1. Always divide the sentences based on who’s speaking
It may happen that two separate speeches from two different people can get merged into one, or that the last sentence from one speaker can be merged with that of the following speaker. Always separate them so that everyone can be noted and quoted for their sentences.
Incorrect
Don't forget to check all transcriptions and make sure that Speakers are correctly assigned.
Correct
Remember to separate Speakers when they are accidentally merged by automated transcriptions.
2. Add punctuation
The automatic transcription can sometimes miss question marks, quotation marks, exclamation marks, and so on. If so, you should add them (only one per sentence, don’t go with “What???” or “Wow!!!). There should always be a punctuation mark at the end of each line.
Incorrent
Hey how are you
Very good, thank you
Correct
Hey, how are you?
Very good, thank you.
4. Use a single dash to indicate an unfinished or censored word or sentence.
When a Speaker doesn’t finish a sentence, just add a single dash (–) to show the word has been interrupted:
Incorrent
I was ready for the moment and I was ready for
Correct
I was ready for the moment and I was ready for -
You can follow the same rule for when you find a censored word in a podcast:
Incorrent
I got people to see and sh** to do
Correct
I got people to see and sh- to do
To keep punctuation consistent across podcasts:
- Always use the En dash (–) where there is a dash or an ellipsis, or when a word is not possible to accurately transcribe.
- Shorter Hyphens (-) are used to separate compound words.
- Do not use the longer Em dash (—).
5. Transcribe music lyrics, but not too many!
You can transcribe music lyrics if it’s just a few lines (5-7 lines) and you are 100% sure of the content. Like this 👇
If the podcast is playing a full song or a large portion of it (but the song is not the main object or topic of the podcast) remove the whole automated transcription and replace it with the name and author of the song, in brackets 👇
Incorrect
Write the whole lyrics of the song
Correct
[🎵 Livin’ on a Prayer - Bon Jovi]
This option above is a valid one even if you are uncertain about the lyrics, as long as you know the correct title and artist name.
If you are not sure about the correctness of the small part of the lyrics added by the artificial intelligence, and you don’t know who the author of the song is, just delete the whole part and go on correcting the rest of the podcast.
6. Don’t describe. Transcribe!
Always transcribe direct speech but never audio description: you can avoid referencing “Music in the background” or “Noises in the background”.
Incorrect
"Hello Conan"
door closes in background
"Hello Michael!"
Correct
"Hello Conan"
"Hello Michael!
Incorrect
It's going to get curly soon (Conan laughs in the background) is that a blowout?
Correct
It's going to get curly soon, is that a blowout?
7. Remove laughter, filler sounds, humming, and so on
Eliminate filler sounds and repetitions while transcribing a Podcast
• Filler sounds are “uhm”, “ehm” and so on. “Yeah” is a filler sound when it is not necessary to state an affirmation. In this case it can also be removed.
• Repetitions are words that come right after one another, as in “very very…”, and you should delete only one.
• Other words that should usually be removed are “like”, “you know”, “I mean”, “Right” and so on – when they are clearly functioning as verbatim or fillers.
Attention!
Sometimes a Speaker expresses assent or dissent right over the voice of someone else. You can and should transcribe those expressions when they're clear and distinct. On occasion, though, these interactions can be difficult to separate. Consider whether to keep these expressions on a case-by-case basis.
Incorrect
Uhm and I came to learn, uhm as I grew up.
Correct
And I came to learn as I grew up.
8. Delete all transcribed repetitions
It may happen that a speaker repeats a part of the speech two or three times in the same sentence. Transcribe the repeated parts only once, unless the repetition itself is instrumental to the concept.
Incorrect
I mean, I mean, do you really think that?
Correct
I mean, do you really think that?
9. Don’t transcribe advertisements
Advertisements (Ads) are treated differently from the rest of the text and don’t follow the rules of Speakers and Tags. Once text is marked as an Ad, there is no need to tag or correct anything within it.
See the screenshots below to get an idea of the process and final result 👇
What is to be marked as an advertisement
Ads may come at the beginning, end, or middle of a podcast
Prerecorded-ads
Pre-produced or prerecorded podcast ads are similar to a traditional radio spot. Usually, you can hear the difference in voice, tone, even volume from the original episode.
This should be marked as an advertisement.
Host-read
Host-read ads are live-read podcast ads read by the podcast host(s) during the recording of a podcast. The ads are often delivered without a script and become permanent parts of the podcast episode. They sometimes include a special offer for podcast listeners e.g. "Song Exploder listeners get 15% off an annual membership."
Promo
If the ads are promoting other content eg. another podcast or episode, or the network itself is considered an advertisement.
10. Credits: to transcribe, or not to transcribe...
Sometimes at the end (or beginning) of a podcast, the host lists all the people who contributed to the production (producers, writers, and so on).
Whether you transcribe these credits or not it is really up to you. We ask only that you're consistent in your decision and that, if you do choose to keep them, you don't tag them.
Attention: when credits have been added automatically by the AI, please remove them all (names’ spelling will likely be wrong).