can't get same result as compute-mfcc-feats.

Question

can't get same result as compute-mfcc-feats.

bjtommychen opened this issue 7 years ago · comments

compute-mfcc-feats --window-type=hamming --dither=0.0 --use-energy=false --sample-frequency=8000 --num-mel-bins=40 --num-ceps=40 --low-freq=40 --raw-energy=false --remove-dc-offset=false --high-freq=3800 scp:wav.scp ark,scp:feats.ark,feats.scp

mfcc(signal=sig, samplerate=rate, winlen=0.025, winstep=0.01, numcep=40, nfilt=40, lowfreq=40, highfreq=3800,
appendEnergy=False, winfunc = lambda x: np.hamming(x) )

is there some difference ?

James Lyons · Answer 1 · Wed Sep 27 2017 09:34:43 GMT+0800 (China Standard Time)

In general, most mfcc programs wont give identical features due to offset and/or scaling differences, as well as any of a number of other implementation differences. When building features though, these differences won't matter, since the classifier will learn the supplied features just fine. The real test is whether you get similar accuracies at the output of the classifier, which should be the case. If you are actually looking for binary compatibility with your other feature generation script, it will likely be a pretty difficult job.

…

On 27 Sep 2017 11:05 AM, "Tommy Chen" ***@***.***> wrote: compute-mfcc-feats --window-type=hamming --dither=0.0 --use-energy=false --sample-frequency=8000 --num-mel-bins=40 --num-ceps=40 --low-freq=40 --raw-energy=false --remove-dc-offset=false --high-freq=3800 scp:wav.scp ark,scp:feats.ark,feats.scp mfcc(signal=sig, samplerate=rate, winlen=0.025, winstep=0.01, numcep=40, nfilt=40, lowfreq=40, highfreq=3800, appendEnergy=False, winfunc = lambda x: np.hamming(x) ) is there some difference ? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#50>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABn1QUUruvDvS6D1Zk9IBNWS3KEtUdyiks5smZ88gaJpZM4PlFA5> .

ZitengWang · Answer 2 · Tue Jan 09 2018 18:25:54 GMT+0800 (China Standard Time)

I am using a modified version of this code to produce the same results as with the compute-mfcc-feats and compute-fbank-feats commands in Kaldi. check here

Tommy Chen · Answer 3 · Tue Jan 09 2018 20:04:24 GMT+0800 (China Standard Time)

Hi, ZitengWang Thanks. I will check this.

…

On Tue, Jan 9, 2018 at 6:25 PM, ZitengWang ***@***.***> wrote: I am using a modified version of this code to produce the same results as with the compute-mfcc-feats and compute-fbank-feats commands in Kaldi. check here <https://github.com/ZitengWang/python_kaldi_features> — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#50 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAz0dPFQ7plLGaufnPMCUmmDjopDh0dVks5tIz60gaJpZM4PlFA5> .

-- Regards, Tommy, from Gmail.