Why the value of `sp` is so large

[candlewill]

I try to extract acoustic parameters using the tool examples/analysis_synthesis/analysis. I can get the three parameters. But when I check the values of the sp file. I found that abs of some value is very very large, for example, -1.59584e+165, 7.15049e+43.

analysis 000001.wav 000001.f0 000001.sp 000001.ap

The way I used to get the value of sp file is: x2x +da 000001.sp. x2x bin is from the latest SPTK tool.

Could anyone help explain if there is something wrong?

The wave to analysis is attached. 000001.zip

[mmorise]

The header of the attached file shows that it is not a linear PCM. File loader in WORLD supports only the simple linear PCM.

http://soundfile.sapp.org/doc/WaveFormat/ Subchunk1Size is generally 16 for PCM, but the attached file's size is not 16. By modifying the header, I could read and analyze the file. The result did not include such large values.

[candlewill]

Thanks @mmorise . That solved my problem.

[megazone87]

The header of the attached file shows that it is not a linear PCM. File loader in WORLD supports only the simple linear PCM.

http://soundfile.sapp.org/doc/WaveFormat/ Subchunk1Size is generally 16 for PCM, but the attached file's size is not 16. By modifying the header, I could read and analyze the file. The result did not include such large values.

Hello @mmorise ! I checked the Subchunk1Size of wav that @candlewill supplied, it is 16.

Could you check this wav again or describe how you modified the header?

[mmorise]

I checked the file and confirmed that it was not the same as the old one. The URL to download the file seems to be changed.

Old: https://github.com/mmorise/World/files/2378030/000001.zip New: https://github.com/mmorise/World/files/2378095/000001.zip

So, you would be able to load new file by using the loader in WORLD.

[megazone87]

I see. thank you！👍🏻

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: mmorise notifications@github.com 发送时间: Thursday, January 24, 2019 8:54:38 PM 收件人: mmorise/World 抄送: Meixu Song; Comment 主题: Re: [mmorise/World] Why the value of sp is so large (#70)

I checked the file and confirmed that it was not the same as the old one. The URL to download the file seems to be changed.

Old: https://github.com/mmorise/World/files/2378030/000001.zip New: https://github.com/mmorise/World/files/2378095/000001.zip

So, you would be able to load new file by using the loader in WORLD.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/mmorise/World/issues/70#issuecomment-457186372, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AArELvxS4AUpIggppeUqC2xq8whRZdDMks5vGa0OgaJpZM4WmsIA.

[candlewill]

Thanks @mmorise . I downloaded the wav from the new 2378095/000001.zip. The sp parameters still contains some big values: command:

./build/analysis 000001.wav  f0 sp ap
x2x +da sp | head

result:

7.90505e-320
-3.80046e-118
-1.10136e+190
-2.02275e-201
-1.60101e-49
-1.61272e-295
1.29486e-216
5.34522e+78
5.95244e+26
58.0803

[mmorise]

I tested the file by using an example in example/parameter_io.

./build/f0analysis.exe 000001.wav
./build/spanalysis.exe 000001.wav output.f0

You can obtain output.sp as the parameter in spectral envelope.

As a result, such big values was not contained in the file. I think that the example contains a bug.

Since I'm not the developer in the example, I can't support to debug it. I recommend you to use the example in parameter_io.

[candlewill]

Thanks, I would try it.