Need help with an interesting project

[pyoussef]

I’m working on a project for school and I could use some advice. I’m an electrical engineering background, but unfortunately, when it comes to audio engineering, I’m a bit in the dark.

Basically the scope of my project is to determine the effect of microphone quality on speech recognition. I’ve got a speech recognition engine already setup and I want to:
a) Reproduce a person’s voice on a speaker/set of speakers (from a 1 channel, PCM wav file recording)
b) Model background noise (a noisy street, background office noise, etc…) using a speaker/set of speakers (also played from a pre-recoded, 1 channel, wav file)
c) Capture that voice using one or several microphones of varying quality and store the result digitally.
d) Do an analysis, using the speech engine, to see if varying levels of microphone quality have any impact on the recognition of words/phrases.

My questions:
1) Can speakers accurately model human voice? If yes, what kind of speakers would I need? Would I be better off using an array of speakers or a single speaker?
2) When modeling background noise, could I use the same speakers I used to reproduce the voice or should I use a different setup?
3) If it is possible to generalize, what effect do cheaper microphones have on voice recording.

Assume I have a budget on the order of a couple of hundred bucks(~500)

Any help you guys could offer would be appreciated!

 

[markw]

just a thought.

Before going off on this project, consider that most people can recognize a voice over a telephone, which has about as limited a response as anything, including a pocket radio.

Now, what's the point of your project?

 

[mtrycrafts]

I’m working on a project for school and I could use some advice. I’m an electrical engineering background, but unfortunately, when it comes to audio engineering, I’m a bit in the dark.

Basically the scope of my project is to determine the effect of microphone quality on speech recognition. I’ve got a speech recognition engine already setup and I want to:
a) Reproduce a person’s voice on a speaker/set of speakers (from a 1 channel, PCM wav file recording)
b) Model background noise (a noisy street, background office noise, etc…) using a speaker/set of speakers (also played from a pre-recoded, 1 channel, wav file)
c) Capture that voice using one or several microphones of varying quality and store the result digitally.
d) Do an analysis, using the speech engine, to see if varying levels of microphone quality have any impact on the recognition of words/phrases.

My questions:
1) Can speakers accurately model human voice? If yes, what kind of speakers would I need? Would I be better off using an array of speakers or a single speaker?
2) When modeling background noise, could I use the same speakers I used to reproduce the voice or should I use a different setup?
3) If it is possible to generalize, what effect do cheaper microphones have on voice recording.

Assume I have a budget on the order of a couple of hundred bucks(~500)

Any help you guys could offer would be appreciated!

While this is a school project, that budget may be a drawback and certainly understandable.
You may want to see if there are any recording studios in your area and talk to them about mics and their impact, what they use, would they be willing to help you on the project by using their mic, perhaps in the studio?

Speakers do affect what you hear and probably make the bigger difference.
Also, there is another factor when you combine in this case speech and background noise. Its intensity makes a difference in word recognition and how much. Do a google search on 'cocktail party' effect. Lots of research in that area; it may or may not help you but worth a check. Also, if you are time limited, you may be in trouble.

 

[billnchristy]

PE has a 3" fullrange on sale for $8.88 I recently bought a pair of these, they sound very good and reproduce voice excellently.

Go to the website, click the sales flier in the top right corner, click on the page with Tang-band and click the 3" full range.

You could even go as minimalistic as cutting a hole in a cardboard box for it.