configure file missing

[mcoto]

I can see the configure.ac file on the project, but I think it also needs the configure file to run the demo.

[MattShannon]

Please follow the instructions in INSTALL, as per the Readme. configure is generated from configure.ac using autoconf. I believe standard practice is to include configure in released tarballs, etc but not check it in to version control, in order to keep the repository clean and ensure it is always consistent. Let me know if you have any other trouble.

On Sat, 03 Jan 2015 08:42:45 -0800 mcoto notifications@github.com wrote:

I can see the configure.ac file on the project, but I think it also needs the configure file to run the demo.

---

Reply to this email directly or view it on GitHub: https://github.com/MattShannon/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree/issues/1

[mcoto]

Thank you so much for your indications on autoconf. Now, I was able to execute the demo, but just at the beggining of embedded re-estimation (monophone) I got this error:

ERROR [+2319] HERest: Unknown switch i FATAL ERROR - Terminating program /home/marvin/HTS/HTS-2.1/htk/HTKTools/HERest Error in /home/marvin/HTS/HTS-2.1/htk/HTKTools/HERest -A -C /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/configs/trn.cnf -D -T 1 -S /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/data/scp/train.scp -I /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/data/labels/mono.mlf -m 1 -u tmvwdmv -w 3 -t 1500 100 5000 -i 1 40 -H /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/models/qst001/ver1/cmp/monophone.mmf -N /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/models/qst001/ver1/dur/monophone.mmf -M /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/models/qst001/ver1/cmp -R /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/models/qst001/ver1/dur /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/data/lists/mono.list /home/marvin/HTS/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree-master/data/lists/mono.list

I'm wondering if the error is due to a bad version of some of the program. I'm working with: HTK-3.4.1 , SPTK-3.4.1, HTS-2.1, hts_engine-API-1.05, Matlab R2013a, STRAIGHTV40pcode.

Thanks in advance for your kind attention.

2015-01-03 12:11 GMT-06:00 Matt Shannon notifications@github.com:

Please follow the instructions in INSTALL, as per the Readme. configure is generated from configure.ac using autoconf. I believe standard practice is to include configure in released tarballs, etc but not check it in to version control, in order to keep the repository clean and ensure it is always consistent. Let me know if you have any other trouble.

On Sat, 03 Jan 2015 08:42:45 -0800 mcoto notifications@github.com wrote:

I can see the configure.ac file on the project, but I think it also needs the configure file to run the demo.

---

Reply to this email directly or view it on GitHub:

https://github.com/MattShannon/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree/issues/1

— Reply to this email directly or view it on GitHub https://github.com/MattShannon/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree/issues/1#issuecomment-68603778 .

[pineking]

mcoto,

you need to patch the HTS tools first, using the auto-regressive HMM for HTS patch file, provided by Matt Shannon.

you can find this patch file from his website: http://mi.eng.cam.ac.uk/research/emime/ar-for-hts/

[MattShannon]

Hi mcoto,

That's correct. In addition, I would say that if you're just starting out in speech synthesis research, it's probably a good idea to first run the conventional HTS demo available from the HTS website before running the autoregressive HTS demo present in this repository.

Good luck!

On Sun, 04 Jan 2015 19:09:19 -0800 pineking notifications@github.com wrote:

mcoto,

you need to patch the HTS tools first, using the auto-regressive HMM for HTS patch file, provided by Matt Shannon.

you can find this patch file from his website: http://mi.eng.cam.ac.uk/research/emime/ar-for-hts/

---

Reply to this email directly or view it on GitHub: https://github.com/MattShannon/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree/issues/1#issuecomment-68663660

[mcoto]

Thank you very much!

I happily confirm the code is running for mexican spanish voices!

The only problem I have now is in changing the code for working with 48 kHz audio files instead of the default 16kHz. For 48 kHz, I changed the lines of the configure.ac file:

if test -z "$FRAMESHIFT"; then FRAMESHIFT=240 fi

if test -z "$SAMPFREQ"; then SAMPFREQ=48000 fi

if test -z "$FREQWARP"; then case $SAMPFREQ in 8000) FREQWARP=0.31 ;; 10000) FREQWARP=0.35 ;; 12000) FREQWARP=0.37 ;; 16000) FREQWARP=0.42 ;; 22050) FREQWARP=0.45 ;; 32000) FREQWARP=0.45 ;; 44100) FREQWARP=0.53 ;; 48000) FREQWARP=0.55 ;; *) FREQWARP=0.00 ;; esac fi

But got the following error, just in the beggining of "Start generating speech parameter sequences (1mix)":

ERROR [+9999] Cholesky_Factorization: (853,853)-th element of W'_U^{-1}_W is negative

Everything works fine if I downsample the files to de default sample rate. Thanks again and best regards,

Marvin.

[MattShannon]

Hi Marvin,

Glad to hear you've got it working for the 16kHz case!

Are you saying parameter generation fails even for the very first utterance it tries to generate for? What if you try a different utterance?

To explain why I'm asking, the error is sometimes caused by a combination of slightly badly trained models and numerical issues (in which case the element referred to will be slightly negative) and sometimes caused by something major going wrong (in which case the element referred to will be very negative). In the first case adding a small constant (say 1e-6) to every element of the diagonal of the matrix before performing the Cholesky factorization should suffice to solve the problem, and I can probably come up with a patch to do that fairly easily, but I'd like to establish this is the problem first.

On Fri, 09 Jan 2015 07:51:00 -0800 mcoto notifications@github.com wrote:

Thank you very much!

I happily confirm the code is running for mexican spanish voices!

The only problem I have now is in changing the code for working with 48 kHz audio files instead of the default 16kHz. For 48 kHz, I changed the lines of the configure.ac file:

if test -z "$FRAMESHIFT"; then FRAMESHIFT=240 fi

if test -z "$SAMPFREQ"; then SAMPFREQ=48000 fi

if test -z "$FREQWARP"; then case $SAMPFREQ in 8000) FREQWARP=0.31 ;; 10000) FREQWARP=0.35 ;; 12000) FREQWARP=0.37 ;; 16000) FREQWARP=0.42 ;; 22050) FREQWARP=0.45 ;; 32000) FREQWARP=0.45 ;; 44100) FREQWARP=0.53 ;; 48000) FREQWARP=0.55 ;; *) FREQWARP=0.00 ;; esac fi

But got the following error, just in the beggining of "Start generating speech parameter sequences (1mix)":

ERROR [+9999] Cholesky_Factorization: (853,853)-th element of W'_U^{-1}_W is negative

Everything works fine if I downsample the files to de default sample rate. Thanks again and best regards,

Marvin.

---

Reply to this email directly or view it on GitHub: https://github.com/MattShannon/HTS-demo_CMU-ARCTIC-SLT-STRAIGHT-AR-decision-tree/issues/1#issuecomment-69352401

[mcoto]

Thanks again for your help.

I've been working with different 48k databases for several days, and the error appears every time with all the utterances. I mean, it appears for the first one, and after deleting it, the problem goes for the second one and so on.