I required ic that is use to remove noise from the voice for speech recognition

 

There is no dedicated chip that I am aware of. You could try building a a bandpass filter that goes from say 300 Hz. to 3000 Hz. My experience with listening to SSB signals on the amateur bands is that it will be of limited utility.

 

The few things I've read about speech recognition software, mentions alot of details on noise reduction. SOFTWARE that is. So, after bandpass, you should expect to need to implement some kind of limit on the data in your sample. "The rise per sample of the digital value must not exceed a certain amount" for example, would be a form of low pass filter. This is another form of digital filtering for noise spikes. Other software changes might be refered to as normalizing the data. For repeated iterations of the data, it might help to have software 'recognize' the high frequency speech portions, like the letter 's', or the letters 'c' and 'k', to know which filter to apply. You are trying to match a preformed/preloaded data pattern for a word, using phonems most often. Having the software make repeated 'tweaks' on the pattern and recompare it to confirm a match is key.

 

It is sort of difficult, unless you have an exact copy of the line noise, as used in early voice-scrambling solutions.

Otherwise, try to get the SNR high enough so filters can remove the noise while leaving a usable signal. DSP filters are rather in-depth, as pointed out above.

 

Could you somehow smooth the audio signal to remove pops and clicks?

 

If you've really got a software-based speech recognition system, this pre-filtering step is a trivial addition. If it doesn't seem like an easy thing to do, just wait till you get into the real programming.