Wrong frames number

[MrChoclate]

Hello I run into an issue with the frames numbers.

I try to decode an audio book (http://ia801400.us.archive.org/7/items/jekyll_and_hyde_klh_0904_librivox/jekyll_01_stevenson_64kb.mp3)

configuration = Pocketsphinx::Configuration.default
decoder = Pocketsphinx::Decoder.new(configuration)
decoder.decode 'jekyll_01_stevenson.wav'

puts decoder.words

There is something wrong with start_frame and end_frame.
http://pastebin.com/a3zRxdH9

I expected the first start_frame to be 0.

This is my audio metadata:

  Channels:              1
  Bits per sample:       16
  Samples per second:    16000
  Bytes per second:      32000
  Block align:           2
  Sample frame count:    13682957

If I use 84749 as my original frame: I got 13682957 / (165414 - 84749.0) = 169.626938573111 which should be 160 since a pocketsphinx frame is 10ms.

Using pocketsphinx_continuous -infile jekyll_01_stevenson.wav -time yes
I got the correct times:

<s> 1.380 1.600 0.999300
chapter 1.610 1.980 0.864662
one 1.990 2.360 0.556734
<sil> 2.370 2.500 0.949038
of 2.510 2.600 0.820433
the 2.610 2.680 0.414533
strange 2.690 3.200 0.998601
case 3.210 3.550 0.999800
of 3.560 3.640 0.567698
dr(2) 3.650 4.000 0.734228
jekyll 4.010 4.350 0.999800
and 4.360 4.480 0.636639
mr 4.490 4.800 0.999900
hyde 4.810 5.340 0.986096
</s> 5.350 5.580 1.000000

[nshmyrev]

Maybe you can use cmusphinx/pocketsphinx-ruby instead?

[MrChoclate]

This repository seems to be ahead. What is wrong with it?

[oneguy]

I'm having the same issue. The craziest part is that it's Feb 13, 2018. Exactly one year.

any luck fixing the issue?