Jose Ignacio Hinestrosa

Please see below for details about the project. I need to be able to deliver the final prototype in about 3 weeks. Yes, it is a paid gig.

Create an application that will "recover" the "natural" voice of those who have lost it due to cigarrette smoking. Many patients suffering from this disease have to go through a tracheostomy operation which implies losing their vocal cords. In most of these cases a "speaking valve" will be used in order to allow patients to speak and communicate. The sound coming out from this mechanism is very "synthetic" and "roboticized" and we want to be able to output a more human like voice and ideally the lost voice of the person.

In very general terms this is what the system will do :

1. A microphone that will be attached to the person and will allow an optimum signal level into the computer.

2. Once MAX captures this signal it will transform speech into text. - speech to text recognition -

3. Recreate the speech in realtime but the sound output coming from the computer will be generated either by samples of the original prerecorded voice or some sound synthesis mechanism that will resemble the acoustic characteristics of the original voice.

I think the first question for whoever is interested would be if this is even possible : reproducing the speech of one voice with the sound of another voice in realtime.

Here are some patches that i have found and that relates to this topic :

Please see below for details about the project. I need to be able to deliver the final prototype in about 3 weeks. Yes, it is a paid gig. 

Create an application that will "recover" the "natural" voice of those who have lost it due to cigarrette smoking. Many patients suffering from this disease have to go through a tracheostomy operation which implies losing their vocal cords.  In most of these cases a "speaking valve" will be used in order to allow patients to speak and communicate. The sound coming out from this mechanism is very "synthetic" and "roboticized" and we want to be able to output a more human like voice and ideally the lost voice of the person. 

In very general terms this is what the system will do : 

1. A microphone that will be attached to the person and will allow an optimum signal level into the computer.
2. Once MAX captures this signal it will transform speech into text. - speech to text recognition -
3. Recreate the speech in realtime but the sound output coming from the computer will be generated either by samples of the original prerecorded voice or some sound synthesis mechanism that will resemble the acoustic characteristics of the original voice. 

I think the first question for whoever is interested would be if this is even possible : reproducing the speech of one voice with the sound of another voice in realtime. 

looking-for-msp-programmers-asap

I'd be interested in this. mail me to ptrk.lechner@gmail.com

I'm quite certain though, that this is a huge task. In 3 weeks, what do you want to deliver? A proof of concept?

I think overcoming the uncanny valley here is very difficult. Making the sound more pleasant won't be hard I guess, but make it more 'human', or make it more like the persons voice will be very hard. Working with simple samples won't be an option I think.

Hi,
I'd be interested in this. mail me to ptrk.lechner@gmail.com
I'm quite certain though, that this is a huge task. In 3 weeks, what do you want to deliver? A proof of concept?
I think overcoming the uncanny valley here is very difficult. Making the sound more pleasant won't be hard I guess, but make it more 'human', or make it more like the persons voice will be very hard. Working with simple samples won't be an option I think.


For sure. I think not reinventing the wheel would be ideal, and somehow hooking into an already existing speech lib would be neccesary to have anything that works in 3 weeks.

So yeah, it's two rather tricky elements; 1 is speech recognition from one of those buzzer voices - can that even be done? And 2 is hooking into a speech lib of some kind. Can you license the one google uses?

For sure. I think not reinventing the wheel would be ideal, and somehow hooking into an already existing speech lib would be neccesary to have anything that works in 3 weeks. 

So yeah, it's two rather tricky elements; 1 is speech recognition from one of those buzzer voices - can that even be done? And 2 is hooking into a speech lib of some kind. Can you license the one google uses?


Thank you guys for your valuable insights. Yes the idea is to not reinvent the wheel and to hook up to an existing speech Lib. This is just to proof the concept of getting the buzzer voice into a speech recognition API and getting an output with a more pleasant and natural voice. For this stage it doesnt have to be the exact persons voice. I was thinking to attach a wireless mic to the person and this way get a more clear sound input into the speech recognition API. Something like a Lavalier mic maybe ?

Hey Woyteg ! Just sent you a personal email. Please check and let me know. thx.

Part one is kind of a dealbreaker, if it doesn't work. I say start by seeing if you can even *do* voice recognition with speech impediments.

Wow, This is a really interesting project.

For the hardware, have you considered using a throat mic like the ones used by soldiers and law enforcement agencies? Supposedly, high end throat mics can clearly transmit a whisper while remaining impervious to high volumes of ambient noise...like guns and bombs and stuff I guess. Sounds nuts right? I have no direct experience with these I just know of their existence.

http://teaheadsets.com/index.php/2-wiresystems/lash.html

also, what of the possibility of simply amplifying and processing the voice coming from the speech valve to humanize and project it? if such were possible the software and subsequently the hardware required to run it would be considerably more light weight.

Just some thoughts that ran through my head

Wow, This is a really interesting project. 

http://teaheadsets.com/index.php/2-wiresystems/lash.html 

also, what of the possibility of simply amplifying and processing the voice coming from the speech valve to humanize and project it? if such were possible the software and subsequently the hardware required to run it would be considerably more light weight. 

Just some thoughts that ran through my head


I also think that that's a more robust solution. I don't think the speech recognition is a good idea for getting this going. I don't know enough about what exactly is going on anatomically/what these robot voices are analyzing/producing. But I think a LPC model would be the right way to do this. Trying to identify exciter/resonator, process the resonator, calculate a new filter and excite it through the incoming extracted exciter signal. ?

the two most important factors of a realistic voice (in the sense of comparing the persons lost real voice against the monotone sound of the current solutions available) are:

at this point i dont see how you could (re)create the intonation part from the whispering the person speaks. there is nothing left what could trigger and control it and using "presets" will not work either.

Thx CPTNFANTASY for your thougthts and link. its an awesome idea.

Hey have you guys seeing this ? : www.vocalid.co

Do you guys think this technology will help to solve some of the issues ?

Any of you guys interested in workin in this project ? Its a paid gig.

Jose this seems like a project that someone would base a PhD on, I'm a bit skeptical as to whether you can hook a few libraries/objects and get anything working within the time you asked. Have you tried advertising the job? Is it for a government or private company?

Hi Mike S ! thx for your comments. the idea at this stage is to show a proof of concept not a fully worgking app with 100% accuracy. Its in the context of a marketing campaign that will develop by the Public Health sector in Miami in order to help create conscious about the risks of Tobacco smoking. As was said before i do think that the starting point is if its even possible to get a clear voice from people with this disease and then use a speech recognition api to transfer this into text. Then do a text to speech conversion and output the text with a more pleasant and natural voice.

Hi Mike S ! thx for your comments. the idea at this stage is to show a proof of concept not a fully worgking app with 100% accuracy. Its in the context of a marketing campaign that will develop by the Public Health sector in Miami in order to help create conscious about the risks of Tobacco smoking. As was said before i do think that the starting point is if its even possible to get a clear voice from people with this disease and then use a speech recognition api to transfer this into text.  Then do a text to speech conversion and output the text with a more pleasant and natural voice.


but I think the result will most probably be a pedagogic one-

if you smoke,you will loose your voice and even max/msp can't bring it back.

if it's not about that,I can offer you to do extensive experimentation and research on the issue,

but I cant guarantee for useful results or even a prototype within 3 weeks..

Hi Jose,
I'm also interested since it's a job--
but I think the result will most probably be a pedagogic one-
if you smoke,you will loose your voice and even max/msp can't bring it back.
if it's about that,I'm in. ;)

if it's not about that,I can offer you to do extensive experimentation and research on the issue,
but I cant guarantee for useful results or even a prototype within 3 weeks..


I understand your point RBRT. Can we do a skype meeting today or tomorrow? Let me know what works for you. You can also write directly to mi email : syncmassim@gmail.com

This is my skype id : joseignacio111 . If youre going to contact me please let me know your id in advance so i know its you..

This is my skype id : joseignacio111 .  If youre going to contact me please let me know your id in advance so i know its you..

"Here are some patches that i have found and that relates to this topic :"

"Here are some patches that i have found and that relates to this topic :"
srsly


yes vichug i have seeing them and very helpful. thc again.

Looking for MSP Programmers ASAP.