8 thoughts on “Thoughts On What’s Happening In Space”
I hate podcasts. It’s usually much faster to just read a transcript.
Me, too, but for those who don’t mind…
That’s what I told him when he asked me. I said I don’t mind doing podcasts, but I hate listening to them.
Maybe he could transcribe it. There ought to be software that can do that these days, with voice recognition, though it would no doubt need a lot of editing to clean it up.
Interesting, thanks. Particularly the historical perspective on the great American observatories.
Youtube has a built-in transcription service. here’s one way to use it to convert audio to text (quality probably not very good)
I tried that for our podcast (shameless plug: talkuniverse.org). It actually works really well, getting only a few words wrong. The only problem is no punctuation. So, I had to go through it and do all the punctuation. Still, it’s faster than trying to transcribe the whole thing. We’ve got over 40 episodes in the can right now, so it’s going to take me about six weeks or so to get them all on YouTube and transcribed, but this is a great way to reach the deaf and those who simply don’t like listening to podcasts. Thanks for the tip.
After transcribing two shows, I can report a few things:
First, YouTube’s auto-transcription feature is very good, but not perfect. But hey, it’s free, so what can you expect? The accuracy is up around 95% to 99%, depending on how fast the person talks. The utterance “uh” normally doesn’t get transcribed, but that means that words like “a” and “the” are often missing as well – it must simply ignore an isolated schwa sound. The most common mistranscription seems to be a substitution of the word “and” for “in” and vice-versa.
There’s no punctuation, so one needs to go through the video and do the punctuation and capitalization of the first word of a sentence oneself. The transcription also doesn’t differentiate between voices, it all just runs together. To make life a little easier, when the text switched from one person to another I simply put the cursor just before the first word spoken by a person and then hit Enter. I also did that at the start of any sentence which I wanted to start a paragraph in the transcription. I didn’t really time myself for this step, but after YouTube did its thing I estimate it took somewhere between four and six hours to get the captions corrected and properly formatted for a 63 minute podcast – not trivial!
Going from the corrected YouTube closed captions to a readable transcript was a little tricky, so I’m going to post my method here in case anyone else wants to do something similar themselves.
When copying the entire YouTube caption page with ctrl-A ctrl-C, the timestamps are also copied. I pasted that into Notepad, stripped off the header and footer information that was copied over from the YouTube page, and then copied and pasted the remainder into an Open Office spreadsheet. The timestamps meant that the first row of text was stored in cell A6, the second in cell A12, and so forth, all the way to A10164. So, I gave cell C1 the following contents:
=indirect(“a”&row()*6)
and then copied that and pasted it into cells C2 to C1694.
From there, it was just a matter of putting the letter P in column B wherever I wanted to start a new paragraph, and H or G in column B if the line in column C was a new statement by the host or guest, respectively. I also put the letter P in column B in the row immediately below the last line of text in column C (in this case, B1695). Then column D was created to either continue the paragraph started on a previous line or start a new paragraph of text. Cell D1 was simply a copy of cell C1, and cell D2 was the following:
=if(b2=”p”;c2;if(b2=”h”;”Host Name: “&c2;if(b2=”g”;”Guest Name: “&c2;d1&” “&c2)))
and then I just copied D2 and pasted it into D3 through D1694.
Finally column E copied column D if the contents of column B on the next row was either P, H, or G. Contents of E1:
=if(or(b2=”p”;b2=”h”;b2=”g”);d1;””)
and copied that and pasted into E2 through E1694. That gave me entire paragraphs or new lines of text by either the guest or host on single lines, with blank lines in between. Then it was just a matter of copying cells E1:E1694 and pasting it into Notepad, and deleting the blank lines.
The steps with Notepad and Open Office only took about half an hour, and the result was a 63 kb text file, almost exactly one kilobyte of text per minute of podcast. It worked out to about an hour of work per ten minutes of podcast, which is probably why you don’t see most podcasters doing this. My hat is off to Rush Limbaugh’s staff, who transcribe a four hour show every day.
Android has really good voice recognition, maybe same as youtube, so not sure why transcriptions would be off. Maybe because less frequently used words based on the topic? Haven’t seen any good apps though.
A good podcast with so many goo points, but you sounded a bit depressed, Rand. Speak slower and more animated next time. 😉
I hate podcasts. It’s usually much faster to just read a transcript.
Me, too, but for those who don’t mind…
That’s what I told him when he asked me. I said I don’t mind doing podcasts, but I hate listening to them.
Maybe he could transcribe it. There ought to be software that can do that these days, with voice recognition, though it would no doubt need a lot of editing to clean it up.
Interesting, thanks. Particularly the historical perspective on the great American observatories.
Youtube has a built-in transcription service. here’s one way to use it to convert audio to text (quality probably not very good)
I tried that for our podcast (shameless plug: talkuniverse.org). It actually works really well, getting only a few words wrong. The only problem is no punctuation. So, I had to go through it and do all the punctuation. Still, it’s faster than trying to transcribe the whole thing. We’ve got over 40 episodes in the can right now, so it’s going to take me about six weeks or so to get them all on YouTube and transcribed, but this is a great way to reach the deaf and those who simply don’t like listening to podcasts. Thanks for the tip.
After transcribing two shows, I can report a few things:
First, YouTube’s auto-transcription feature is very good, but not perfect. But hey, it’s free, so what can you expect? The accuracy is up around 95% to 99%, depending on how fast the person talks. The utterance “uh” normally doesn’t get transcribed, but that means that words like “a” and “the” are often missing as well – it must simply ignore an isolated schwa sound. The most common mistranscription seems to be a substitution of the word “and” for “in” and vice-versa.
There’s no punctuation, so one needs to go through the video and do the punctuation and capitalization of the first word of a sentence oneself. The transcription also doesn’t differentiate between voices, it all just runs together. To make life a little easier, when the text switched from one person to another I simply put the cursor just before the first word spoken by a person and then hit Enter. I also did that at the start of any sentence which I wanted to start a paragraph in the transcription. I didn’t really time myself for this step, but after YouTube did its thing I estimate it took somewhere between four and six hours to get the captions corrected and properly formatted for a 63 minute podcast – not trivial!
Going from the corrected YouTube closed captions to a readable transcript was a little tricky, so I’m going to post my method here in case anyone else wants to do something similar themselves.
When copying the entire YouTube caption page with ctrl-A ctrl-C, the timestamps are also copied. I pasted that into Notepad, stripped off the header and footer information that was copied over from the YouTube page, and then copied and pasted the remainder into an Open Office spreadsheet. The timestamps meant that the first row of text was stored in cell A6, the second in cell A12, and so forth, all the way to A10164. So, I gave cell C1 the following contents:
=indirect(“a”&row()*6)
and then copied that and pasted it into cells C2 to C1694.
From there, it was just a matter of putting the letter P in column B wherever I wanted to start a new paragraph, and H or G in column B if the line in column C was a new statement by the host or guest, respectively. I also put the letter P in column B in the row immediately below the last line of text in column C (in this case, B1695). Then column D was created to either continue the paragraph started on a previous line or start a new paragraph of text. Cell D1 was simply a copy of cell C1, and cell D2 was the following:
=if(b2=”p”;c2;if(b2=”h”;”Host Name: “&c2;if(b2=”g”;”Guest Name: “&c2;d1&” “&c2)))
and then I just copied D2 and pasted it into D3 through D1694.
Finally column E copied column D if the contents of column B on the next row was either P, H, or G. Contents of E1:
=if(or(b2=”p”;b2=”h”;b2=”g”);d1;””)
and copied that and pasted into E2 through E1694. That gave me entire paragraphs or new lines of text by either the guest or host on single lines, with blank lines in between. Then it was just a matter of copying cells E1:E1694 and pasting it into Notepad, and deleting the blank lines.
The steps with Notepad and Open Office only took about half an hour, and the result was a 63 kb text file, almost exactly one kilobyte of text per minute of podcast. It worked out to about an hour of work per ten minutes of podcast, which is probably why you don’t see most podcasters doing this. My hat is off to Rush Limbaugh’s staff, who transcribe a four hour show every day.
Android has really good voice recognition, maybe same as youtube, so not sure why transcriptions would be off. Maybe because less frequently used words based on the topic? Haven’t seen any good apps though.
A good podcast with so many goo points, but you sounded a bit depressed, Rand. Speak slower and more animated next time. 😉