Speech Recognition Gurbanee Search Program

[HSingh]

Vaheguroo Jee Ka Khalsa! Vaheguroo Jee Kee Fateh!

Now I already know that we have sikhitothemax and many more programs like that now, however I was thinking wouldnt it be nice to have a program like that but with speach recognition?

For those of you who dont know what that is, its basicly when you talk it will recognise what your saying and input it into the system. This is normally used mainly for programs like word, so instead of typing you just speak and it will type it for you. Now for a search program we would want it to search what is being said instead of type, and then it will bring up the search result, in our case the gurbanee.

Now this would be most useful at gurdwaras or house programs. For example at akhand paths screens (when I say screen I mean sttm) are not normally done all the way threw, only at the begining and end. However with speach recognision, the program could follow the akhand path and nobody will have to sit at the computer all the time to opertate the screens, instead the program will scroll threw the slides itself instead, (obviosly it wont be using powerpoint and will need to be built into the program instead of useing powerpoint for this to work properly, as the default office speach recognition tool is rubbish and wont recognise gurmukhi or punjabi correctly). So in this case people would stay longer becouse they can follow along with the banee.

Another example would be if there is a kirtan program at the gurdwara one day and nobody is available that day to operate the screens, then this would come in handy becouse all the people who are there would need to do is start up the program and it will do the rest itself. So it will bring up the kirtan that is being played and etc.

Now this program will obviosly need to be self learning so it can keep improving itself over time with the speach recognition.

However my quetion is, if this kind of software was made then would anyone really use it? Because there is no point in it being made and then nobody using it. Do you think it would be a good idea to do something like this? If yes then please post any sugetions you would want to add to it, if no then please KINDLY state the reason why not.

Vaheguroo Jee Ka Khalsa! Vaheguroo Jee Kee Fateh!

[JaMdooT]

As far as the speech recognition software available in the market, it tends to be ignorant or simply make some next level shyt up depending on who is speaking it and how something is said. I worked on active noise cancellation for my design project and I would say that speech recognition is pretty intense problem, due to the signal variability due to more than a single source.

The issues related to low signal-to-noise ratio is a factor which seems to be the major problem to most of the system till date. If the software isn't able to distinguish the spoken words distinctly or if there is any extra noise in the background, the interference simply causes flawed results.

Since you are suggesting gurdawara kirtan programs and house kirtans, I am pretty sure the overlapping speech would be the issue you should consider when progressing with this idea. Also, there are tonnes of similar sounding words, which we only distinguish in the complete phrase and it will be the worst nightmare for speech recognition system, so do look into it to make sure to avoid projecting a different shabad on different notes.

I am in no way discouraging the idea bro, I think its a Wonderful idea but if it doesn't deliver at the end, it wouldn't be useful at all. I just wanted to give you my 2cents and wish you the best, if you do choose to proceed with this project.

[HSingh]

Vaheguroo Jee Ka Khalsa! Vaheguroo Jee Kee Fateh!

I know most speach recognition tools out there are rubbish, however there are some that are very good and self learning as well. So as you say it may get mixed up, however it will learn over time (now when I say learn over time I dont mean it will automatically correct its mistakes, it wont because it wont know if its wrong or not). What I mean is, to start of with we can make it so it will give you results like the other gurbanee search programs do and then you click on the correct result. Over time it it will get better at searching and as we will keep on releasing regular updates to it, (if we do start this that is) then it will start to learn the best results and we will start it to bring up the shabad automatically. So yes it will at times make mistakes, and will to begin with need someone there just to check up on it, however as we keep on improving it then it shall get better at it. So to achieve this what we will need to do is for everyone who does use it, to be connected to the internet when using the program so we can collect data from it, however we know not all gurdwara's have net connection so there will be a of line mode as well.

As for background noise, there is already noise reduction technology out there, however yes at times it can get a bit too noisy, for that I would suggest instead of using a microphone, to use line in, (connect the computer to the amp where the main mics are attached to so you wont get much background noise.

Vaheguroo Jee Ka Khalsa! Vaheguroo Jee Kee Fateh!