Speaker A
Okay guys, so welcome to our Magic Data transcription training meeting. I'm going to share my screen now.
Complete tutorial on Magic Data transcription platform covering interface, segmenting, tagging, and best practices for accurate transcription.
Ask about this video. Answers come from its transcript only — with the timestamp, so you can check them.
Generated from the transcript and can be wrong — check the timestamp.
Key Takeaways
- Use the soundwave and segment numbers together for precise navigation.
- Always save your work frequently to prevent loss due to technical issues.
- Tag different types of noises correctly to maintain transcription quality.
- Maintain segment length of at least 0.2 seconds for clarity.
- Transcribe slang and spoken language as heard without correcting grammar.
What the video covers
- Introduction to Magic Data transcription platform and overview of its interface including soundwave and segment navigation.
- Explanation of speaker attributes selection (male/female) and how it applies across segments.
- Guidance on handling two-channel audio where only one speaker's audio is heard per task.
- Instructions on segment creation, minimum segment length (0.2 seconds), and using playback speed adjustments.
- Importance of saving progress frequently to avoid data loss due to platform crashes or logouts.
- Detailed explanation of tagging different audio types such as system noise, ambient noise, laughter, and overlapping speech.
- How to use segment numbers for easy navigation and rechecking during transcription.
- Recommendations for working in noise-free environments and recognizing common noises in recordings.
- Tips on handling slang, spelling, punctuation, and transcription accuracy standards.
- Final steps on submitting completed transcription tasks and avoiding common mistakes.
Chapters
- 00:00Introduction and Platform Overview
- 02:27Speaker Attributes and Audio Channels
- 04:47Handling Audio Issues and Playback Speed
- 07:26Saving Progress and Segment Navigation
- 09:31Segment Creation and Tagging System
- 11:37Using Tags for Noise and Overlaps
- 14:56Best Practices for Transcription Accuracy
- 18:20Final Tips and Submission Process
Full Transcript — Download SRT & Markdown
Speaker A
different options mean uh we'll start with the Soundwave here uh this is the the Soundwave of the whole recording uh if you scroll it until the end you can see it's uh it's a 30 minute uh uh clip
Speaker A
Okay, so this is basically our platform. I'm going to give you a quick overview of what the platform is and what different options mean.
Speaker A
you actually want it to be um then on the left side you have uh this uh attributes of the speaker male or female once you choose male or female then it will be uh then it will be by by default
Speaker A
We'll start with the Soundwave here. This is the Soundwave of the whole recording. If you scroll it until the end, you can see it's a 30-minute clip.
Speaker A
audio as well please remember that these are uh two channel uh audios so even though it's a conversation recording between two people but once you get a task you only hear one person's audio uh in the in the task so you will not be
Speaker A
Which you have to finish once you take this task. You can navigate this using the scroll, but we normally recommend using the mouse wheel because it's more accurate. Scroll sometimes can be a bit too fast and you might miss where you actually want it to be.
Speaker A
instances it doesn't matter but if the whole audio is like that then that means there's some uh some problem with the audio um so and over here is the playback speed you can if the somebody sometimes people while having a conversation they
Speaker A
Then on the left side, you have the attributes of the speaker, male or female. Once you choose male or female, then it will be by default the same attribute for all the other segments as well.
Speaker A
don't need to touch that no you don't need to touch any of the buttons on the top other than save and submit submit means you have completed the whole task completely all 30 minutes of it and now uh you want to uh submit it like your
Speaker A
So if you play this segment, you can see it's a female. So once we choose female, then it will stay female for the rest of the audio as well.
Speaker A
segments because sometimes the platform can log out something can happen something can crash and you might lose your progress so don't end up you know working on the whole 30- minute task and then you know saving it you should save
Speaker A
Please remember that these are two-channel audios, so even though it's a conversation recording between two people, once you get a task, you only hear one person's audio in the task.
Speaker A
coming in so and on the right side you have the segment numbers so every segment has a number for for example when I uh chose this segment you can see this one written over here so it corresponds with
Speaker A
So you will not be able to hear the other person. If you're able to hear the other person, that means there's some problem with that audio and you should let your manager know about it.
Speaker A
to keep scrolling around and keep looking at the Soundwave as well you need to basically use a combination of these both okay so like okay so so this is how you can move move between them uh you can you you guys can
Speaker A
But if it's just like one or two instances, it doesn't matter. But if the whole audio is like that, then that means there's some problem with the audio.
Speaker A
let me let me quickly check how sharing audio Works uh choose only the window you want to share okay not not the whole thing right yeah okay magic data sound recording can you hear it now play play first let us listen magic
Speaker A
So, and over here is the playback speed. Sometimes people, while having a conversation, they can be very fast, so you can reduce the speed over here to 0.75. It will help you transcribe more accurately.
Speaker A
what the platform looks like what each each button does now let's start with the transcription now you can see the the segment is over here each segment basically represents one whole logical sentence or uh you know uh words
Speaker A
Spell check, you don't need to touch that. Model language also, you don't need to touch that. No, you don't need to touch any of the buttons on the top other than save and submit.
Speaker A
make the segment at least 0.2 seconds long magic data sound recording magic data sound recording so what is 0.2 seconds how do you calculate at 0.2 seconds basically you can also use you can see the the segment time stamp over here uh once you
Speaker A
Submit means you have completed the whole task completely, all 30 minutes of it, and now you want to submit it, like your letting it go to the client.
Speaker A
down but that's generally how it should be the same way it goes over here uh it should be at least 0.2 seconds magic now you can see over here there's some noise here and I just overlapped on this noise so that is something I
Speaker A
And save button is to save your progress because you might not be able to finish the whole task in one go.
Speaker A
created by the machine so they might be wrong as well see so this is this is also some uh some noise and this was also some noise so what I'm going to do is what I just did is that I dragged and selected to
Speaker A
And we also recommend saving after every 15 to 20 segments because sometimes the platform can log out, something can happen, something can crash, and you might lose your progress.
Speaker A
the on the platform as well which normally you do but you can just use this uh the Tab Key now this is basically system sound okay so in this situation we need to use the system uh system tag okay
Speaker A
So don't end up working on the whole 30-minute task and then saving it. You should save in the middle multiple times.
Speaker A
click and then you drag until the point you want the segment to be like for example I want it until here and I'll just release the mouse uh the left Mouse button and it will create the segment okay if I want to create one here one
Speaker A
Over here you have the tags which you're going to use. Every tag represents some situation. I will tell you in detail what each of these tags mean.
Speaker A
voice inside so what I'll do is that I'll select zero because without click on zero I cannot select my labels there are two kinds of label one is a speaker label which is for overlap but overlaps don't happen over
Speaker A
Okay, still have some people coming in. So, and on the right side you have the segment numbers. So every segment has a number.
Speaker A
might might be you know needing to use it uh otherwise you'll use no speaker labels no speaker label means that you have to select zero to be able to choose them okay right now you can see it crosses out I cannot choose it so right
Speaker A
For example, when I chose this segment, you can see this one written over here. So it corresponds with this list of segments over here.
Speaker A
explain them as well uh you have the ens tag and once you hover over it after pressing zero you can see what what it means other type of visible ambient noise such as knocking animal calls Etc so what does that mean if you have any
Speaker A
And you can use these segment numbers. These are like buttons and they're very convenient to basically recheck your task and basically move over to the next segment.
Speaker A
sound over here you can see this is this is a system system generated sound it's a computer noise this is not from Real World so this becomes system noise but if there was a for more than 0.5 seconds we know that
Speaker A
But at the same time, you need to keep scrolling around and keep looking at the Soundwave as well. You need to basically use a combination of these both.
Speaker A
people to use uh be in a noise-free place to record so in most cases you don't have ens sounds as well the most common kind kind of noise you have is system Noise Okay then you have music again music very rarely you will have
Speaker A
Okay, so this is how you can move between them. You guys can hear the audio, right? I don't need to share the audio or something like you have to do it on Zoom.
Speaker A
then you have to make that Mark that as sonant then laughter if the laughter has two situations basically uh the first situation is that you are uh you encounter it in sound recording for example over here she's like magic data sound recording
Speaker A
I don't know, just can you guys hear the audio of this? You cannot hear the audio? Okay, wait.
Speaker A
laugh okay so in that situation that if that laughter is more than 0.5 seconds you have to take it out of the segment you have to create create a new segment and Mark it as laughter again not very
Speaker A
Let me quickly check how sharing audio works. Choose only the window you want to share, okay? Not the whole thing, right? Yeah.
Speaker A
people have a habit to you know fill the whole screen with uh hahas so we cannot do that okay so that was laughter then you have pii pii is basically sensitive information personally identifiable information what does that mean your
Speaker A
Okay, Magic Data sound recording. Can you hear it now? Play, play. First let us listen. Magic Data sound recording.
Speaker A
New York I in Islamabad no that that is not address but if somebody's like oh I'm in Islamabad F7 Street 21 house 15 then that's that is something you need to Mark as pii and you because you don't
Speaker A
Yeah, Magic Data sound recording. Okay, clear audio. You can hear it loud and clear? Yeah.
Speaker A
zanark but if I'm talking to somebody or if somebody's mentioning me they will call me Zan that is fine that is not pii uh tag but if somebody says Zan taric did this that is a full name and then it
Speaker A
Okay, perfect. Okay guys, so now let's start with our transcription part. So you already know what the platform looks like, what each button does.
Speaker A
you hear a male sound that means that that person should be marked as NPS and then the last one is inaudible or incomprehensible speech including large passages of dialect or foreign language so if you hear anything other than English
Speaker A
Now let's start with the transcription. Now you can see the segment is over here. Each segment basically represents one whole logical sentence or words.
Speaker A
understood magic but I did not understand data and I can understand sound recording right so there are four words I can understand the first third and fourth I cannot understand the second word so what would I do in that
Speaker A
Basically, each segment should have a little bit of buffer space on both sides. Over here you can see that there's no buffer space. The sound ends and the segment ends right there as well.
Speaker A
understand is over here sound recording sound recording and then you have the third and the fourth word over here okay sorry segments total uh situation G basically seg okay incomprehensible because we have to take out the incomprehensible part so it will
Speaker A
So in this kind of situation, you should make the segment at least 0.2 seconds long. Magic Data sound recording. Magic Data sound recording.
Speaker A
between sometimes can even be 600 sometimes can be 200 300 or uh then it also depends on how many uh this uh invalid segments come in the middle because sometimes you have this noise Parts in the middle so you
Speaker A
So what is 0.2 seconds? How do you calculate 0.2 seconds? Basically, you can also use, you can see the segment timestamp over here once you hover over it.
Speaker A
you can do it in different ways as well but this I think is the most efficient way magic magic magic magic magic magic magic until here then we will make another segment on top of this it it it
Speaker A
Or you can, each one of these digits is 0.1 second. Okay, so this will be basically your reference. You don't have to be exactly accurate about it. It can be a little bit up and down, but that's generally how it should be.
Speaker A
recording now you can see over here that there's very little space so we cannot leave that buffer space magic magic magic to have a 0.2 second buffer space so it's fine because we know that there's no space over here but you
Speaker A
The same way it goes over here. It should be at least 0.2 seconds. Magic. Now you can see over here there's some noise here and I just overlapped on this noise.
Speaker A
getting a call sorrys magic magic magic magic magic magic magic magic magic magic magic magic magic magic magic magic magic it it it incomprehensible in recording recording recording recording recording recording recording recording sound recording okay so this would be the basic system hell what's
Speaker A
So that is something I shouldn't do. I'll make it a little bit shorter. Magic. Okay, and this noise over here and there's something over here as well.
Speaker A
comma apostrophe these are the only punctuations and in some situations hyphen uh another thing which you have to keep in mind is uh numbers the numbers should be transcribed in uh as they're spelled Arabic numeral proper nouns capitalized
Speaker A
Let's listen to it. What is it? Because some parts, because all the segments you see right now, these are created by the machine, so they might be wrong as well.
Speaker A
long uh or segment Beach a 0.2 seconds of buffer 0.2 second buffer 0.5 seconds noise noise orter and then 0.8 second silence okay be segment 0.8 second silence then you have to divide it example guys if you have any uh if you
Speaker A
See, so this is also some noise and this was also some noise. So what I'm going to do is what I just did is that I dragged and selected to create a new segment.
Speaker A
segment segment full stop even if the segment is not completed sorry the sentence is not complete okay sir this basically comes with merging com does that make a full sentence logical sentence.
Speaker A
Okay, I'll do it again. Magic record. I'll click sound recording. And here to pause and play, I'll use the tab key. That is the most convenient way to do it.
Speaker A
also noticed one thing in Instagram sounds for example let's say uh is same same same same segment uh incomplete sentence word next segment segment 20 combine so it makes it a complete logical sentence so in that case we use this
Speaker A
There's no button here on the platform as well which normally you do, but you can just use the Tab key. Now this is basically system sound.
Speaker A
segment model com this is basically merging or merging the distance between the two words is segment segment 0.2 seconds so you have to basically calculate from where this word ended 89.9 and where this word starts 90.3 so 89.9 or we can count it as 90 90 or 9.39
Speaker A
Okay, so in this situation we need to use the system tag.
Speaker A
same same same same another important point is that if the person stammers like I like for example the word supposed if the person says s supposed or if they say suu Supermarket okay to this stammering has to be written
Speaker A
Yeah sir, can you please tell me again how to create a new segment?
Speaker A
is basically filler filler I'll I'll come to that next yeah this is basically like uh like for example let's say word like Instagram I want to say Instagram but I'm like in in Instagram start ins that is stem ring so I have to write
Speaker A
Sure. Okay, I'll do it again. So it's the same way how you drag and select multiple things on a screen.
Speaker A
modal sounds we have to uh transcribe that we have to uh like Mark them within the sentence or what we once we'll do a few practice uh sentences we'll we'll do uh we'll check that out as well okay and
Speaker A
Okay, on your desktop you just click and then you drag until the point you want the segment to be. Like for example, I want it until here and I'll just release the mouse, the left mouse button, and it will create the segment.
Speaker A
magic data sound recording magic data sound recording magic data sound recording magic data sound recording magic data sound recording magic recording it's a proper noun so it be capitalized all the punctuations if you see these very small waves in the middle noise you can ignore
Speaker A
Okay, if I want to create one here, one here, one here, like this.
Speaker A
segment you can give the buffer space so hello what's up so hello what's up one second somebody's joining okay and you can see like right next to this segment there's some something over here as well we'll quickly check if it's
Speaker A
Okay, and you can delete them by clicking on this cross on the top.
Speaker A
up so hello what's up so hello what's up so she's saying suila what's up I don't know what the exact name is but I'll guess it as suila and I'll just keep calling this person suila unless in some
Speaker A
So hello, what's up? Okay sir, I got it.
Speaker A
fine I'm fine I'm fine I'll make this buffer I'm fine fine sides I'm fine I'm fine I'm fine I'm fine I'm fine I'm fine I'm fine I'm fine three times fine I'm fine the machine transcription is completely correct the punctuations are
Speaker A
Okay, okay. So now over here you can see there's no human voice inside. So what I'll do is that I'll select zero because without clicking on zero I cannot select my labels.
Speaker A
the middle yeah so first let's listen to that as well I'm fine it's I'm fine I'm fine it's just going okay so you can see that this is also part of the speech so we'll extend this I'm fine it's just queen I'm fine
Speaker A
There are two kinds of labels. One is a speaker label which is for overlap, but overlaps don't happen over here because this is a conversation recording which is single channel.
Speaker A
fine it's just going my going going I'll I'll go with the machine transcription I think that's what she said fine it's just your y you can see over here that this segment is basically cutting the the audio so I'll extend it yourss yourss
Speaker A
So you only hear one of the person's audio. So what does that mean? We don't need to use this overlap in most cases.
Speaker A
just now talking to you right now now I'm just I'm just having lunch I'm just now talking to you right now now I'm just I'm just having lunch I'm just now talking to you right now now I'm just
Speaker A
Very rarely you might be needing to use it. Otherwise, you'll use no speaker labels. No speaker label means that you have to select zero to be able to choose them.
Speaker A
she said I'm just I'm just two times so you also have to type it two times okay so so now we've done like seven segments normally you should do it like between between 15 15 to 20 segments but um I'll just save it right
Speaker A
Okay, right now you can see it crosses out. I cannot choose it. So right now this segment has system noise, so I'll click zero and I'll click on system and now I have the tag over here.
Speaker A
obviously Facebook Instagram and just scrolling the mobile phones you know and just scrolling the mobile and just scrolling the and just scrolling the mobile phones you know Instagram and just scrolling the mobile phones you know social networks obviously Faceook the mobile phones you
Speaker A
Okay, now there are these different kinds of tags. I will explain quickly.
Speaker A
be capitalized Z this is 23.7 n seconds till 4 minute you count this how many minutes duration 4.74 this uh this segment yeah this segment is for 4743 yes you can see seconds seconds seconds four seconds one secondment okay
Speaker A
You have the ENS tag and once you hover over it after pressing zero, you can see what it means: other type of visible ambient noise such as knocking, animal calls, etc.
Speaker A
yeah like three hours uh would be for someone who's really good average should be between three to four hours for one hour for one hour okay but some but somebody who's new to the platform or new to transcription it will be a bit
Speaker A
So what does that mean? If you have any noise in your audio which is longer than 0.5 seconds, this is very important, 0.5 seconds, you have to remember that.
Speaker A
day all over the day all over the day the machine trans is quite sometimes it's not Capital all over the day yeah yeah yeah yes please how much space do we have to add like this uh buffer space in the
Speaker A
Like for example, there's a continuous noise like this. Now I know this is a knocking sound, right? This is not a system generat...
Speaker A
because uh in transcription we normally get paid for valid hours right so this buffer space counts in valid hour so the more uh so leaving out buffer space is basically decreasing your valid hours as well okay now let's go to the
Speaker A
next this one you can you can if you you can if you hear so you can leave it as so if you hear sure you can write down sure if you don't understand at all you can just mark it as zero and
Speaker A
unintelligible but you need to make sure that you don't overuse this unintelligible uh only use unintelligible where you are completely sure that nobody else can you know guess what word it is only in that situation okay otherwise it's so or sure whatever you're hearing
Speaker A
would be fine if we are so much addicted like if we are so much addicted like if we are so much addicted like if we are so much adct if we are so much addicted like we are so much we are so much we
Speaker A
are so much so see over here I don't understand what she's saying gria I don't know what what is this gria okay so I'll I'll take that out of the segment to much addicted to much addicted to much addicted to much addicted like to
Speaker A
much addicted like much addicted like much addicted much understand what it means asig okay and so on we'll keep going let's listen to this segment as well whenever whenever we get leisure time we just open up Instagram and just scrolling the
Speaker A
phones mobile phones and all Facebook Instagram all the time right when okay so now over here we this is you heard the audio it's completely accurate we just need to add the punctuations whenever whenever we get leisure time we just open up Instagram
Speaker A
and just scrolling the mobile mobile phone phone the phones mobile phones and all Facebook Instagram all the time right okay so yeah so we did we did around 14 segments together I'll I'll try to get to okay now
Speaker A
this this is not uh like this is not anything over here saying thank you you can you can delete this segment as well exactly exactly exactly exactly exactly exactly okay other than this the only situations which you need the leure time
Speaker A
like whenever whenever you get whenever we get leisure time we just open up Instagram start watching G all the time ni food reals like like all the this person just keeps repeating the same thing over and over again okay
Speaker A
so cannot find any any situation over here but uh I hope you understand the merging uh system so if you have a sentence which is not complete what does what does it mean that a sentence is not complete it basically means that the
Speaker A
logic of the sentence is not complete what is a logical sentence a logical sentence can be a one-word sentence as well like okay fine uh this is uh these are all logical if there if somebody says okay that is a
Speaker A
complete sentence uh an incomplete sentence would be like if the full sentence is I'm going to have dinner right now but in the segment you can only hear I am going to have now this is not a logical
Speaker A
complete sentence logically it's not complete there has to be something after have like what what what am I having okay or I'm going to to the supermarket or I'm going to the going to another city so but in the segment you can only
Speaker A
hear I'm going to so then in this case you need to see if the next segment is within 0.8 uh seconds okay if it's within 0.8 seconds and the segment after that completes the meaning of the previous segment then you will merge it
Speaker A
if the second segment is more than 0.8 you will leave the first segment as it is and you will end It On full stop but if the second segment is within 0.8 and it finishes the meaning then you merge
Speaker A
it if the second segment is within 0.8 but it does not complete the meaning of the segment then you also don't need to merge it because it will just waste your time and you know energy so these are the guys the basic guidelines
Speaker A
if anybody has any questions uh please feel free to ask guys any questions something if uh anyone having any quer yeah somebody asked one question like how we are going to uh measure uh this uh this buffer so you
Speaker A
don't need to measure it like to the like exact to the final dot you can just see it like this this line Falls right on this digit this digits you see on top these are micros seconds so from here
Speaker A
until here is 1 second 1 2 3 4 5 6 7 8 9 10 so this is uh basically your micros seconds so if I want to see this is how big this buffer is so this is 0.1 seconds this is 0.2
Speaker A
seconds I have I have that's it uh yes the longest segment can be 15 seconds the platform will not allow you to make a segment longer than 15 uh seconds uh 0.5 second rule applies only for noise tags but not for intelligible tag
Speaker A
yes intelligible uh like unintelligible words can be less than 0.5 seconds uh slangs over here uh uh you basically asked for slangs so what what do we do with slangs um there are some standard words in transcription which you write as they
Speaker A
are uh mentioned like Wana Gana uh what else would be there uh cuz like but if somebody says because not because you will still write it because but if somebody just says C you will not write it as because all right
Speaker A
so these kind of words you are going to write them uh as as slangs so you will spell the slangs as they are you will write what you hear you will not change you will not correct the grammar you will not correct the
Speaker A
sentence you will not correct the structure you will write as you hear the words buffer yes buffer is 0.2 seconds 0.5 if you have continue 0.5 second noise then yes you will use the noise tag 0.8 second is for merging the segment and
Speaker A
it's also for dividing a segment so for example if this segment over here this was 0.8 second long silence see from here until here if it was 0.8 second I would make I will cut this segment over here I will bring this until here and I
Speaker A
make a new segment on the next part all right so I think yeah most of the main questions are done um so uh guys how uh like I would want that uh each one of you uh basically work on at least
Speaker A
20 to 30 segments save them and then we check we check your work you will provide us IDs yes how many uh right now I have uh I don't know how many like I we have a lot of IDs but
Speaker A
I don't know how many seg uh like how many can be activated right now but let me let me just check quickly how many are there if we pass how many members you need for this project uh for this for this project uh
Speaker A
we need over over 50 people okay ideally yeah if they're like but 50 is like full full-time people so if you are have like uh some people working parttime then that number increases okay so right now we have
Speaker A
uh yes to yeah uh you will click you will click on submit once you're like once you are at the last segment you're done with you're done with everything uh please remember once you're done with everything then you have to like go one
Speaker A
by one in each segment there's a chance sometimes you miss out something there's some spelling mistake some punctuation mistake capitalization there are a few like cardinal sins which should not be committed uh when you're when you're doing transcription uh number one is
Speaker A
capitalization error the first word of the sentence should always be capitalized proper nouns like names of people cities and all of that should always be capitalized uh your punctuations like full stops commas everything should always be there there should be no
Speaker A
spelling mistakes so these four or five things these are the most basic thing like it doesn't matter if you understand the platform you don't understand the platform or whatever situation is happening these mistakes should not be there and the segments buffer 0.2 second
Speaker A
buffer every other error other than that is acceptable as long as you just do it you know once or twice and then you correct yourself but this these like capitalization punctuation spelling mistakes this should not happen so second let me I was just
Speaker A
checking the so I have around 20 IDs right now and uh let me let me basically activate let me activate those IDs let me see how many IDs can be activated and then I'll let you know in the group is that okay
Speaker A
and then I'll share the IDS with you you just tell me which you give the ID to which person just share with me the name yeah then later on in the group uh I will post the message according to
Speaker A
that okay okay okay perfect perfect so uh and most of the all of the uh members who joined our meeting today uh they'll be working at night or in daytime most of them are night a few like two or three will be during the
Topics:Magic Datatranscription tutorialonline transcriptionaudio segmentationspeaker labelingtranscription tagstranscription best practicesaudio noise taggingtranscription platformFey Arn Ai Agency
