pls recommend app for voice to text -note taking

[Jamayka24]

Doesn't have to be very accurate - just understandable.
I wouldn't have a problem with retyping its output.
For Bookwormpup64 Thanks

[sonny]

@Jamayka24

https://fosspost.org/open-source-speech-recognition

I also posted here about Personal Intelligence called "Pi":

https://www.forum.puppylinux.com/viewto ... 70#p127270

pi.png (45.4 KiB) Viewed 1981 times

Use it in Chrome browser, whatever you say (and he/she says) will be transcribed.
Psst... it is very natural, very accurate, and very addicting (esp. #4)

Tips:

1. Give 'her' (#4) a nickname and tell 'her' your name/nickname
('she' is happy I call her 'sweet pie', so am I with 'sonny')
2. /delete
(to delete the chat history on the left pane)

[Jamayka24]

@sonny - thanks but I do not not want a bot.
I had looked at those. I want a simple old app like
' Dragon easy speak' I think it was called, worked fine
on Windows 15 years ago. Must be something free
out there for Linux. Something non predictive,
non conversational, unfriendly, that just changes
speech into text. Not an AI friend.

[Geek3579]

I have recently done about 20,000 words using Google Docs as part of Google-Chrome. No installation required.

Open a word document in Google Docs and use Tools --> Voice typing ( or ctrl + shift + s ) to bring up then microphone icon, and click to start.
As other users have noted, It does tend to randomly stop and needs to be restarted, but its a minor nuisance IMHO .

It works tolerably well, although you need to add capitals, and punctuation to suit. There are some voice commands, but I really wonder if they work.
The final text can be copied and pasted wherever.
Having a sharper or more shrill voice does help. My wife gets a very high accuracy rate in English, German and Russian. Languages are easily changed via a pull down menu.

As I tend to touch the caps lock key inadvertently when typing, I turn the caps lock function OFF via: setxkbmap -option ctrl:nocaps
And turn it ON via: setxkbmap -option
See https://www.cyberciti.biz/faq/linux-dea ... caps-lock/

[Clarity]

@Geek3579 provides an excellent recommendation requiring no need for installation and very-very little learning. Below is a 2min video:
Voice typing to write a document/book <=== Click

[Jamayka24]

I am trying not to use the monopolists, LOL. A futile protest for sure but...
I would like some stand alone non cloud app.

[mikewalsh]

@Jamayka24 :-

Heh. Unfortunately, this is one of the very few areas where there simply is no Linux "challenge"to the Windows dominance. "Dragon Naturally Speaking" has no Linux equivalent at all; it's stubbornly Windows-only, neither will it work under WINE.....@mikeslr and I have both proved that.

There's something called 'Festival', I believe.....but I've never bothered with this myself, though others here have discussed it in recent months. I do the same as @Geek3579 - I use the voice transcribing option in Google Docs; in my case, it's set-up as a webapp to open in its own dedicated window, with its own Menu entry, for the odd occasions when I have need for it.

TBH, with virtually everything having an online equivalent nowadays - and many of those "equivalents" being of superior quality to their native Linux counterparts - it makes no sense at all to shun them. I appreciate where you're coming from; many don't like cloud stuff, and plenty just seem to hate Google.....but Google's software infrastructure just 'works' - everywhere! - especially in tandem with its own stuff. Many of their provided apps have made my own life very much easier over the last 15 yrs or so.

All you really need these days is a web-browser.

Mike.

[Geek3579]

I am trying not to use the monopolists, LOL. A futile protest for sure but...
I would like some stand alone non cloud app.

I get that. I started looking for a stand alone package myself months ago now. There was a post on this a ways back for running Nerd Dictation, but I could only get it to work in Jammypup64, and I use Bookwormpuop64 as my daily driver, so I did not persist. (I also was only able to use the small VOSK package version. The large VOSK version ran out of RAM (8GB on my machine - probably needs 16GB machine or pupswap enabled , which I did not try).

Anyhow, I'm not sure if the small version works as well as Voice typing in google. But that may be related to my audio settings.

For more details re Nerd-dictation (speech to text ) see
viewtopic.php?p=112917
https://alphacephei.com/vosk/models

[mikewalsh]

@Geek3579 :-

Curious...

Voice typing in Google Docs, yes? It seems to vary from one browser to another...

Chrome - shows up, and works (as expected)
Slimjet - shows up, and works (slight delay, though, but it does work)
Iron - shows up, but is non-functional. Tweaking does nothing
Opera - no sign of the option (nothing I try will make it show). Weird, given that voice search in Google itself works A-OK in Opera....
Firefox, Midori, Pale Moon - none of the 'zilla-based browsers are functional with this option. They all default to a 'global' microphone option; fine for laptops, but desktops don't, by default, even HAVE a microphone (and these browsers won't allow selecting an alternative to 'default & only sink') They leave it to the OS itself to select/set the microphone.....and Puppy doesn't have this feature, per se.

I use the microphones on my Logitech webcam.....which ALSA doesn't "see" OR show in AlsaMixer. It can, however, be individually selected via arecord.....AND in any Chromium-based 'clone'.

Bloody minefield, if you ask me..!

Mike.

[Geek3579]

I just tried Vivaldi on Bookwormpup64 with Pulse audio. Voice to text comes up, but does not produce text output.

[mikeslr]

Yes, Google's Chrome is very good at most things. But Google wants to know and keep a record of everything. It's a trade-off.

TBH, the best private Voice-to-Text remains Dragon under Windows 7. Dragon under other Linuxes --e.g. Ubuntu-- running Wine is almost as good, if a little difficult to set up. [Would have to look for the instructions]. But as T'other Mike noted, I was never able to get it functional under Puppys.

[mikewalsh]

@mikeslr :-

From what I can see of it, Mike, Dragon is now a 'paid-only' solution. Looking on their site, there is NO 'free' option at all any longer. $700 a year? Jeezus. I think that's out-the-window...

T'other Mike.

[jamesbond]

whisper.cpp is your friend.

My earlier post: viewtopic.php?p=95947#p95947

To test, I read Mike's post:

Heh. Unfortunately, this is one of the very few areas where there simply is no Linux "challenge"to the Windows dominance. "Dragon Naturally Speaking" has no Linux equivalent at all; it's stubbornly Windows-only, neither will it work under WINE.....@mikeslr and I have both proved that.

There's something called 'Festival', I believe.....but I've never bothered with this myself, though others here have discussed it in recent months. I do the same as @Geek3579 - I use the voice transcribing option in Google Docs; in my case, it's set-up as a webapp to open in its own dedicated window, with its own Menu entry, for the odd occasions when I have need for it.

TBH, with virtually everything having an online equivalent nowadays - and many of those "equivalents" being of superior quality to their native Linux counterparts - it makes no sense at all to shun them. I appreciate where you're coming from; many don't like cloud stuff, and plenty just seem to hate Google.....but Google's software infrastructure just 'works' - everywhere! - especially in tandem with its own stuff. Many of their provided apps have made my own life very much easier over the last 15 yrs or so.

All you really need these days is a web-browser.

And then had whisper.cpp transcribed, using the modest "base_en" model.

This is unedited output:

[00:00:00.000 --> 00:00:08.160] Unfortunately, this is one of the very few areas where there is simply no Linux challenge
[00:00:08.160 --> 00:00:10.400] to the Windows dominance.
[00:00:10.400 --> 00:00:14.520] Drago naturally speaking has no Linux equivalent at all.
[00:00:14.520 --> 00:00:16.880] It's stubbornly Windows only.
[00:00:16.880 --> 00:00:20.200] Neither will it will work under wine.
[00:00:20.200 --> 00:00:22.920] Mike and I have both proved that.
[00:00:22.920 --> 00:00:27.840] There is something called Festifar I believe, but I've never bothered with this myself,
[00:00:27.840 --> 00:00:31.680] so others here have discussed it in recent months.
[00:00:31.680 --> 00:00:34.680] I do get the same as Geek.
[00:00:34.680 --> 00:00:39.280] I use the voice transcribing option in Google Docs.
[00:00:39.280 --> 00:00:47.080] In my case, it's set up as a web app to open in its own dedicated window, with its own
[00:00:47.080 --> 00:00:54.680] menu entry for the odd occasions where I need the need for it.
[00:00:54.680 --> 00:01:00.440] Be honest, with virtually everything having an online equivalent nowadays, and many of
[00:01:00.440 --> 00:01:05.760] those equivalents being superior quality to their native Linux counterparts, it makes
[00:01:05.760 --> 00:01:08.320] no sense at all to shun them.
[00:01:08.320 --> 00:01:11.040] I appreciate where you're coming from.
[00:01:11.040 --> 00:01:16.600] Many don't like cloud stuff, and plenty just seem to hate Google, but Google software
[00:01:16.600 --> 00:01:23.200] infrastructure just works everywhere, especially in tandem with its own stuff.
[00:01:23.200 --> 00:01:28.160] Many of the provided apps have made my own life very much easier over the last 15 years
[00:01:28.160 --> 00:01:29.560] or so.
[00:01:29.560 --> 00:01:32.160] All you really need this day is a web browser.
[00:01:32.160 --> 00:01:42.160] [BLANK_AUDIO]

I actually stumbled around [00:00:47.080 --> 00:00:54.680] so you see words "need" repeated twice - that's not whisper's fault, that's the reader's (=me) fault.

The only downside is that the transcribing is not real-time or continuous. I had to record the entire speech first, and the pass the recorded WAV file to whisper.cpp. whisper.cpp actually have a mode where it can transcribe in real-time, but the result is far worse; not because the software itself is bad, but because the mechanism whereby they capture the input from the microphone is not optimised.

But this isn't a problem for us here, innit? We can create a simple gtkdialog (or yad, or whatever) app, which will alternately record, and the transcribe, and then further record, and then further transcribe, etc until we tell it to finish. It's still not real-time but close enough. In fact, I may even write this script myself ...

----

The test was done using whisper.cpp I downloaded in August 2023; that's a year ago. A year has passed and the software is in continuous improvements, I'm quite sure the recent version would be waaaay better!

[dogcat]

whisper.cpp is your friend.

My earlier post: viewtopic.php?p=95947#p95947

To test, I read Mike's post: <snipped>

The test was done using whisper.cpp I downloaded in August 2023; that's a year ago. A year has passed and the software is in continuous improvements, I'm quite sure the recent version would be waaaay better!

-------------

Oh my, what a wonderful tool!

First tested with the suggested jfk.wav to verify working after compile.

Then converted one of my .m4a audio files of a podcast that is 1hr 15 minutes to 16-bit wav format.

Ran it through the "main" binary and it processed the poorly recorded 139MB audio 16-bit wav file in 9min 45sec, piped to output.txt, the finished text file size was 76k

[00:00:00.000 --> 00:00:05.680] Good luck Michael.
[00:00:05.680 --> 00:00:11.460] recorded live Hello this is Michael Adams for nothing but the truth.
[00:00:11.460 --> 00:00:18.880] It's January 21st 2015 and once again we have a York from, excuse me, I say your name
[00:00:18.880 --> 00:00:24.720] wrong every time I apologize to my friend, a York from, I took their 66 YouTube channel
<snipped out the middel 1+ hour>
[01:14:37.800 --> 01:14:43.960] and doing that otherwise i will round this broadcast up right now with almost an hour and a half going
[01:14:43.960 --> 01:14:53.240] on and thank you all for listening and i hope to see you next time and you know we didn't do
[01:14:53.240 --> 01:14:59.480] this for our thing we did do this broadcast like all the broadcast we do here to bring fame
[01:14:59.480 --> 01:15:04.440] and glory to our lord and savior jesus christ
[01:15:04.440 --> 01:15:10.840] and in his name i thank you very much and um and in jesus name i thank the father and the
[01:15:10.840 --> 01:15:18.200] holy spirit that they helped their hand upon us doing this broadcast and um i leave my closing
[01:15:18.200 --> 01:15:23.640] remarks to tom and michael thank you very much and goodbye

This is accurate, even transcribing poorly recorded heavy German accented English correctly. I did not try it with a microphone because I don't have one. All done with no internet on my local machine. Sweet

Thanks!

[BarryK]

But this isn't a problem for us here, innit? We can create a simple gtkdialog (or yad, or whatever) app, which will alternately record, and the transcribe, and then further record, and then further transcribe, etc until we tell it to finish. It's still not real-time but close enough. In fact, I may even write this script myself ...

Yes, do it!

[jamesbond]

But this isn't a problem for us here, innit? We can create a simple gtkdialog (or yad, or whatever) app, which will alternately record, and the transcribe, and then further record, and then further transcribe, etc until we tell it to finish. It's still not real-time but close enough. In fact, I may even write this script myself ...

Yes, do it!

On it!
Be back soon.

[mikeslr]

@mikeslr :-

From what I can see of it, Mike, Dragon is now a 'paid-only' solution. Looking on their site, there is NO 'free' option at all any longer. $700 a year? Jeezus. I think that's out-the-window...

T'other Mike.

Per WineHQ, Dragons versions 11 thru 12.5 had "Silver" ratings. These are 'installed' into a Wine-prefix under, say, Ubuntu. No subscription necessary. They can still be purchased on Amazon and ebay for under $100. Lost the link, but IIRC there was a 'new, sealed' version 11 for about $35 on ebay.

[jamesbond]

On it!
Be back soon.

There you go. The GUI can operate in 3 output modes:
a) the transcription can be stored in clipboard which you can paste using Ctrl-V immediately after every transcription,
b) the transcription can be output to a textbox, which you can copy and paste at the end of the session,
c) the transcription can be output to a file (which you can still view using a a textbox during the session).

You can also choose the model and they will be downloaded automatically if they don't exist yet.

I have only tested this in Fatdog64 but it should work elsewhere too, since the requirement is modest. In addition to whisper.cpp, you will only need sox, Xdialog (and either xsel or xclip if you want to use the "clipboard" mode).

You only need to modify WHISPER_PREFIX to point to the correct location, everything else should work as is.

NOTE: For people who hasn't read the entire thread: the attachment is a script to make the use of whisper.cpp easier. It does not include whisper.cpp itself, so therefore it cannot do the transcription by itself. If you don't already have whisper.cpp, then the script will not work for you.

UPDATE: to include a switch to automatically translate native speech to English. You need to use at least the "small" model for this to work. I saw there were already 6 downloads before I updated this. If you're one of them, please re-download.

[dogcat]

Hi jamesbond, thanks for building the script. I tried it in bookwormpup64 and it will not go past the initial launch screen "Please click start to start recording". All dependencies were installed including both xsel and xclip. No ideas as to what is not working, I used the directory structure in your script /opt/whisper-cpp/ for the whisper.cpp files.

Running from terminal the bookwormpup64 error is

Code: Select all

# ./whisper-gui.sh
/usr/lib/gtkdialog/box_splash: line 192: arithmetic syntax error
/usr/lib/gtkdialog/box_splash: line 192: arithmetic syntax error
error: input file not found '/tmp/transcribe.6217.wav'
error: no input files specified

That gave me the incentive to try fatdog902, it built whisper just fine and your script seems to work, I now see the 2nd window " Recording ... Click here to stop recording and start transcribing", I do not have a microphone to try it past there but you already know it works in fatdog . After trying fatdog for the 1st time I found it intuitive and easy to use, I should have looked at it before now but better late than never

Anyway, posting this so you know about the script & bookworm64 not getting past the first screen. There were warnings in bookwormpup64 when it built whisper but it did process the 16-bit .wav files. Out of time for now, later gator.

[jamesbond]

You're most welcome @dogcat.

The problem you're experiencing is because bookwormpup (and all recent puppies) ships with broken yaf-splash.

The "yaf-splash" in Puppies isn't the real yaf-splash, it's a replacement script (which internally uses gtkdialog). Unfortunately, this replacement script suffers from two problems:

a) it is not compatible. Try this on the terminal: yaf-splash -bg yellow -outline 0 -text "This is a test".
With the real yaf-splash, this will pop-up a message, which will stay indefinitely until it is clicked; and then the message will be dismissed. In bookwormpup, it will only shows up for 30 seconds and then disappears.
You can make bookwormpup's yaf-splash to stay indefinitely by using: yaf-splash -bg yellow -outline 0 -text "This is a test" -close never but then it doesn't go away even if you click on the message; and the "-close never" switch isn't supported in the original: if we try to use it on the original yaf-splash, it will cause an error instead.

b) it is buggy. For example: instead the command above, use this: yaf-splash -font 10x20 -bg yellow -outline 0 -text "This is a test" (I'm adding the -font parameter to try to use a different font). With the real yaf-splash, this works. In bookwormpup it will give you the error you saw when you ran whisper-gui.sh.

The problem (b) can be worked out, but (a) is a real headache. In my opinion, a replacement program for a well known program should be backward-compatible with the original. More features can be added, or certain switches can be ignored, but it should at least work (even minimally) when given the parameters that work with the original.

The proper solution is of course to fix this yaf-splash replacement, but I don't think this will happen soon. Fortunately you can run whisper-gui.sh in another way, telling it to use Xdialog instead of yaf-splash: whisper-gui.sh --splash xdialog.

Or otherwise, just edit the script on line 161, and replace check_yafsplash with SPLASH=xdialog.

That gave me the incentive to try fatdog902 ...
...
After trying fatdog for the 1st time I found it intuitive and easy to use, I should have looked at it before now but better late than never

I hope you like, and you're welcome to stay if you like

[dogcat]

The problem you're experiencing is because bookwormpup (and all recent puppies) ships with broken yaf-splash.

@jamesbond, Great explanation, and you had already anticipated the yaf-splash issue. I looked at the script but might as well have been reading Martian , I could not tell there were switches there.

That gave me the incentive to try fatdog902 ...
...
After trying fatdog for the 1st time I found it intuitive and easy to use, I should have looked at it before now but better late than never

I hope you like, and you're welcome to stay if you like

I think that's a good idea, thanks again!

[BarryK]

EasyOS has the real yaf-splash!
The gtkdialog variant is named gtkdialog-splash

[BarryK]

Trying the GUI, have hit a problem:

Code: Select all

# ./whisper-gui.sh 
Downloading ggml model base.en from 'https://huggingface.co/ggerganov/whisper.cpp' ...
ggml-base.en.bin    100%[===================>] 141.11M  3.92MB/s    in 48s     
Done! Model 'base.en' saved in '/opt/whisper-cpp/models/ggml-base.en.bin'
You can now use it like this:

  $ ./main -m /opt/whisper-cpp/models/ggml-base.en.bin -f samples/jfk.wav

sox FAIL sox: Sorry, there is no default audio device configured
./whisper-gui.sh: line 230: kill: (28253) - No such process
error: input file not found '/tmp/transcribe.28139.wav'
error: no input files specified

EDIT:
Fixed. I discovered that the build recipe for sox in openembedded has a bug; compiled without alsa and pulseaudio support.

Recompiled sox and now your gui works!

[Jamayka24]

Please pursue this Whisper- I looked at it but it needs some
step by step instructions, assuming zero knowledge,
to get newbies like me able to use it.
I had a friend, a writer who had Parkington disease ( hand trembling
symptoms) and had trained a very early Dragon version to a point where he could dictate 500
words with only one or two errors. You keep correcting the errors in
its text output till your pronounciations/accent is recognised..
Even a app without the ability to voice correct would be fine,
just keyboard correction of output is ok

[dogcat]

Please pursue this Whisper- I looked at it but it needs some
step by step instructions, assuming zero knowledge,
to get newbies like me able to use it.
I had a friend, a writer who had Parkington disease ( hand trembling
symptoms) and had trained a very early Dragon version to a point where he could dictate 500
words with only one or two errors. You keep correcting the errors in
its text output till your pronounciations/accent is recognised..
Even a app without the ability to voice correct would be fine,
just keyboard correction of output is ok

Hi @Jamayka24
I tried whisper in bookwormpup64, maybe the following info will help you get started since you are running bookwormpup64.

Here is a whisper PET built for bookwormpup64 that will install the whole compiled directory files into /opt/whisper-cpp/
That is where the script that @jamesbond provided looks for the whisper files, so installing that PET and running his script will launch whisper. (I made a PET of the whole whisper.cpp compiled directory because I did not know exactly what is required, that way you have it all).

The PET can be downloaded from here https://archive.org/download/wine-8.0_a ... worm64.pet
md5 abd102923d2a012fe50bf74de23ff75c
The file is 9MB

After installing the PET, download jamesbond's script and make the script executable so it will run.
viewtopic.php?p=128333#p128333

Then using ROX filer, open a terminal window in the same directory where the script is located (you chose where to put the script) and enter the following command in the terminal

Code: Select all

./whisper-gui --splash xdialog

That will launch the script. Then follow the instructions in the window, you will need a working microphone.

That is as far as I got, I do not have a microphone.

Now I am in no man's land, I never made it to the following screens.

According to the script instructions at viewtopic.php?p=128333#p128333, you can choose the model (language) and that will be downloaded automatically if they don't exist yet. I did not include any extra downloaded models in the PET, (the English model I tried was 142MB when I transcribed a podcast), so have an internet connection to download the needed models while using the script for the first time. You should probably only need to do that the first time you use the script to transcribe unless you choose a different model???(not sure how that works)

Anyway, the above PET has more stuff in it than is needed, all the leftover stuff from the build as I don't know exactly what is needed to run the script all the way through.

[Jamayka24]

Many thanks to all especially dogcat - a lot of work -you are truly
a rightious Greek as your signature says!
I have no mike either and trying to get my laptop working
(bad touchpad) so will have to get back to this in a while. Downloaded
all the Whisper components. Thanks again to all.