Looking for speech-to-text software.

[cobaka]

Hello all!

Anyone here know about/used speech to text software for Puppy?
This is microphone to UTF-8, not the other way around.
Thanks!

cobaka

[mikewalsh]

@cobaka :-

Hi, Les.

You're going to be unlucky on this one, I fear.

TTS - text-to-speech - is not easy to find for Linux. STT - speech-to-text - is all but non-existent.

What STT stuff there is seems to be all created for developers to build training models for. Take a look here:-

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

.....and it'll give you some idea of the state of play ATM. It's mostly development models & libraries, and you're definitely not going to find a nice GUI to gently guide you through it..!

Mike.

[ravensrest]

I don't know of any stand-alone STT programs for Linux, but the Chrome browser can be equipped with "Speech to text" and "Dictation for Gmail" extensions. I use both with Fossapup and Mike Walsh's portable Chrome, and both work quite well.
BS

[wiak]

Long time ago (ten years or more) I (as old forum member mcewanw) published a deb and dotpet of a special version of flite that used Hidden Markov Models to provide pretty good TTS. Was a bit of work compiling and assembling all the bits and pieces since from a university research project with little in the way of instructions, but was really worth the effort - tiny compared to using Festival and considerably superior to the robot-like voices of espeak. I've lost the dotpet and the deb but the flite+hts_engine sourceforge site is still out there. However I have no time at present to build this again, but I can assure you - all things considered, it was pretty good:

https://oldforum.puppylinux.com/viewtop ... 30#p365897

https://hts-engine.sourceforge.net/

But that is Text to Speech not Speech to Text. I remember there was something called Sphinx for Speech to Text and a Pocket Sphinx small version of that... https://github.com/cmusphinx/pocketsphinx

[stemsee]

I once contacted the developer of 'Simon' speech recognition software and he gave me permission to compile and develope simon on puppy! Which i never did....but I was building a puppy which had everything needed for blind users. At that time access to the source code was only through him directly. Now it's available on https://simon.kde.org/
stemsee

[mikeslr]

I have several versions of Dragon Naturally Speaking which run under Windows (thru 7). I could run them under Wine in Ubuntu. But never under Puppys. Naturally, I search unsuccessfully for alternatives. wiak's post reminded me of the applications I looked into. But as MikeWalsh noted, 'it's all but non-existent'.

Dragon is/was one of the reasons I don't wipe Windows from the computers I acquire, having the thought that being loquacious and able to speak faster than type, a speech to text app might save some time. Well, that's the theory. The reality is that despite the hype in practice I found I spent more time making corrections of documents produced by Dragon than I would have if I just typed it.

There is another route of speech-to-text you might consider. Again something I explored but not very much. [The idea was to be able to take notes about things which occurred to me while driving. I decided that not driving into things was more important]. If you have an android device there are several applications which work reasonably well. https://www.makeuseof.com/tag/best-andr ... tion-apps/. Your dictated text files can then be transferred to Puppys. They might be sufficient to provide a 'frame-work' you can 'flesh-out' in using a word-processor.

[dogcat]

https://github.com/petewarden/spchcat
goes to the readme, specifically for debian/ubuntu and needs pulse audio.
has a deb available for download but is a whole gigabyte

[mikewalsh]

https://github.com/petewarden/spchcat
goes to the readme, specifically for debian/ubuntu and needs pulse audio.
has a deb available for download but is a whole gigabyte

A gigabyte? *Ouch*

I'm not surprised, though. As I found out when sourcing "voices" for TextAloud! a while back, even the older AT&T 'NaturalVoices' run out anywhere between 600-800 MB each...

(*shrug*)

This isn't something you can really tackle when operating under limited resources, using older hardware.

Mike.

[cobaka]

@dogcat

https://github.com/petewarden/spchcat
Go to the readme, specifically for debian/ubuntu and needs pulse audio.
Has a deb available for download but is a whole gigabyte

Some made this STT work. Here is an interesting comment:

While installing on an RPi3, it ran out of disk space when copying etc/spchcat/models to /etc. The total size of this dir. is indeed larger (2.3GB).

Well .. 2.3GiB doesn't stress my HDD (even the smaller capacity drives I use) - but wow! 2.3GiB is a LOT of code. Makes my 1.3kG brain seem efficient. Recently I read an interview with Janet Clapp. Here's Clapp on memory:

Abbate (interviewing Clapp): Did you have more memory to work with at that point?

Clapp:
Oh yes. And we moved to one thousand twenty-four, [bytes] then two thousand forty-eight. [bytes] Giant leaps! Oh, it was interesting for another reason, because you (we) spent so much time trying to cram things in, being clever rather than being straightforward, [so] that we used to have a contest. We’d post something, [a piece of code,] and say — “Can you make this shorter?” or “Can you figure out what this does?” [laughter.]

To put this in context: Clapp is writing about her experience working on a military contract for national (US) defence.
This is (almost) within living memory. I know some-one who serviced a computer running with vacuum tubes.
When a maintenance session was over Ed had a bucket of 9-pin twin triodes to take home. 12AT7, I would guess.

So we moved from 1k bytes to 2.3GiB in a life-time. Amazing.

cobaka

[Clarity]

Hello @cobaka
For speech-to-text (STT) I have done this for many years: I use my Google Drive as follows:

1. Open a blank document

2. On its Menu, goto Tools>Voice Typing

3. Click the Microphone that appears to start a recording

4. When finished, click the Microphone to stop/restart at any time

NO software to install/maintain. It is 'reasonably' competent. It is Free. This feature is ALSO in recent MS Office (cloud version too).

I have used it for recording seminars-speakers. As well I use it from time to time having it dictate while I walking-talking around the room. Used, as well, in a classroom setting to record the info presented.

Hope this is helpful.