This is my first topic here on this rich forum and I am writing to seek help. Here is the background for a software program I want to develop. Please advise:
I am a (constantly learning) computer programmer with no background in audio programming :-(. A family friend’s dad is suffering from the Alzheimer’s Disease (AD). His case appears worse than moderate. His mom and his family have been the primary caregivers. The caregiver’s life, as you could perhaps imagine, is hard. I recently had a privilege to stay overnight with them to get a glimpse of how their life is, and I felt, in spite of the hardships, that perhaps I can develop an increasingly useful program to help them. The basic theme is to write a program to fool his dad into thinking that he’s conversing with his primary caregiver, his mom in this case. In order to do this, I, of course, need good hardware to record the conversations between his mom and dad and then after there is enough data, a (rudimentary) machine learning model would be developed. I hope this program then converses with dad in mom’s voice. I call this a very (very) rudimentary audio incarnation of the great program from 1976 by Joe Weizenbaum named ELIZA.
This is perhaps a grandiose goal, but one has got to start somewhere. So, here’s what I am thinking:
In the Record mode, it records the conversation between mom and dad. We then manually identify their voices and train a machine learning model (again, very rudimentary, but progressively more capable).
In the Play mode, the program will converse (“unintelligently”, but hopefully acceptable to his dad) with him. The hope is the mom gets a breather from her “duties”.
Is this worth it?
Is this at least somewhat feasible?
Any tips on the hardware, especially the mic 2) above?
Any other general advice on using Audacity is welcome.
During the initial development stage I would suggest using any old cheap mic that works.
We had some surprisingly good samples of voice recordings posted to the forum using a $15 Logitec USB desktop mic - this may not be the exact model, but it was this sort of thing:
The sound quality of those samples would certainly be good enough while you are developing the software. You could consider upgrading later if you need better quality.
Thanks, Steve. After reading Audacity wiki however, I thought that I need a good sound card. So, I opted for 1) which accepts the mic audio mini-plug. It does not accept a USB mic.
The recording schematic I am visualizing is:
|Mic with audio mini-plug| → |Syba external USB sound card on the laptop| → |laptop|
Or are you suggesting that I buy the sound card 1), connect it to the recording laptop and buy the simple USB mic you suggest and plug it directly into the laptop’s USB port?
I expect that the major problems with your project will be with the software development, so I’d suggest that you hold off spending money on equipment until such time as you have at least a working prototype. Once you have that, then there are several different approaches to getting good quality audio in and out of the computer. The sound quality of the mic input of most “on board” sound cards is very poor, but even that will be good enough to know if your prototype works.
For recording, there are many non-technological considerations as well as technological ones - for example, are you able to make the recordings in a studio? We can discus these issues when you get to that stage in your project, but at this stage I would recommend that you concentrate your efforts on making a prototype and so assessing the feasibility of the project.
Sounds good! I too don’t want to overcommit anything at this stage. Like always, I will start with sticking to minimalism. Will report back on this thread, hopefully soon.
I agree with others - don’t get hung up on the audio quality of the microphone and input, any reasonable quality will work to debug the software.
Once you are getting closer though, I would think a bigger consideration would be the sound output. Reproduction of single voice speaking requires a decent amplifier and speakers to sound realistic. There is a lot of energy at the fundamental and a few harmonics, that takes some peak power and speakers that can handle that peak power. But you won’t need deep bass (especially for a typical female voice). Music is different, that sound is spread more evenly across the spectrum. That is why PA speakers are designed different from high-fidelity music reproducers.
And even though the subject has dementia, at some basic level they might recognize reproduced sound from original if it isn’t of high quality, I don’t know. Recall that some people with very poor, or dysfunctional memories can sing the lyrics to songs. Sound and our brain have a very tight,basic connection.
This isn’t difficult to achieve, it’s just a consideration - and laptop speakers won’t come close (but again, fine for testing the software). Most of the cheap ‘computer speakers’ focus on a (relatively) big subwoofer to fool people into thinking they are getting ‘big’ sound, but the satellites don’t handle much power - and that is where the voice will be.
Again, once you get near that point, I’m sure people here can offer suggestions.
Good luck with the project, my FIL passed away recently, before his dementia got too bad (from an illness unrelated to the dementia), so I know a little of what these people are going through. It is good of you to try to help, I hope your project brings them some relief.
For a first pass of the software, maybe even skip the ‘intelligent’ ELIZA format, just detect the subject’s voice and play a random recording as a ‘reply’ - that might even be enough to soothe the subject?
Good luck with the project, my FIL passed away recently,
Sorry to hear about your loss
before his dementia got too bad (from an illness unrelated to the dementia), so I know a little of what these people are going through. It is good of you to try to help, I hope your project brings them some relief.
Exactly! I just want to write a rule-based program to begin with. Thank you for the encouragement.
For a first pass of the software, maybe even skip the ‘intelligent’ ELIZA format, just detect the subject’s voice and play a random recording as a ‘reply’ - that might even be enough to soothe the subject?
(http://www.amazon.com/gp/product/B001MSS6CS/ref=ox_sc_act_title_2?ie=UTF8&psc=1&smid=ATVPDKIKX0DER)