Hi,
I extracted BAP from AP, and then reconstructed AP from BAP without HTS-training
[Makefile]
echo "Converting aperiodicity file $${ap} to band aperiodicity file bap/$${base}.bap"; \
$(X2X) +af $${ap} | $(BCP) +f -n 512 -L 64 -s 0 -e 63 -S 0 | $(AVERAGE) -l 64 > bap1; \
$(X2X) +af $${ap} | $(BCP) +f -n 512 -L 64 -s 64 -e 127 -S 0 | $(AVERAGE) -l 64 > bap2; \
$(X2X) +af $${ap} | $(BCP) +f -n 512 -L 128 -s 128 -e 255 -S 0 | $(AVERAGE) -l 128 > bap3; \
$(X2X) +af $${ap} | $(BCP) +f -n 512 -L 128 -s 256 -e 383 -S 0 | $(AVERAGE) -l 128 > bap4; \
$(X2X) +af $${ap} | $(BCP) +f -n 512 -L 129 -s 384 -e 512 -S 0 | $(AVERAGE) -l 129 > bap5; \
$(MERGE) -s 0 -l 1 -L 1 bap1 bap2 | \
$(MERGE) -s 2 -l 2 -L 1 bap3 | \
$(MERGE) -s 3 -l 3 -L 1 bap4 | \
$(MERGE) -s 4 -l 4 -L 1 bap5 > bap/$${base}.bap; \
[Training.pl]
# convert band-aperiodicity to aperiodicity
$bap = "$gendir/$base.bap";
shell("$BCP +f -l 5 -L 1 -s 0 -e 0 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 64 | ${DFS} -a 1 -1 > $gendir/$base.ap1");
shell("$BCP +f -l 5 -L 1 -s 1 -e 1 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 64 | ${DFS} -a 1 -1 > $gendir/$base.ap2");
shell("$BCP +f -l 5 -L 1 -s 2 -e 2 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 128 | ${DFS} -a 1 -1 > $gendir/$base.ap3");
shell("$BCP +f -l 5 -L 1 -s 3 -e 3 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 128 | ${DFS} -a 1 -1 > $gendir/$base.ap4");
shell("$BCP +f -l 5 -L 1 -s 4 -e 4 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 129 | ${DFS} -a 1 -1 > $gendir/$base.ap5");
$line = "$MERGE -s 0 -l 64 -L 64 $gendir/$base.ap1 $gendir/$base.ap2 | "
. "$MERGE -s 128 -l 128 -L 128 $gendir/$base.ap3 | "
. "$MERGE -s 256 -l 256 -L 128 $gendir/$base.ap4 | "
. "$MERGE -s 384 -l 384 -L 129 $gendir/$base.ap5 > $gendir/$base.ap";
shell($line);
The commands above are from the scripts in "HTS-demo_CMU-ARCTIC-SLT+STRAIGHT" package.
I found that the reconstructed AP data is very much different from original AP.
The reconstructed AP data seems rather abnormal (or unstable).
I also guess that the problem might be cause by "${DFS} -a 1 -1"
Even though the reconstructed AP looks strange, the speech is still synthesized.
Is there anyone who has similiar results?
My test environment is "XP + Cygwin"
SangJin Kim
On Sun, Aug 31, 2008 at 12:23 PM, Xingchi Xian
<yeahxxc@xxxxxxxxx> wrote:
Hi all,
I am wondering if cmd "$BCP +f -l 5 -L 1 -s 0 -e 0 -S 0 $bap | ${DFS} -b 1 -1 | ${INTERPOLATE} -p 64 | ${DFS} -a 1 -1 > $gendir/$base.ap1" in training.pl will cause unstable output, and I think it did happen in my training process which generated abnormal ap coefficients.
The problem might be cause by "${DFS} -a 1 -1".
Xian Xingchi