DL C5W3 subtitle errata

I noticed some mistranscriptions in the transcript (subtitles, captions) for the course videos. Below is the errata for the 3rd week.

The format for the errata is as follows. The approximate time is in braces, then the mistranscribed word or phrase follows, followed by a hyphen, then followed by the correct word or phrase. Like this:

[time] mistranscribed word or phrase - correct word or phrase

video 1 - basic models
in this video “encoder” and “encoding” are often mixed up, along with decoder and decoding, but it is a small issue.
[0:15] beam surge - beam search
[2:16] taken - token
[2:25] Knicks - next?
[2:30] techs - text
[3:29] face - phrase
[3:33] calls - course
[3:34] confidence - convnets
[3:57] free trade - pre-trained
[5:40] image capturing - image captioning

video 2 - picking the most likely sequence
[1:20] coded - encoder
[1:21] decoded - decoder
[2:34] censors - sentence

video 3 - beam search
[0:49] operate - output
[1:05] coalition - encoder
[1:05] decoalition - decoder
[1:08] for a square - first word
[1:37] with the three - equal to three
[1:38] cause that - consider
[2:09] said - set
[3:22] new network - neural network
[4:04] this time were fragment - this network fragment
[4:18] we also need help out - we ultimately care about
[4:19] second step would be assertions - second step of beam search
[5:32] your network - neural network
[6:23] although words - all the words
[7:43] being surge - beam search
[11:20] being searched - beam search
[11:49] from our partners - for refinements

video 4 - refinements to beam search
[2:57] obviously objective - original objective
[3:07] loss - lots
[4:32] putting - outputting
[5:09] hydrofracking - hyperparmeter
[5:14] continued - can tune
[5:28] it the worth one practice - it works well in practice
[6:00] I’ll put - output
[6:04] be with a - beam width of
[6:21] I’ll put - output
[6:44] to the cheese - that achieves
[6:56] all put - output
[7:26] consuming - considering
[8:17] systems - research systems
[8:21] publish people - publish a paper with
[8:53] basically research - basically greedy search[?]
[8:58] thousands of thousand - thousand to three thousand
[9:11] social programs - search algorithms
[9:12] like PAFs Breakfast or DFS DEFA search - like BFS (breadth first search) or DFS (depth first search)
[9:27] But even further proof of search with emphasis, then on those algorithms, - But if you’ve heard of breadth first search or depth first search, unlike those algorithms,
[9:40] augments - arg max
[9:44] the of social deficits, don’t about outside - breadth first search or depth first search, don’t worry about it
[10:03] cause into sequence of causes - course in the sequence of courses

video - Bleu score
[2:51] depth set - dev set

video - attention model intuition
[0:30] in your network - neural network
[2:59] Yoshe Bengio - Yoshua Bengio
[3:30] R and N - RNN
[3:46] R and N - RNN
[3:54] R and N - RNN
[4:20] R and N - RNN
[4:24] R and N - RNN
[4:37] R and N - RNN
[5:55] work - word
[6:12] R and N - RNN
[6:22] R and N - RNN
[6:55] the free - l’Afrique
[8:15] R and N - RNN
[8:27] R and N - RNN
[8:55] R and N - RNN

video - attention model
[1:20] factor - vector
[3:07] way to some - weighted sum
[3:14] waited - weighted
[3:15] waits - weights
[3:19] waits - weights
[3:45] waited - weighted
[3:58] waits - weights
[4:51] waits - weights
[5:45] ways - weights
[5:53] waits - weights
[6:21] soft pass - softmax
[6:24] waits - weights
[7:07] that just fell - that’s fed
[7:46] hidden stages - hidden state
[7:49] upper - output
[8:15] that obligation - back propagation
[8:16] wait and descent - gradient descent
[9:47] up in - I’ve been
[9:57] capturing - captioning
[10:14] Andrew Benjo - Yoshua Bengio
[10:37] prior - programming
[11:10] prior - programming
[11:26] waits - weights
[11:39] waits - weights
[12:08] propagation - back propagation

video - speech recognition
[0:02] were - with
[1:56] false back outputs - filter back outputs
[2:50] the Greek language - to break language
[3:52] voice - hours
[4:10] were - where
[5:05] new network - neural network
[5:19] LSP - LSTM
[5:21] GIU - GRU
[7:59] upwards - output
[8:06] by those - Baidu’s
[8:29] where production skills - or production scale

video - trigger word detection
[1:51] Why? - Y.

Thanks for reporting them, @Thomas_A_W. I am going to share this thread with the course team.

Raymond