'll have to
663.36 -> go back and make some tweaks, but overall, I'd say
it's pretty solid. As we've been walking through
668.28 -> this, I've been using lots of different arguments
like dash dash language or dash dash task, and if
675.06 -> you'd like to see a list of all of the different
arguments that you could use with Whisper,
678.66 -> simply type in whisper dash dash, and then help.
This one I'll spit out a list of all the different
685.5 -> arguments that you have available to you. And
it also includes a description of what all of
690.3 -> them do. So, for instance, you can choose where
you want to save all of your additional files,
694.98 -> and there are lots of other settings here. So,
feel free to look through here to see what all of
699.54 -> the different options are. On the following page,
and you'll find a link in the description, you can
704.76 -> see all of the different languages that Whisper
AI supports. In general, the lower the number,
709.92 -> the higher the quality that you'll see. Typically,
when I finish transcribing, I’ll listen to the
716.22 -> audio and look at the text just to make sure that
it's accurate. Overall, it works incredibly well,
721.08 -> but I do find that I have to go back and make
a few small tweaks here and there. If you
725.76 -> decide that you no longer want Whisper AI on your
computer, you can uninstall it by walking through
730.68 -> all of these different steps. You'll also find
this in the description of this video. All right,
736.14 -> well, let me know down below in the comments,
were you successful at transcribing audio? To
741.42 -> watch more videos like this one, please consider
subscribing and I'll see you in the next video.
Source: https://www.youtube.com/watch?v=ABFqbY_rmEk