Speech recognition for a newbie

Started by ngolehung84 July 17, 2009
Hi all,

I'm very new with both DSP and Speech recognition. So could you please
let me know:

+ What type of filters used to remove noise since my input is taken in a
noisy environment.
+ Some documents that discuss the frequency range of human voice?
+ Other types of filters?
+ Algorithms/techniques/models used to retrieve and match
sample/pattern.

These are all I think I need to work with speech recognition. If I miss
anything, please let me know.

I really appreciate if you guys can give me a hand to start studying
this field.

Thanks in advance,
Pete.
Hey man!

To answer a few of your queries:
-----------------------------
1) Noise Cancellation is a very hot research field and there are numerous methods to do so. I will suggest a few noise cancellation methods in the ascending order of difficulty - spectral subtraction (multiband inclusive), frequency scaling methods, kalman filtering, global soft decision method, signal subspace methods and auditory scene analysis / blind source separation. For just a start you would want to work on spectral subtraction.

References:
Multiband spectral subtraction for speech enhancement [Sunil Devdas Kamant, MS Thesis]

speech enhancement for personal communications using an adaptive gain equalizer. Neils Westerlund, Mattias Dahl and Ingvar Claesson
-----------------------------
2) Regarding the frequency range of human utterance, take it from me that it wont exceed 4kHz, so 8kHz sampling is just fine.
-----------------------------
3) Other types of filters for noise cancellation: well, wavelets. but most filterbanks are used to split the signal into sub-bands.
-----------------------------
4) Pattern Matching: just go through the chapter on HMMs from rabiner, or the famous tutorial for that matter... You just need to know how to fit a cluster of points in space into gaussians. That should give you a good start for speech recognition.
-----------------------------