Research

The Nine Billion Names of God

Research

In the beginning was the file.

Getting Started

All of this is best done on a linux machine.

Downloading a video file from any online service can be done with youtube-dl. Youtube-dl works with Youtube, Vimeo as well as a lot of other online video platforms including all the popular adult video ones.

Splitting the video file into video and audio is done through ffmpeg.

Commands:
$ youtube-dl https://video.com/abc/ -o abc.mp4
$ ffmpeg -i abc.mp4 -an -c copy abc-video.mp4
$ ffmpeg -i abc.mp4 abc-audio.mp3

Audiogrep

Audiogrep uses CMU PocketSphinx for speech recognition. Sphinx is based on an older heuristic approach to voice recognition and the results were quite poor. I was able to recognize a word in audio files only probably 1 in 10-20 times. The word it recognized was "shit" from George Carlin's video. But It was unable to recognize "god" from some 10 other audio inputs.

Deep Learning

Services such as Google Speech and Youtube's auto-captioning service use deep learning neural networks to handle speech to text conversion. Google has open sourced its deep learning library called tensorflow and projects such as this are attempting to develop open models for speech recognition. For although the software libraries are open, the efficiency of the speech recognition depends on the neural network models. A model is a particular configuration of neural network 'black
boxes' that can be trained to perform a task. In this case speech recognition.

We found another open source project that had developed a model and trained it based on an open VCTK corpus. We downloaded the code and ran it, training it on the same VCTK corpus. But the neural network seems to perform well only on the audio samples from the VCTK training data set. When fed a random audio sample it's output is garbage, even worse than the accuracy of audiogrep mentioned earlier.

Other Approaches

There does not appear to be a good open source deep learning model for speech recognition. But there appear to be a number of different approaches and attempts. There are a number of papers being written on deep-learning and speech recognition. We can try and implement these models, but the big time spent in neural networks is training. Training the VCTK corpus above took 4 days. Training each model will likely take a similar amount of time.

Another option might be to try some kind of heuristic approach to recognizing just this one word alone rather than a generic speech recognition system. I'm not sure what this entails.

Yet another option is to create a sufficiently large training data set to train the deep learning neural network to just recognize this one word.

 

Audiogrep

Steps:
1. Download the video clip
$ youtube-dl https://www.youtube.com/watch?v=kyBH5oNQOS0



2. Extract the audio file
$ ffmpeg -i seven.mp4 seven.mp3

3. Install audiogrep
$ sudo pip install audiogrep

pip is the python package manager. Instructions on installing it along
with python can be found here -
https://packaging.python.org/installing/

4. Transcribe the audio file
$ audiogrep --input seven.mp3 --transcribe

This is the audio file:

This is the output :

at got out there in our own
<s> 4.070 4.090 0.999300
at 4.100 4.820 0.006280
got 4.830 5.140 0.034559
out 5.150 5.200 0.020001
there 5.210 5.390 0.182301
in 5.400 5.500 0.132545
our(3) 5.510 5.850 0.061469
own 5.860 6.360 0.147457
</s> 6.370 6.510 1.000000
shit
<s> 6.700 6.720 0.999700
shit 6.730 7.260 0.051409
<sil> 7.270 7.370 0.699534
</s> 7.380 7.660 1.000000
as an artist and a bit actually
<s> 7.970 7.990 0.998301
as(2) 8.000 8.230 0.043917
an(2) 8.240 8.330 0.003088
artist(2) 8.340 9.030 0.003956
and 9.040 9.460 0.083686
[SPEECH] 9.470 9.490 0.031024
a 9.500 9.620 0.036143
bit 9.630 9.860 0.002704
actually(2) 9.870 10.250 0.048779
</s> 10.260 10.680 1.000000
this
<s> 11.170 11.190 0.999900
this 11.200 11.660 0.482013
</s> 11.670 12.040 1.000000
that it is like a racehorse
<s> 12.460 12.520 0.999200
that 12.530 12.980 0.982847
it 12.990 13.220 0.485691
is 13.230 13.490 0.246927
<sil> 13.500 13.590 0.731516
like 13.600 13.780 0.409056
a 13.790 13.860 0.175678
racehorse 13.870 14.640 0.077297
</s> 14.650 15.000 1.000000
bach
<s> 15.480 15.510 0.999400
bach 15.520 15.880 0.001699
</s> 15.890 16.310 1.000000
are you
<s> 16.890 16.910 0.999700
are 16.920 17.120 0.021652
<sil> 17.130 17.160 0.448883
you 17.170 17.470 0.544509
</s> 17.480 17.870 1.000000
ah
<s> 18.800 18.880 1.000600
ah 18.890 19.170 0.049546
</s> 19.180 19.480 1.000000
he has a beautiful caught
<s> 19.650 19.670 1.000100
he 19.680 19.800 0.316848
has(2) 19.810 19.990 0.180866
a 20.000 20.050 0.609292
beautiful 20.060 20.530 1.000100
caught 20.540 21.030 0.056178
</s> 21.040 21.480 1.000000
huh
<s> 22.370 22.560 0.998801
huh 22.570 23.060 0.019486
</s> 23.070 23.430 1.000000
go to their own car accident
<s> 23.730 23.750 0.999300
go 23.760 23.880 0.045368
to 23.890 24.090 0.411806
their 24.100 24.340 0.033240
own 24.350 24.480 0.122610
<sil> 24.490 24.830 0.997403
car 24.840 25.010 0.037682
accident 25.020 25.560 0.048231
</s> 25.570 25.980 1.000000
no doctor
<s> 26.920 26.940 0.999200
no 26.950 27.160 0.031161
doctor(2) 27.170 27.610 0.007543
</s> 27.620 28.020 1.000000
you
<s> 28.160 28.180 0.999900
you 28.190 28.800 0.798815
</s> 28.810 28.890 1.000000
the relevant topic
<s> 29.220 29.240 0.998900
the 29.250 29.330 0.152618
relevant 29.340 29.710 0.001479
topic 29.720 30.050 0.001058
</s> 30.060 30.430 1.000000
dead
<s> 31.370 31.390 0.999400
dead 31.400 32.010 0.006404
</s> 32.020 32.370 1.000000
hey
<s> 32.510 32.530 0.999800
hey 32.540 33.120 0.203013
</s> 33.130 33.220 1.000000
it then
<s> 33.410 33.480 0.999500
it 33.490 33.800 0.805715
then 33.810 34.620 0.498885
</s> 34.630 34.800 1.000000
it in
<s> 36.190 36.210 0.999400
it 36.220 36.410 0.487637
in 36.420 36.800 0.455987
<sil> 36.810 36.830 0.395697
</s> 36.840 37.210 1.000000
and to
<s> 37.750 37.780 0.999500
and 37.790 37.900 0.365307
to(3) 37.910 38.480 0.238839
<sil> 38.490 38.510 0.238219
</s> 38.520 38.860 1.000000
the in the
<s> 39.640 39.660 0.999400
the(2) 39.670 39.940 0.207177
in 39.950 40.250 0.226803
<sil> 40.260 40.890 0.999500
the(2) 40.900 41.360 0.997603
<sil> 41.370 41.540 0.988565
</s> 41.550 42.300 1.000000

Deep Learning

Steps
1. Install git and python for your platform.

2. Clone the repository
$ git clone https://github.com/itzikgili/speech-to-text-wavenet
(Change directory to the new cloned directory - speech-to-text-wavenet)

3. Install pip (see above) and the required libraries
$ pip install -r requirements.txt

4. Create the required asset/data directory in the main directory of the project. speech-to-text-wavenet/asset/data
$ mkdir -p asset/data

5. Download the VCTK corpus files.
$ wget http://homepages.inf.ed.ac.uk/jyamagis/release/VCTK-Corpus.tar.gz -O
asset/data/VCTK-Corpus.tar.gz

6. Unzip the Corpus
$ cd asset/data && tar VCTK-Corpus.tar.gz && cd ../..

7. Train the neural network.
$ python train.py

Wait (it took me 4 days to finish training the network on a fast dedicated server)

8. Transcribe file
$ python recognize.py --file seven.mp3

The output of this transcription is complete garbage, compared to the output from audiogrep. But when used against audio files in the VCTK training corpus the transcription seems reasonably accurate.

$ python recognize.py --file asset/data/wav48/p225/p225_003.wav

Gonzo approach to Machine Learning - Part 1

I downloaded a test video from https://www.pornhub.com/ using the search phrase "Oh god"


I downloaded this video - http://www.pornhub.com/view_video.php?viewkey=1631870315 that contained the phrase multiple times and split the audio and video. The process for this has already been mentioned earlier (youtube-dl & ffmpeg).

Attempt #1 - Using Youtube's autocaptioning feature 

Combine the audio file above with a static image.To circumvent the Youtube guidelines against pornographic images. [This approach looks promising]

$ ffmpeg -loop 1 -i staticimage.jpg -i ohgod.mp3 -c:v libx264 -tune stillimage -c:a aac -strict experimental -b:a 64k -shortest ohgod.flv

Upload the video to youtube - https://www.youtube.com/upload


Once uploaded visit - https://www.youtube.com/my_videos locate the video and click Edit. Under the 'Subtitles/CC' tab you should see the auto-transcription as shown below:

Unfortunately for this video no such caption appeared. Reading the support FAQ for auto captioning service for Youtube, I found some suggestions stating that it would work only if there was not much background noise, and the speech began within 12 seconds of the video. I tried to scrub the noise using Audacity, and also clipped the bits of silence in the beginning. But none of these were auto captioned by Youtube.

Attempt #2 - Deepgram

A paid service with a free tier that performs transcription of audio files.

Signed up for the service then uploaded the ohgod.mp3.

Transcription result:

00:00:00

Big moment oh no [noise] [noise] you you you see a oh [noise]. Uh i i you know [noise] really really. That iraq or [noise] you yeah [noise] oh cool so and and.

00:01:45

You know u._s. us oh oh girl didn't. Really [noise] [noise] [noise] [noise] great [noise] who. Are [noise] are yeah [noise] oh oh oh. Oh oh to go for a for a year [noise] [noise] earl world and.

00:03:14

[noise] maybe [noise] [noise] oh as a as a um [noise] uh. I'm a [noise] a [noise] a. A oh as a i know and um and. Are are are um are um.

00:04:18

[noise] and er er oh oh. Oh [noise] and money yeah but oh. Oh oh yeah she uh oh it is oh go ahead our . Our our um there [noise] a food or or.

00:04:54

Your um your yeah i'm i'm like oh no oh no.

Notes:

1. use audiogrep for transcription and download say the first 100 videos using the search phrase "oh god" and see how many audiogrep is able to detect with the phrase.

2. the transcription for this test video seems to accurately detect "oh" but not "god". maybe we can use some kind of heuristics to flag videos containing a certain number of "oh"s

Using AudioGrep Again

here is the code from the latest experiment https://github.com/ninebillionnamesofgod/godiogrep 

about 60 files were downloaded and the transcription ran and found only one instance of "god". This is about 2GB of video.

The transcription is here:

ah how it ha ha
<s> 0.080 0.100 0.999900
ah 0.110 0.400 0.116048
how 0.410 0.640 0.149581
it 0.650 0.910 0.425326
<sil> 0.920 0.970 0.571172
ha 0.980 1.630 0.425581
<sil> 1.640 2.080 0.999900
ha 2.090 3.210 0.997004
</s> 3.220 3.260 1.000000
and it has to it don't get then it hit him alone
<s> 3.410 3.430 0.998002
and 3.440 4.160 0.102186
it 4.170 4.360 0.091385
has(2) 4.370 4.590 0.021798
to 4.600 5.590 1.000300
<sil> 5.600 5.620 0.772109
<sil> 5.630 5.720 1.000300
<sil> 5.730 5.760 0.749737
<sil> 5.770 5.790 0.556289
[SPEECH] 5.800 6.000 0.914013
it 6.010 6.460 0.983634
don't(2) 6.470 6.730 0.613695
<sil> 6.740 6.990 0.880897
get(2) 7.000 7.820 1.000000
then 7.830 8.780 0.700234
<sil> 8.790 8.930 0.851446
it 8.940 9.340 0.987181
hit 9.350 9.650 0.001968
him 9.660 9.900 0.010872
alone 9.910 11.200 1.000000
well and it it's a good loan hard at the top but it had lol like your eyes
<s> 11.270 11.290 0.999300
well 11.300 12.280 0.691119
<sil> 12.290 12.310 0.482881
and 12.320 13.340 0.999300
<sil> 13.350 13.590 0.997702
<sil> 13.600 13.620 0.411600
<sil> 13.630 13.650 0.371423
<sil> 13.660 13.680 0.421599
<sil> 13.690 13.760 0.677772
it 13.770 15.660 0.995211
<sil> 15.670 15.910 0.996904
it's 15.920 16.020 0.014744
a(2) 16.030 16.160 0.122414
<sil> 16.170 16.480 0.990148
good 16.490 16.880 0.020876
loan 16.890 17.640 0.009441
<sil> 17.650 18.680 0.999700
hard 18.690 20.630 0.139411
at 20.640 20.870 0.227758
<sil> 20.880 20.960 0.655178
<sil> 20.970 21.470 0.999700
the 21.480 21.530 0.071633
<sil> 21.540 22.110 0.998301
top(2) 22.120 22.410 0.087180
<sil> 22.420 22.580 0.908363
but 22.590 24.260 0.999400
it 24.270 25.190 0.078387
had 25.200 25.580 0.075245
lol 25.590 26.050 0.054300
like 26.060 26.390 0.292543
[SPEECH] 26.400 26.550 0.593059
your(2) 26.560 26.810 0.524938
eyes 26.820 27.490 0.940440
[SPEECH] 27.500 27.660 0.569347
<sil> 27.670 28.420 0.999000
</s> 28.430 28.910 1.000000
and now and the boeing plant the candidate and not only i'd known at out crack open
<s> 29.120 29.150 0.997303
and 29.160 29.710 0.961939
now 29.720 30.040 0.302692
[SPEECH] 30.050 30.230 0.724889
and 30.240 30.600 0.699534
<sil> 30.610 30.960 0.979315
the 30.970 31.280 0.276858
<sil> 31.290 31.640 0.999000
boeing 31.650 33.360 0.780884
<sil> 33.370 33.510 0.779713
plant 33.520 34.220 0.575875
the(2) 34.230 34.530 0.144305
<sil> 34.540 34.570 0.114675
candidate(2) 34.580 35.600 0.092915
and 35.610 36.230 0.484526
not 36.240 36.630 0.484429
only 36.640 37.050 0.169482
i'd 37.060 37.290 0.004449
known 37.300 38.120 0.010176
at 38.130 38.760 0.003394
out 38.770 39.530 0.208528
crack 39.540 40.200 0.041888
<sil> 40.210 42.590 0.999400
open 42.600 43.280 0.852128
</s> 43.290 44.220 1.000000
oh well the day and then in at noon and good i'm not ruling but i didn't whoa oh my god
<s> 44.320 44.340 0.997004
oh 44.350 44.770 0.959153
well 44.780 45.870 0.560982
the 45.880 46.070 0.173322
<sil> 46.080 46.130 0.168907
day 46.140 46.600 0.077104
and(2) 46.610 46.830 0.069162
then 46.840 47.280 0.032615
in 47.290 47.570 0.003361
at 47.580 47.750 0.009015
noon 47.760 48.130 0.000247
<sil> 48.140 48.320 0.961073
and(2) 48.330 48.900 0.795864
<sil> 48.910 48.940 0.478267
good 48.950 50.840 0.262408
i'm 50.850 51.120 0.120507
<sil> 51.130 52.240 0.998801
not 52.250 52.420 0.122292
ruling 52.430 53.060 0.101971
but 53.070 53.360 0.069857
i 53.370 53.880 0.584463
didn't(2) 53.890 54.730 0.170315
whoa 54.740 57.010 0.999100
<sil> 57.020 57.880 0.999300
oh 57.890 58.200 0.176153
my 58.210 58.660 0.524728
<sil> 58.670 58.860 0.575991
god 58.870 59.430 0.744134
</s> 59.440 59.680 1.000000

Here is the AudioFile:

We found False Gods

the audio transcription had not yet finished for all the videos, it found one more instance of "god". but on examining the video it appears to be a false signal. the woman in the video is speaking in Russian and one of the phrases has gotten translated to "god".

so final count, videos downloaded: 58

false gods found: 1

Here is the AudioFile:

 

Here is the transcript for the Russian God:

and we were we were ah i need
<s> 0.080 0.100 0.999800
and 0.110 0.660 0.050821
we 0.670 0.920 0.068302
were 0.930 1.120 0.065114
we 1.130 1.330 0.184133
were 1.340 1.430 0.025547
<sil> 1.440 1.460 0.331632
ah 1.470 1.830 0.651453
i 1.840 2.050 0.528203
need 2.060 3.710 1.000000
mm to be good you are the good in a word were it won't ruin like what what in the or the or at lunch line of them were he at the top of that the you know what a go i i chalk good but lower chalet on a golf thing she eat yeah the the optimal out what we're going to do move god of course the one i'm in good in and i'm it would and do anything about it
<s> 4.990 5.010 0.983535
mm 5.020 5.190 0.017286
<sil> 5.200 5.420 0.979706
<sil> 5.430 5.820 0.976185
to(3) 5.830 5.950 0.070622
<sil> 5.960 6.150 0.955993
be 6.160 6.470 0.493772
<sil> 6.480 6.790 0.973650
good(2) 6.800 7.780 0.978727
<sil> 7.790 7.820 0.323800
<sil> 7.830 7.850 0.264199
<sil> 7.860 7.880 0.634161
<sil> 7.890 8.700 0.984322
you 8.710 8.960 0.186954
<sil> 8.970 9.070 0.899594
are(2) 9.080 9.280 0.029573
<sil> 9.290 9.390 0.885136
the 9.400 9.750 0.866913
<sil> 9.760 9.880 0.976478
good 9.890 10.200 0.063208
<sil> 10.210 10.580 0.977553
in 10.590 10.840 0.114480
<sil> 10.850 10.940 0.909090
a 10.950 11.140 0.099046
<sil> 11.150 11.760 0.978727
word 11.770 14.180 0.987280
were 14.190 14.240 0.987280
it 14.250 14.360 0.987181
won't 14.370 14.680 0.456443
ruin 14.690 14.960 0.010475
<sil> 14.970 15.040 0.899504
like 15.050 15.330 0.015020
<sil> 15.340 15.510 0.699464
what 15.520 15.880 0.068432
what 15.890 16.010 0.034826
<sil> 16.020 16.180 0.884605
in 16.190 16.440 0.003931
<sil> 16.450 16.520 0.846183
the 16.530 16.760 0.138038
<sil> 16.770 16.880 0.383502
or(2) 16.890 17.120 0.091504
the 17.130 17.440 0.428143
or(2) 17.450 17.530 0.147118
at 17.540 18.020 0.915752
<sil> 18.030 18.370 0.951795
lunch 18.380 18.730 0.200732
line 18.740 18.970 0.086450
of 18.980 19.110 0.030467
them(2) 19.120 19.430 0.230301
were 19.440 19.710 0.413126
[SPEECH] 19.720 19.740 0.510034
<sil> 19.750 19.930 0.899504
he 19.940 20.370 0.119332
<sil> 20.380 20.440 0.187010
at 20.450 20.870 0.006599
the(2) 20.880 20.970 0.016398
top(2) 20.980 21.140 0.006073
of 21.150 21.430 0.667613
<sil> 21.440 21.810 0.812431
that(2) 21.820 22.020 0.151826
<sil> 22.030 22.190 0.909545
the(2) 22.200 22.480 0.868562
<sil> 22.490 22.560 0.771183
you 22.570 22.680 0.036099
<sil> 22.690 22.760 0.780806
know 22.770 23.020 0.809349
what 23.030 23.250 0.609353
a(2) 23.260 23.560 0.885401
go 23.570 24.030 0.018150
i 24.040 24.060 0.007262
i 24.070 24.190 0.085112
chalk(2) 24.200 24.680 0.089468
<sil> 24.690 24.840 0.964154
good(2) 24.850 25.230 0.354864
<sil> 25.240 25.730 0.862071
but 25.740 25.920 0.183912
lower 25.930 26.150 0.004546
chalet 26.160 26.490 0.010230
<sil> 26.500 26.590 0.584931
on(2) 26.600 26.750 0.478267
a(2) 26.760 27.220 0.883721
golf(2) 27.230 27.500 0.346136
thing 27.510 27.680 0.189552
she 27.690 28.440 0.997702
<sil> 28.450 28.880 0.996705
<sil> 28.890 29.110 0.994713
<sil> 29.120 29.490 0.997503
eat 29.500 29.990 0.168182
yeah 30.000 30.480 0.946195
the 30.490 30.650 0.067684
the 30.660 30.730 0.037554
optimal 30.740 31.490 0.021635
out 31.500 31.870 0.009128
what(2) 31.880 32.370 0.321508
we're(3) 32.380 33.000 0.022245
<sil> 33.010 33.170 0.635176
going(2) 33.180 33.610 0.090947
<sil> 33.620 34.220 0.998900
to 34.230 34.310 0.400434
do 34.320 34.780 0.958769
<sil> 34.790 34.910 0.910819
move 34.920 35.520 0.102441
<sil> 35.530 35.660 0.926901
god 35.670 36.080 0.183435
of 36.090 36.220 0.209176
course 36.230 36.420 0.210624
the(2) 36.430 37.010 0.160059
<sil> 37.020 37.150 0.996008
one 37.160 37.370 0.013268
<sil> 37.380 37.510 0.781431
i'm(2) 37.520 38.420 0.045741
<sil> 38.430 38.680 0.978433
in 38.690 38.870 0.090675
<sil> 38.880 39.000 0.946953
good 39.010 40.190 0.999400
[SPEECH] 40.200 40.320 0.822076
<sil> 40.330 40.520 0.935375
in 40.530 40.810 0.513155
<sil> 40.820 40.910 0.893586
and 40.920 41.590 0.488369
<sil> 41.600 41.730 0.980000
i'm(2) 41.740 42.050 0.926437
<sil> 42.060 42.420 0.999800
it 42.430 42.840 0.004572
would 42.850 42.980 0.007453
<sil> 42.990 43.080 0.980294
and(2) 43.090 43.480 0.041875
[SPEECH] 43.490 43.770 0.606132
<sil> 43.780 43.920 0.629107
<sil> 43.930 44.340 0.995410
do 44.350 44.570 0.028004
<sil> 44.580 44.600 0.441978
anything 44.610 45.000 0.351825
about 45.010 45.640 0.454985
it 45.650 45.790 0.199311
[SPEECH] 45.800 45.960 0.715883
</s> 45.970 46.120 1.000000
when you been in women who are you in what might be the ah i thought it could be good and that i got caught and you got that much what what what in the i'm glad the mightiest to rachel
<s> 46.210 46.350 0.990049
<sil> 46.360 46.480 0.840196
when(3) 46.490 46.720 0.028243
<sil> 46.730 46.860 0.926530
you 46.870 47.180 0.308037
been(2) 47.190 47.650 0.085719
<sil> 47.660 48.020 0.756591
in 48.030 48.250 0.060047
<sil> 48.260 48.350 0.967825
women 48.360 48.590 0.001302
<sil> 48.600 48.720 0.947142
who 48.730 48.880 0.531223
<sil> 48.890 49.030 0.694515
<sil> 49.040 49.430 0.962612
are(2) 49.440 49.460 0.057244
<sil> 49.470 49.560 0.625343
you 49.570 49.730 0.317800
<sil> 49.740 49.890 0.950748
in 49.900 50.120 0.033303
<sil> 50.130 50.360 0.926901
<sil> 50.370 50.620 0.973845
what(2) 50.630 51.080 0.019696
<sil> 51.090 51.270 0.875364
might 51.280 51.420 0.010140
<sil> 51.430 51.740 0.993024
be 51.750 51.850 0.216151
<sil> 51.860 52.110 0.992726
the 52.120 52.220 0.069606
<sil> 52.230 52.360 0.538175
<sil> 52.370 52.410 0.864143
ah 52.420 53.160 0.183251
i 53.170 53.270 0.069738
<sil> 53.280 53.740 0.993620
thought 53.750 54.200 0.048444
<sil> 54.210 54.590 0.992130
it 54.600 54.690 0.132121
<sil> 54.700 55.010 0.985504
could 55.020 55.140 0.077004
<sil> 55.150 55.230 0.705082
<sil> 55.240 55.490 0.992626
be 55.500 55.660 0.671631
<sil> 55.670 55.850 0.985603
good 55.860 56.120 0.090756
<sil> 56.130 56.330 0.956758
and 56.340 56.600 0.023920
that 56.610 56.880 0.036895
i 56.890 57.040 0.324643
got 57.050 57.210 0.114297
caught(2) 57.220 57.990 0.022334
<sil> 58.000 58.110 0.666212
and 58.120 58.460 0.162202
<sil> 58.470 58.630 0.877643
<sil> 58.640 58.780 0.894301
you 58.790 59.190 0.244128
<sil> 59.200 59.220 0.102605
got 59.230 59.370 0.076337
that(2) 59.380 59.750 0.448390
much 59.760 59.970 0.125225
what(2) 59.980 60.340 0.023285
what(2) 60.350 60.550 0.101656
what(2) 60.560 60.830 0.040739
<sil> 60.840 61.140 0.998102
in 61.150 61.250 0.002065
<sil> 61.260 61.560 0.993520
the 61.570 61.890 0.977455
i'm(2) 61.900 62.030 0.034784
glad 62.040 62.310 0.050275
the 62.320 62.390 0.080147
mightiest 62.400 62.930 0.020059
to(3) 62.940 63.160 0.309457
<sil> 63.170 63.190 0.115747
rachel 63.200 63.760 0.003979
<sil> 63.770 63.910 0.638041
</s> 63.920 64.320 1.000000
hello what should she be a good ish could be be be a hobby and we ha e two the time we get it runs to ha ha beep beep
<s> 64.800 64.820 0.995012
hello 64.830 65.150 0.035427
what(2) 65.160 65.420 0.248637
should 65.430 65.670 0.019820
she 65.680 66.430 0.968600
<sil> 66.440 66.730 0.996107
be 66.740 66.860 0.033058
<sil> 66.870 66.940 0.611918
a(2) 66.950 67.030 0.168958
good 67.040 67.230 0.296312
ish 67.240 67.420 0.001740
could 67.430 67.790 0.019572
<sil> 67.800 68.130 0.996506
be 68.140 68.280 0.634732
<sil> 68.290 68.610 0.993222
be 68.620 68.750 0.010204
<sil> 68.760 69.070 0.996406
be 69.080 69.250 0.002951
<sil> 69.260 69.360 0.975405
a 69.370 69.540 0.118642
hobby 69.550 70.130 0.004713
<sil> 70.140 70.490 0.997702
and 70.500 70.770 0.122684
<sil> 70.780 70.840 0.839104
we 70.850 71.080 0.154770
<sil> 71.090 71.510 0.998102
ha 71.520 72.200 0.005741
<sil> 72.210 72.380 0.794830
e 72.390 72.470 0.002329
two 72.480 73.140 0.996008
[SPEECH] 73.150 73.170 0.582129
<sil> 73.180 73.300 0.989653
the(2) 73.310 73.450 0.059724
<sil> 73.460 73.920 0.991039
time 73.930 74.410 0.039432
<sil> 74.420 74.580 0.959825
we 74.590 74.840 0.105161
<sil> 74.850 75.300 0.997104
<sil> 75.310 75.530 0.996904
<sil> 75.540 75.620 0.746445
get(2) 75.630 76.000 0.055959
<sil> 76.010 76.110 0.599440
it 76.120 76.550 0.625594
runs 76.560 76.730 0.008810
to 76.740 76.890 0.034174
<sil> 76.900 77.030 0.984224
ha 77.040 77.430 0.002040
ha 77.440 77.840 0.929407
<sil> 77.850 77.970 0.500334
beep 77.980 78.410 0.000092
beep 78.420 78.610 0.391211
</s> 78.620 79.030 1.000000
your net worth the end will go meet a half
<s> 79.380 79.430 0.998102
your 79.440 80.210 0.168485
net 80.220 80.390 0.015491
worth 80.400 81.040 0.316278
<sil> 81.050 81.430 0.998301
the 81.440 82.310 0.999000
<sil> 82.320 82.340 0.942700
end 82.350 82.530 0.034075
will(2) 82.540 82.940 0.998401
go 82.950 83.180 0.961939
meet 83.190 83.570 0.022849
a 83.580 83.630 0.111989
half 83.640 83.780 0.023475
<sil> 83.790 84.070 0.965119
</s> 84.080 84.220 1.000000
hi nice adequate what what what word but most news doesn't work in haiti who it the things that could not how it it could do that i didn't hm
<s> 84.860 84.880 0.997603
hi 84.890 85.250 0.255928
nice 85.260 85.720 0.087678
adequate 85.730 86.630 0.001194
what(2) 86.640 87.110 0.048458
what 87.120 87.330 0.042922
what 87.340 87.680 0.029608
word 87.690 88.470 0.000780
but 88.480 88.650 0.010616
most(2) 88.660 88.910 0.007556
news 88.920 89.140 0.004289
doesn't(2) 89.150 89.550 0.048512
work 89.560 89.780 0.286005
in 89.790 89.940 0.280875
haiti 89.950 90.570 0.021056
who 90.580 90.770 0.010317
it 90.780 91.290 0.888417
[NOISE] 91.300 91.320 0.034725
the 91.330 91.380 0.003794
things 91.390 92.260 0.018441
that 92.270 92.350 0.040273
could 92.360 92.580 0.004982
not 92.590 92.840 0.135250
how 92.850 93.170 0.264463
<sil> 93.180 93.200 0.167528
it 93.210 93.830 0.222824
<sil> 93.840 93.890 0.296253
<sil> 93.900 93.990 0.509881
<sil> 94.000 94.420 0.895644
it 94.430 97.320 0.815035
could 97.330 98.980 0.706777
do 98.990 99.110 0.757272
<sil> 99.120 99.300 0.696671
that(2) 99.310 99.480 0.249359
i 99.490 99.700 0.808217
<sil> 99.710 100.050 0.947995
didn't(3) 100.060 100.340 0.113002
hm 100.350 100.780 0.043043
</s> 100.790 100.810 1.000000
it an egg that are that when done and not to read it when it will all become to think what i thought i am
<s> 101.130 101.390 0.993818
[SPEECH] 101.400 101.730 0.983732
<sil> 101.740 102.060 0.955993
<sil> 102.070 102.210 0.596987
it 102.220 102.990 0.993818
an(2) 103.000 103.290 0.161393
<sil> 103.300 103.450 0.683900
egg 103.460 104.540 0.450322
<sil> 104.550 104.790 0.846353
that(2) 104.800 105.300 0.728523
[NOISE] 105.310 105.400 0.686435
are(2) 105.410 105.520 0.173444
<sil> 105.530 105.610 0.893407
that 105.620 106.740 0.850425
when(3) 106.750 107.160 0.359436
[SPEECH] 107.170 107.190 0.653934
done 107.200 107.810 0.655440
<sil> 107.820 108.120 0.995808
and 108.130 109.190 0.058823
<sil> 109.200 109.380 0.782839
not 109.390 110.350 0.943738
<sil> 110.360 110.640 0.891354
<sil> 110.650 110.910 0.835671
[SPEECH] 110.920 111.470 0.971899
<sil> 111.480 111.570 0.498436
<sil> 111.580 111.780 0.984913
<sil> 111.790 111.910 0.954559
to 111.920 112.110 0.078426
<sil> 112.120 112.290 0.947900
read(2) 112.300 113.120 0.972580
it 113.130 113.190 0.232429
when(3) 113.200 113.330 0.004951
it 113.340 113.530 0.295336
will(2) 113.540 114.250 0.997104
all 114.260 114.760 0.462972
<sil> 114.770 115.020 0.894748
[SPEECH] 115.030 115.300 0.864748
become 115.310 115.920 0.002191
to(3) 115.930 116.150 0.812919
think 116.160 116.360 0.785270
<sil> 116.370 116.450 0.745177
what(2) 116.460 116.660 0.176683
i 116.670 117.010 0.579805
thought 117.020 117.410 0.338400
i 117.420 117.570 0.394353
<sil> 117.580 117.650 0.517277
am(2) 117.660 117.960 0.651909
<sil> 117.970 118.650 0.998900
<sil> 118.660 118.780 0.733421
</s> 118.790 119.050 1.000000
only as sure as a while to learn that that got i got i got greedy and they had gone hand in the news a lot
<s> 119.500 119.520 0.995211
only 119.530 119.990 0.737171
as(2) 120.000 120.330 0.014199
sure 120.340 120.560 0.187028
as(2) 120.570 120.920 0.124252
a 120.930 120.990 0.098041
while(2) 121.000 121.410 0.089262
to(2) 121.420 121.610 0.585867
learn 121.620 121.920 0.030980
<sil> 121.930 122.140 0.969763
[SPEECH] 122.150 122.390 0.587628
that(2) 122.400 122.580 0.489886
that 122.590 122.950 0.580676
got 122.960 123.240 0.039884
i 123.250 123.400 0.269267
got 123.410 123.700 0.351755
i 123.710 123.970 0.249210
[SPEECH] 123.980 124.260 0.979510
<sil> 124.270 124.480 0.925789
<sil> 124.490 124.860 0.871782
got 124.870 125.050 0.336847
greedy 125.060 125.370 0.009557
and 125.380 125.610 0.021025
they 125.620 125.750 0.093999
had 125.760 127.010 0.868562
gone 127.020 127.250 0.404136
hand 127.260 127.900 0.056183
in 127.910 128.120 0.062236
the 128.130 128.400 0.558854
news 128.410 128.680 0.447359
a(2) 128.690 129.010 0.868215
<sil> 129.020 129.880 1.000100
lot 129.890 130.070 0.412631
</s> 130.080 130.160 1.000000