Hospitals no fun, not talking worse.

[bensimmo]

May give you some ideas
http://www.hawking.org.uk/the-computer.html

[DavidS]

May give you some ideas
http://www.hawking.org.uk/the-computer.html

While there is a need for situations like that of Dr Hawkins (still do not understand why he embrases the doctoret title), I am attempting to provide a solution for those of us that still have some level of control of our upper extremities.

I do thank you for that though. It reinforces the need to include a Win32 x86 target.

[DavidS]

Ok I am sticking with FreeBASIC for the front end, and using flite, as it is the best of the speech synthesis back-ends that I have been able to find with a reasonable license that is available for all needed targets.

At present using flite with kal16 or slt played to a wav file and then played at 90% speed and file volume clipped to 85% is very clear and understandable. I will have to upgrade to flite 2.0 so that I can try more voices, I will only include options for the voices that are clear when spoken.

I am sticking to FreeBASIC because I realize that most have a Windows or x86 Linux laptop, NetBook, or Tablet. As Laptop and Netbook give the most input options (keyboard, mouse/touchpad, touch screen) by default without any aditional HW it makes sense to target Windows and x86 Linux primarily.

Thank you all for the various input that is being considered in the design of this UI. I will attempt to make sure that the UI is compatible with Dasher, in a simple to use way.

[DavidS]

I am spending 80% of my waking time working on this project, as I really have nothing else I am able to do at this time (I do not have the dexterity and strength to remove items from a 3D printer).

It is going quite slow primarily because of my current limits, and secondarily because of the fact I am doing everything I can to make sure that everything is working in a reliable and predictable way. I do not want to find any bugs that are enough to cause problems. Though it is moving forward.

With my spelling I am certain that there will be some bugs, in the form of spelling and grammatical mistakes in text that is displayed.

Question:
What is the lowest screen resolution on a Windows based tablet still in common use? I ask to make sure that I get the UI design correct. I already know that the lowest commonly used resolution for a Raspberry Pi display in a desktop setup is 800x480 so I have that one figured.

[DavidS]

Also my decision to stick with Windows, x86 Linux, Linux armhf, and FreeDOS means that I am assured the availability of at least 800 pixels width on 99.95% of systems (approximately). This means that I can afford to have three levels of categories, all on screen at the same time, thus allowing for up to 16 to the power of three categories equals 4096 categories, times 32 items per category equals 131072 items total maximum. I do think that this solves the issue of how many items to allow for, without putting any limits on speed of navigation.

Input is welcome, I am not yet sure of what I will end up with in this area of the design. I have changed the layout a couple of times as I have realized ways to simplify it, and as I am manually drawing all the UI elements it takes a little work to make such changes.

[DavidS]

I am sharing every detail the way I am, for the dual reasons of:
1 : Getting as much practice typing as I can.
2 : Hoping that someone else is also working on a similar project, inspired by this thread.

On number 2:
I am moving very slow, and working within a specific of limits. As such the more projects there may be with this same kind of goal the better. Different people will view the problem from different angles, and different results will be of more aid to different people in situations where something like this is extremely helpful.

going as slow on anything as I am now going on everything is frustrating, it takes some restraint to keep myself from getting very frustrated. Often have to remind myself of the positive points in the progress that has already been made, regardless of how long it is taking.

[bensimmo]

Windows 10 Tablets have a minimum​ spec of 800x600 (8" physical minimum too)
There is technically a minimum resolution on cheap tables as that is the 'free' limit.
At least that what it used to be
Most 8/10" in the cheap part seem to use 1280x800 last time I looked (e.g.linx around £50)

[DavidS]

Windows 10 Tablets have a minimum​ spec of 800x600 (8" physical minimum too)
There is technically a minimum resolution on cheap tables as that is the 'free' limit.
At least that what it used to be
Most 8/10" in the cheap part seem to use 1280x800 last time I looked (e.g.linx around £50)

Thank you for that. And here I am using a Raspberry Pi 3B with a 800x480 5 inch display, I guess I really am behind.

[DavidS]

Still working on my UI. I am continuing to use flite for the moment, as I am figuring out how to make its speech output acceptable for now.

I am a little frustrated at how slow the UI is progressing. I am using some of my pause time to begin building a lexicon of word to phoneme translations, that has eight forms of each word, based on the class of phoneme at the end of the previous word, and class of phoneme at the beginning of the next word.

Secondary Related Project:
I decided (while asleep overnight), that I should begin to study Chip Gracey's vocal tract simulation, and begin writing a speech synthesizer based on these concepts. It may be possible to produce very good legibility of speech this way.

This would also give a high level of configurability to the voice being used, thus making it possible for each user to use a voice that they are happy with. It would make since to have a few hundred preset voices so that those that do not wish to set there own can easily find something close enough, as well as to give an idea of how parameters for setting voices effect the result.

Implementing a vocal tract simulation should not be difficult, as Chip Gracey provides every bit of needed information, between his source code and his lectures and demonstrations on youtube.

Implementing the phoneme to vocal tract layer will be a bit more work, as there is a lot more to consider than with a simple diphone synthesis system. Now the transitions will become more of an issue, and we will be using a continuous change means of producing speech.

This is just a concept for implementing a speech synthesizer that should be able to produce reasonably clear speech that is smooth enough that it does NOT cause the listener to jolt in understanding what is being said.

Another Sub Project for Both:
I am also working on a layer to use the Amiga Narrator Device phoneme representations for whatever back-end speech synthesizer may be used. This way it should be possible to use the same lexicons with multiple possible back-ends in all cases.

This relates to the creation of a speech synthesizer in that it is the phonetic alphabet to be used for the synthesizer. The Amiga Narrator Device includes some things not given by a purely phonetic system, such as the emphasis of phonemes in words, thus making it possible to better form a lexicon, especially if you take the time to include the different ways in which words tend to be spoken depending on usage.

Eventually I would like to figure out how to add the phonemes and transition rules for other langues, as the Amiga Narrator Device rules only cover English, Germanic, and Latin based languages. It would be nice to be able to include Indic, Cyrillic and others eventually, thus allowing for people to extend the system for other languages.

[DavidS]

WOW. Looking at the rules for phonetic pronunciation, and how transitions are made in more detail than before gives a small beautifully simple, over-complex huge set of considerations to the implementation of correctly articulated speech synthesis.

The considerations of where most cheat in pronunciation are as important to clarity, as we are trained to understand word-phrases in the way they are commonly pronounced. This seems to be the single issue most missed by current speech synthesis.

[mikerr]

First sorry to hear about your problems, David -sounds like you're working through it.

There is a very simple android app called type and speak my grandad used:
https://play.google.com/store/apps/deta ... k&hl=en_GB (open source too !)

which has the advantage of android's word prediction engine - so you often only type 2 letters
Make sure to go into the config and select speak while typing, and it'll speak when each word is pressed, or when you press space between words.

[DavidS]

First sorry to hear about your problems, David -sounds like you're working through it.

There is a very simple android app called type and speak my grandad used:
https://play.google.com/store/apps/deta ... k&hl=en_GB (open source too !)

which has the advantage of android's word prediction engine - so you often only type 2 letters
Make sure to go into the config and select speak while typing, and it'll speak when each word is pressed, or when you press space between words.

That is the one I am using on Android (Type and Speak), it does the job, though the way that stored speech is accessed is difficult to quickly navigate, using a single monolithic list. I did write the author of Type and Speak an E-Mail thanking them for the app, as it has been of a great deal of aid.

Though typing on a touch screen is more difficult than typing on a keyboard, and I am having enough trouble typing do to my loss of dexterity at this time.

Also that app is a large part of the prod that is pushing me on this project.

I do thank you for the input.

[DavidS]

Sorry I typed that backwards three times. Type and Speak is what I meant.

[Gavinmc42]

Though typing on a touch screen is more difficult than typing on a keyboard,

Have you tried a bluetooth keyboard?
I know most are tiny and have pretty horrible keys, but typing on a touch screen with a reduced sense of touch has got to be worse.

Grab a bluetooth module and code it as a keyboard with real keys or gesture input
https://www.adafruit.com/category/255
Does BBC micro have HID mode?

Something like this has serious grunt to do a bit more than just keyboard.
https://www.adafruit.com/product/2995
It could send words , but making predictive code on it might be harder.
More use as ??? input to HID BT keyboard

[DavidS]

A bit more information, about the progress I am making thus far, in the form of the in progress documentation of the thought process that is going into this project.

So attached is a PDF of the thought process to this point, by me.

[DavidS]

Though typing on a touch screen is more difficult than typing on a keyboard,

Have you tried a bluetooth keyboard?
I know most are tiny and have pretty horrible keys, but typing on a touch screen with a reduced sense of touch has got to be worse.

Grab a bluetooth module and code it as a keyboard with real keys or gesture input
https://www.adafruit.com/category/255
Does BBC micro have HID mode?

Something like this has serious grunt to do a bit more than just keyboard.
https://www.adafruit.com/product/2995
It could send words , but making predictive code on it might be harder.
More use as ??? input to HID BT keyboard

Interesting ideas. I would need a newer tablet. My tablet does not have bluetooth, and only has 802.11g for WiFi.

I am actually looking for a newer tablet, or small laptop for that very reason. Though thank you.

[hippy]

This thread has inspired me to get off my backside and write a speech client for my Android phone using the in-built PicoTTS engine. That repeatedly polls a HTTP server for a text file, if the data is different to the last read, it speak its contents.

That will work with any host running a HTTP server. Any app can produce the text file which will then be spoken.

Not very fancy but it works and not bad for an hour's work with MIT App Inventor 2. Once I have it tidied up I'll look at the best way to publish that because it might be useful for someone.

AndroidThings for the Pi includes the PicoTTS engine so that should serve as well as an Android phone if one prefers a more Pi-oriented solution.

[DavidS]

Though typing on a touch screen is more difficult than typing on a keyboard,

Have you tried a bluetooth keyboard?
I know most are tiny and have pretty horrible keys, but typing on a touch screen with a reduced sense of touch has got to be worse.

Yes typing on a touch screen with a reduced sense of touch is very very difficult, with a normal keyboard my brain knows the layout well enough it is not as bad. Still a little work on a normal keyboard, though at least doable.

[DavidS]

This thread has inspired me to get off my backside and write a speech client for my Android phone using the in-built PicoTTS engine. That repeatedly polls a HTTP server for a text file, if the data is different to the last read, it speak its contents.

That will work with any host running a HTTP server. Any app can produce the text file which will then be spoken.

Not very fancy but it works and not bad for an hour's work with MIT App Inventor 2. Once I have it tidied up I'll look at the best way to publish that because it might be useful for someone.

AndroidThings for the Pi includes the PicoTTS engine so that should serve as well as an Android phone if one prefers a more Pi-oriented solution.

Very nice. It is to bad that hospitals block all internet and phone connections, otherwise something like that could be very useful for someone in my situation.

I am glad to see that this thread has inspired someone.

As posts are happening fast at the moment, a reminder of the PDF I posted a few back (PDF is attached to the post quoted bellow):

A bit more information, about the progress I am making thus far, in the form of the in progress documentation of the thought process that is going into this project.

So attached is a PDF of the thought process to this point, by me.

[Gavinmc42]

So attached is a PDF of the thought process to this point, by me.

I had forgotten the Prop sound stuff, seem to remember seven signing voices at the same time?
I can remember my first speech chip
https://en.wikipedia.org/wiki/General_Instrument_SP0256
That leads to LPC
https://en.wikipedia.org/wiki/Linear_predictive_coding
Which lead to CELP ACELP
https://en.wikipedia.org/wiki/Code-exci ... prediction
https://en.wikipedia.org/wiki/Algebraic ... prediction
A lot of this was for compressing voice, but these days we have serious CPU grunt to enhance voice.

Although you cannot speak this might be useful for analysis the TTS output for tuning.
https://en.wikipedia.org/wiki/Praat

Having married someone who speaks 8 or so languages, tones are important in other languages
I barely manage English and as a husband only need grunts now

https://en.wikipedia.org/wiki/Standard_ ... _phonology
That leads to phonology
https://en.wikipedia.org/wiki/Phonology

https://en.wikipedia.org/wiki/English_phonology
Horrible language English, glad I am a native speaker.
You might want to adopt the notation system for a dictionary of sounds.

[DavidS]

So attached is a PDF of the thought process to this point, by me.

I had forgotten the Prop sound stuff, seem to remember seven signing voices at the same time?
I can remember my first speech chip
https://en.wikipedia.org/wiki/General_Instrument_SP0256
That leads to LPC
https://en.wikipedia.org/wiki/Linear_predictive_coding
Which lead to CELP ACELP
https://en.wikipedia.org/wiki/Code-exci ... prediction
https://en.wikipedia.org/wiki/Algebraic ... prediction
A lot of this was for compressing voice, but these days we have serious CPU grunt to enhance voice.

A lot has changed. Now I can use the processing power to do a lot better than the old methods .

Although you cannot speak this might be useful for analysis the TTS output for tuning.
https://en.wikipedia.org/wiki/Praat

Yes that may be useful for the purpose of analyzing the output of speech synthesis.

Having married someone who speaks 8 or so languages, tones are important in other languages
I barely manage English and as a husband only need grunts now

https://en.wikipedia.org/wiki/Standard_ ... _phonology
That leads to phonology
https://en.wikipedia.org/wiki/Phonology

https://en.wikipedia.org/wiki/English_phonology
Horrible language English, glad I am a native speaker.
You might want to adopt the notation system for a dictionary of sounds.

Well the dictionary representations only cover 44 phonemes for English, to do English speech well I need at least 94 phonemes, and around 500 diphones, plus a lot more rules for how to manipulate the vocal tract simulation.

So the dictionary representations do not quite fit the bill. We have a horrible language, the most difficult of the 18 languages I know (19 if you count American Standard Sign Language).

[Gavinmc42]

We have a horrible language, the most difficult of the 18 languages I know

Now we know at least one of those degrees is linguistics and I will shut up about phonemes, tones etc.
That be like me teaching Einstein addition

If you can crack the language issues for natural sounding TTS, my robots will sound better
AI in games, virtual movie actors, translation programs, offline Babelfish....

Praat had some NN stuff in it, machine learning linguistics?
Instead of totally controlling what the TTS says, get it to learn to speak.

Which gets us to the high end voice recog stuff
http://www.slate.com/articles/technolog ... ingly.html

Instead of voice input they have keypad etc input.
So you only need the voice output bit.

[DavidS]

We have a horrible language, the most difficult of the 18 languages I know

Now we know at least one of those degrees is linguistics and I will shut up about phonemes, tones etc.
That be like me teaching Einstein addition

Rumor is that Einstein was not that good at processing simple arithmetic operations. Though that is likely nothing more than a rumor.

Linguistics is an interesting area of study. There is a lot more to it than most seem to realize. Though it was never a major for my studies, I minored in linguistics at one time, though never took linguistics as a primary study. Thus linguistics is not one of my degrees.

If you can crack the language issues for natural sounding TTS, my robots will sound better
AI in games, virtual movie actors, translation programs, offline Babelfish....

Not going for complete natural TTS, just more acceptable than the average if I can achieve that goal. The results should be interesting, as there is a lot to spoken language.

Praat had some NN stuff in it, machine learning linguistics?
Instead of totally controlling what the TTS says, get it to learn to speak.

Good point of view, and one that I am already looking at. I am taking many samples of spoken language (with the permission of the speakers) from daily conversation in order to attempt to create a model for simplified machine learning, in order to simplify the work of getting the rules for phoneme and diphone transitions in a vocal tract simulation correct. Though I am not sure of how much it will actually simplify the task yet.

Which gets us to the high end voice recog stuff
http://www.slate.com/articles/technolog ... ingly.html

Instead of voice input they have keypad etc input.
So you only need the voice output bit.

Interesting point of view. Thank you again for your input, helpful and thought provoking as always.

[Gavinmc42]

Interesting Amazon Alexa is Android based, so google and I found this.
https://github.com/willblaschko/AlexaAndroid
http://embedded-computing.com/28691-dec ... nd-beyond/
https://developer.amazon.com/public/sol ... ce-service
https://www.sayspring.com/

Not one for talking, even to machines I have pretty much ignored this area.
But there is some serious coding going on.

Like Windows in the early days, accessibility will be an issue.
Expect to see Alexa, Siri, Cortana etc take non verbal input, people will insist
As they do not need to process voice input in your case, off line for speech output?

More suited to the blind at the moment
https://www.abilitynet.org.uk/news-blog ... led-people
https://verysmartgadget.com/amazon-echo ... isability/
Alternative input for the mute?

http://fortune.com/2017/04/17/amazon-alexa-autism/
Leapmotion and sign language
https://devpost.com/software/hack-the-a ... ity-pgfc9a

Learned some magic words to google Lex, Polly
http://tech.firstpost.com/news-analysis ... 50758.html
https://inews.co.uk/essentials/news/tec ... mazons-ai/
https://aws.amazon.com/lex/
How to use Polly with no cloud?

Oh dear, another language for me to learn
https://en.wikipedia.org/wiki/Speech_Sy ... p_Language

How to use SSML off line?
http://www.cepstral.com/en/tutorials/view/ssml

Been putting audio amps into my Pi IoT stuff, was going to be for voice warnings.
Small speakers don't take up much room and TI have a nice i2s to amp chip.
People ignore flashing LEDs, not so device that start talking and saying what's wrong.
On my list of things to figure out. Maybe do TTS?
Pascal source for TTS?, er that's where I came in

Oh well, I learned something, things have been happening while I ignored speech.
Wonder what Alexa, Siri, Cortana look like to Praat?

[DavidS]

Interesting Amazon Alexa is Android based, so google and I found this.
https://github.com/willblaschko/AlexaAndroid
http://embedded-computing.com/28691-dec ... nd-beyond/
https://developer.amazon.com/public/sol ... ce-service
https://www.sayspring.com/

Not one for talking, even to machines I have pretty much ignored this area.
But there is some serious coding going on.

The one huge error in this new gen of speech recognition (I had used speech recognition as far back as the late 1980's), is that it relies on a remote server database for its versatility. This is the one thing that could be used to create problems by some company that wants to spy on people, and releases a product of a similar form that has the purpose of spying.

Though overall it is a great forward leap in technology.

Been putting audio amps into my Pi IoT stuff, was going to be for voice warnings.
Small speakers don't take up much room and TI have a nice i2s to amp chip.
People ignore flashing LEDs, not so device that start talking and saying what's wrong.
On my list of things to figure out. Maybe do TTS?
Pascal source for TTS?, er that's where I came in

Oh well, I learned something, things have been happening while I ignored speech.
Wonder what Alexa, Siri, Cortana look like to Praat?

Needless to say I am looking into Speech Synthesis and TTS quite a bit at this time . Ok maybe not as much as some.

Though I did come across a pascal implementation that was open source the other day, I am not remembering where, and I did not bookmark it.