|
first question, In the HTS2.2 SI straight Demo，the makefile.in file in the data folder, MGC is got from the sp file using the command as followed:
$(X2X) +af $${sp} | \ $(MGCEP) -a $(FREQWARP) -m $(MGCORDER) -l 2048 -e 1.0E-08 -j 0 -f 0.0 -q 3 > mgc/$${base}.mgc; \ the "-l" option is set 2048, maybe this demo's raw file sample rate is 48k. when I checked the HTS2.1 demo, the "-l" is set 1024 and HTS2.1 demo's raw file sample rate is 16k. I know the "-l" option is the fft length, and I remember in the SI Demo "-l" is set 512 for the 16k raw file, so why here is set 1024? why not set the others like 512? the second question is in HTS2.2, bap width is set 26, but in the HTS2.1, it is set 5, so why? for 48k speech, bap width is 26, but for 18k speech, the bap width is 5, so the "5" width is enough for the 16k speech? |