|
Hello all, Can you plz tell whether Huffmann Coding (or adaptive Huffmann) method be directly applied to speech samples ? If yes, how much it helps to reduce bitrate ? and if no, why it can not be applied ? I am eagerly waiting for answer(although the question may be stupid). Thanx in advance. regards Jaydeep |
|
|
Can Huffmann coding be directly applied to speech samples ?
Started by ●January 16, 2004
Reply by ●February 3, 20042004-02-03
|
hi Jaydeep to derive the huffman codes for a given source, you need to know the probablity distribution for various symbols. Speech signal is highly random.. the loudness of speech can vary over abt 30 dB for the same speaker and generally there will be more than one speaker. hence I doubt if it is possible to estimate an optimal model of the probability distribution at the source, as far as speech is concerned. moreover, huffman coding is a lossless coding technique. most standard speech codecs use codebooks, which are lossy, but still give great compression. hence, directly representing speech samples using hufman coding may not be a very effective way of compressing speech. best regards, Sameer. -----Original Message----- From: Jaydeep Inamdar [mailto:] Sent: Friday, January 16, 2004 11:12 AM To: Subject: [speechcoding] Can Huffmann coding be directly applied to speech samples ? Hello all, Can you plz tell whether Huffmann Coding (or adaptive Huffmann) method be directly applied to speech samples ? If yes, how much it helps to reduce bitrate ? and if no, why it can not be applied ? I am eagerly waiting for answer(although the question may be stupid). Thanx in advance. regards Jaydeep |
Reply by ●February 3, 20042004-02-03
|
Sameer- > to derive the huffman codes for a given source, you need to know the > probablity distribution for various symbols. Speech signal is highly > random.. the loudness of speech can vary over abt 30 dB for the same > speaker and generally there will be more than one speaker. hence I doubt if > it is possible to estimate an optimal model of the probability distribution > at the source, as far as speech is concerned. > > moreover, huffman coding is a lossless coding technique. most standard > speech codecs use codebooks, which are lossy, but still give great > compression. hence, directly representing speech samples using hufman coding > may not be a very effective way of compressing speech. That is a great answer. I might add that I have heard of people doing research on applying lossless methods to the resulting bitstream; i.e. looking for patterns and consistent probability distributions in the compressed stream. The idea is that the codec itself removes certain variations in human speech due to speakers, loudness, background noise, etc. The tradeoff in these approaches seems to be latency, as a lot of bits need to be considered, making real-time and full-duplex communication difficult. U can kind of think like "zipping" up a chunk of compressed bitstream -- and how much delay that might pose in the worst-case. -Jeff > -----Original Message----- > From: Jaydeep Inamdar [mailto:] > Sent: Friday, January 16, 2004 11:12 AM > To: > Subject: [speechcoding] Can Huffmann coding be directly applied to > speech samples ? > > Hello all, > > Can you plz tell whether Huffmann Coding (or adaptive Huffmann) > method be directly applied to speech samples ? > If yes, how much it helps to reduce bitrate ? and if no, why it > can not be applied ? > I am eagerly waiting for answer(although the question may be stupid). > Thanx in advance. > > regards > Jaydeep |
|
|
Reply by ●February 4, 20042004-02-04
|
Dear All One more point to add on to this is regarding the concept of lossy and lossless data compression techniques. Although lossy techniques are branded as 'information lossy' the point to note is that we should address the compression from a psycoacoustic point of view with regards to speech. The question to ask in speech compression is can our ears actually perceive the "lossy"ness? Meaning is it really necessary to duplicate the original PDF function at the receiver end or can we use loudness percpetion techniques which say that the human ear works in logarithm scale. Speech as mentioned earlier exhibits a laplacian pdf function and techniques like mu-law, ADPCM, A-law among others swear by this.These exhibit artifacts in the error signal noise but if we take a look at mp3 coders, ac3 etc its basically using the quantization tables BUT with a physcoacoustic touch, its lossy but since the tables encompass filter functions catered for human acoustic system, allowing greater quantization noise to exist in certain frequency zones (critical band) than other areas, the ear simply cannot tell the difference. In short, Speech cannot be treated like raw binary data communication data and the use of loseless coding such as Huffman would be an overkill. Also as Jeff mentioned in real cases, codecs which use 1-bit based over sampling sigma delta techniques (using shaping filters) inherently condition the signal to keep the quantized noise away from the in-band speech region. If you were to process this signal with Huffman, I am not sure what will result. The condition signal will for one have greater SNR as compared to a SAR approximated codec......... anybody want to give a shot at this?? Dr. Arijit (from TU/Eindoven) its your call dude :) cheers Jeff Brower <> wrote: Sameer- > to derive the huffman codes for a given source, you need to know the > probablity distribution for various symbols. Speech signal is highly > random.. the loudness of speech can vary over abt 30 dB for the same > speaker and generally there will be more than one speaker. hence I doubt if > it is possible to estimate an optimal model of the probability distribution > at the source, as far as speech is concerned. > > moreover, huffman coding is a lossless coding technique. most standard > speech codecs use codebooks, which are lossy, but still give great > compression. hence, directly representing speech samples using hufman coding > may not be a very effective way of compressing speech. That is a great answer. I might add that I have heard of people doing research on applying lossless methods to the resulting bitstream; i.e. looking for patterns and consistent probability distributions in the compressed stream. The idea is that the codec itself removes certain variations in human speech due to speakers, loudness, background noise, etc. The tradeoff in these approaches seems to be latency, as a lot of bits need to be considered, making real-time and full-duplex communication difficult. U can kind of think like "zipping" up a chunk of compressed bitstream -- and how much delay that might pose in the worst-case. -Jeff > -----Original Message----- > From: Jaydeep Inamdar [mailto:] > Sent: Friday, January 16, 2004 11:12 AM > To: > Subject: [speechcoding] Can Huffmann coding be directly applied to > speech samples ? > > Hello all, > > Can you plz tell whether Huffmann Coding (or adaptive Huffmann) > method be directly applied to speech samples ? > If yes, how much it helps to reduce bitrate ? and if no, why it > can not be applied ? > I am eagerly waiting for answer(although the question may be stupid). > Thanx in advance. > > regards > Jaydeep _____________________________________ Note: If you do a simple "reply" with your email client, only the author of this message will receive your answer. You need to do a "reply all" if you want your answer to be distributed to the entire group. _____________________________________ About this discussion group: To Join: To Post: To Leave: Archives: http://www.yahoogroups.com/group/speechcoding Other DSP-Related Groups: http://www.dsprelated.com --------------------------------- Yahoo! Groups Links To Shree Jaisimha --------------------------------- |
