Text to speech voice is horrible

[Eric23]

Hi, I was wondering if there are any better, more natural sounding voices for text to speech? The one that the system uses to read the screen is horrible to listen to. I can't seem to see any options to change the voice, and have no idea how to go about finding a better one. Assuming there is one.

Thanks

[rossdv8]

There are some simple things to do to play with the existing voice assuming you are using espeak
sudo apt install espeak

I use this quite a lot in software I write simply because it is easy and I am lazy these days. Everything is done with a few simple speed and pitch changes and a few commas or capitals here and there for emphasis.

for example compare:

espeak "Get Coffee, Before Lunch Today"

espeak "Get Coffee Before Lunch Today"

espeak "Get Coffee, Before Lunch Today" -s 140

espeak -ven-sc -a 25 "Get Coffee, Before Lunch Today" -s 140

espeak -ven+f3 "Get Coffee Before Lunch Today" -s 140 -g 5


There are other arguably better ways, but this is simple and easy to adjust for something a little 'less Stephen', although you can simulate Stephen Hawking pretty closely.

Many of the nice voices you hear on GPS car navigation apps are specially read by paid voice over people hired by software companies. You can find some downloadable voice files for various programs.

On the other hand, for simple text to speech there's a good resource here that goes over several pages here. The bad news is that it covers a lot, the good news is that once you have a voice you can tolerate there's not a lot of need to change it much:

http://espeak.sourceforge.net/index.html
:

[Eric23]

Thanks for the reply.
I installed Espeak and it sounds exactly the same as the default voice that the system uses to read the screen. I still cant find any options to change anything. Its just too robotic and harsh sounding.
I understand that GPS voices are pre-recorded, but I have heard better synthesized voices than this. Even Microsoft Sam was nicer than this one. Is there really nothing better on offer?

Many Thanks

[djph]

Is there really nothing better on offer?

It's a bit of the situation of "you get what you pay for"

[rossdv8]

There are other arguably better ways, but . . .

One of the 'arguably better ways' used to be Mycroft mimic. I liked it and it is capable of sounding pretty acceptable in Linux, sort oof like listeningto my GPS - but good luck trying to get it to install the correct dependencies these days.
There is an experimental Snap package that might eventually overcome the dependency issue, but it is a work in progress, and Mint prefers not to do the Snap thing.

If they ever do an AppImage I will be trying it.

[rossdv8]

I suppose one could just explore the available TTS options in the repository.
None are really going to give you perfect grammar and enunciation, but some are adequate if you don't want to listen to Stephen Hawking proof reading your thesis.

Here's where I found the voice below:
https://darkshadow.io/2020/08/01/speech ... linux.html

Code: Select all

sudo apt-get install festival festvox-us-slt-hts

will give you a reasonable voice with perhaps easier voice changing options.
It is a little gentler on the ears than the Stephen Hawkingish default thing in espeak:

Code: Select all

festival -b '(voice_cmu_us_slt_arctic_hts)' \
    '(SayText "Linux Mint is the bestest ever operating system in the whole wide world.")'

It is easy enough to read the manual (if you have a spare week or so) https://www.cstr.ed.ac.uk/projects/festival/manual/ then if you can follow the simple steps - (I can't find a Smiley for ROFLMAO) - and once you get a voice you like, create a bash script, or add a keyboard shortcut to your /home/.inputrc and have a quick way to input text.
You could probably even use zenity to make a quick and dirty GUI.

Here's where I found some very basic into stuff about festival:
https://www.maketecheasier.com/make-com ... -festival/

There are 'lots' pf TTS voices and programs, that are not horrible.
Some are just not horribler than others . . .

I tried installing all the voices I could find in the Mint repos and they are awful.
I'm looking for some like our lady here, but with better cadence, because festival can be set to read a book or file.

Code: Select all

festival -b '(voice_cmu_us_slt_arctic_hts)'  '(SayText "wobbegongs are great eating")'festival -b '(voice_cmu_us_slt_arctic_hts)'  '(SayText "wobbegongs are great eating")'

***

E D I T

I corrected the Apt command to install festival (I had left out the voice and some other bits) and found the link to where I found it.
I'm still exploring other possible solutions to this question.

Here's another link for Festival, which seems to be the most promising Open Source and FREE program I have found so far, but setting up voices other than the one in the example I used seems a nightmare:
https://ubuntuforums.org/showthread.php?t=677277

[rossdv8]

Exploring this issue further. I've tried a number of options. There are lots of FREE web based readers, but most of the stand alone Free options do sound pretty bad.
So far, Festival has the easiest on the ear voices, albeit tricky to configure. The female voice in the festival links in the earlier post at least is less grating than some of the espeak voices.

This link makes some good reading to follow up as it is based on options tested for/on Ubuntu so it is relevant to Mint and debian. You might have read it, but if not the links in it open some exploration.

https://help.ubuntu.com/community/TextToSpeech

[all41]

Over 20 years ago Microsoft asapi14 speech sysnthesis was better
than the defaults we have today. Microsoft Sam. Anyone remember?

[rossdv8]

I used TTS years ago that sounded less artificial, in Linux or BSD or OS/2 - I can't remember. But this question got me curious as to what is out there.

Here are another couple of Ubuntu links for anyone interested in playing with TTS in Mint.
This link has links to voices and other info, plus how to install voices and a bit on using them.
https://askubuntu.com/questions/1324381 ... untu-20-04

This one has packs of voices in various languages foir festival 2.5 (A lot of the links I found only worked on earlier versions):
http://www.festvox.org/packed/festival/2.5/voices/
However - READ the comments. A lot of the high quality voices (hts) that used to be available are not current any more.
If you find voices with cmu and arctic_hts in the name, they are worth a try. Youo might need to check for us and gb or whatever for country.

sudo apt remove --purge festival 'should' get you back to the beginning if you mess up festival.

[Green man]

I've got a decent sounding female voice in orca. But I've forgotten the nitty-gritty of how I set it up, I'll post back once I re-familiarize myself.

That said, whenever possible on Linux, I usually find it easier to cheat and use Firefox as a screen reader. You can either use the TTS built directly into reader mode, or install this extension.
https://addons.mozilla.org/en-US/firefo ... ead-aloud/
On Linux, the extension defaults to your system voice, but it's easy to change in the extension settings.
You can even open PDFs from your hard drive in Firefox and read them with the extension. To my knowledge, orca still doesn't support PDFs

[Green man]

Okay, the first thing is to grab all the mbrola packages from the software manager. Then, enter the command orca --s into the run dialog box. Then in the voice tab select espeak-mbrola-generic. Then in the person dialogue box select us1 and click apply.
See image:

I personally find that voice pretty tolerable. But I also recognize that that's somewhat subjective.
Hope this is helpful.

[rossdv8]

That was fun.I had no idea orca was installed by default.
Now every key I hit has itself spoken. Weird!

[Green man]

Yeah, if you're on Cinnamon, there should be a nice toggle in the accessibility settings to turn that on or off. On other desktops, you can use [insert+s] to toggle speech on and off.

Correction: [insert+s] is only relevant if it's already running. If your system doesn't have an Orca icon by default, or a nice toggle, you can bind the orca command to a custom keyboard shortcut to start the screen reader.

[rossdv8]

Every now and then there's a comment or question here on the forum that opens a thread which leads to some interesting discoveries.
"Text to speech voice is horrible" is one of those things. I mean, I use espeak all the time for popup reminders that I write that throw a big red on white sign on the screen on login when I have, say, a hospital appointment due.
Because I have a number of Virtual Desltops I might miss the popup, so I also add a command for espeak to tell me time and stuff.

Until this Topic I hadn't bothered looking for a change from Stephen Hawking. However, the script for the hts voice from Nagoya IT sounds just a bit more pleasant, and has got me rewriting my shell script to utilise it.

I'd completely forgotten about festival, didn't know about orca and discovered a few other little alternatives to experiment with while trying to find a reason the OP shouldn't have to live with a harsh voice.

[Green man]

Do any of the better voice options you've found integrate with orca? I remember when I first dug into this, finding some better voices, maybe from the festival family, maybe from elsewhere, but then only being able to invoke them through a terminal emulator. The voice I'm using now is the best one I could find that would integrate,but I'd love an improvement too.

[Eric23]

Wow thanks for all the replies everyone. After the first 2 replies I thought the thread would die a death. I just happened to come back and take another look and noticed it kind of took off.

As a novice user it is a lot to take in, but I will have to take a deeper look at all the useful info you all have been posting.

I thank you all for the experimentation that some how occurred. I'm sure I will have fun/frustration trying out your suggestions.

[AZgl1800]

I tired espeak, and found it useless.

I much prefer my cellphone, it knows how to talk in Plain English without distortion.

which reminds me, it is time to take the Trash out to the street. BEEP!

[rossdv8]

I'm sure I will have fun/frustration trying out your suggestions.

As an old bloke who likes 'playing with Linux' I can assure you there will be plenty of frustration.
It is important I guess, to remember that what you get in Linux is mostly the result of people putting their personal, and unpaid time and effort into trying stuff - unlike Windows, Mac and Google (Android) where there are huge budgets.

It is a shame that the Nagoya Institute of Technology text to speech project has moved from simply Text to Speech to voices for Artificial Intelligence.
But it must be remembered that this is an educational institution and I think the hts project was a group of people, mostly students, working on something that was only a small part of bigger things.

The 'fairly pleasant' female voice that I posted in one of my links was just one of many of the voices that were once available on Git-Hub, but which seem to have been removed.
The bulk of their work is probably now in places like Android, Mac and other commercial technology.

[Green man]

I recently discovered something, that I thought it was worth adding to the thread. If anyone decides to go the browser-based screen reader route, this version of the read aloud extension provides a smoother experience when combined with the Brave Browser.
https://chrome.google.com/webstore/deta ... aplp?hl=en
it seems to do the authentication permission part automatically, so that you don't have to prove you're not a robot to use the screen reader. I'm sure the same is probably true of regular chrome, but since brave has come into its own as a browser, I don't use chrome on any of my computers.

[linuxdummie]

Year 2021 and TTS in Linux is still in the 90's.
I gave it up with non-robotic TTS in Linux.

Btw, I really like ReadAloud Browser-Extension (Vivaldi and Firefox) and I am using it everyday on my Windows.
Good thing about this extension is that you can select texts and then have them read aloud. In Windows in a humane voice.
In Linux unfortunately in robotic voice.

With PDF-files: I use PDF-Reader Okular where you can select texts too and read alouad - in Windows without problem. But in Linux Mint it doesn't work even work. In Okular settings I cannot change from "flite" to espeak-ng. There is only flite.

And I've read Gespeaker and Espeak are dead projects. Espeak-ng is a fork of Espeak.
Would like to try MBROLA with MBROLA voices but don't know how to install it properly without brick my Linux Mint installation. There are hundreds of different tutorials about TTS on Linux.