Cheap, Easy Audio Transcription with Mechanical Turk

After recording last week’s interview, I was left with a 36-minute MP3 and a profound feeling of dread. You see, I hate transcribing audio. I used to transcribe interviews in high school, and it’s always tedious, taking upwards of eight times the length of the clip itself.

Bracing for a good four or five hours of rewinding and writing and rewinding, I remembered that this is The Future! So, instead, I tossed the job over to the global anonymous workforce at Amazon Mechanical Turk instead.

The result: my 36-minute recording was transcribed while I slept, in less than three hours, for a grand total of $15.40.

This is a fraction of the cost/time of any other transcription service online, including the Turk-driven Casting Words, though you potentially sacrifice some quality. In my experience, though, there were virtually no errors.

Here’s how to do it yourself, with no programming knowledge required. The instructions below are verbose, but using my template, it shouldn’t take you more than five minutes of setup per job.

Step 1: Prepare your audio.

First, I split my 35-minute audio file into seven five-minute MP3s. Why? Mechanical Turk workers are all working in parallel, so the more discrete tasks, the faster the job gets done. This also diminishes the risk of one bad worker ruining your whole job. (Though you’re always allowed to reject bad submissions, and you’ll never have to pay for those.)

I used the open-source Audacity to split the files, but you could just as easily use any audio utility or editing software. Optionally, you might want to make each clip overlap by a few seconds, so you’ll be able to easily recognize where each segment of the transcript starts and stops.

Name the files sequentially. In my case, they were interview_1.mp3 through interview_7.mp3. When you’re done, upload the files somewhere they can be downloaded publicly. You’ll need the full URLs later.

Step 2: Design your HIT template.

Mechanical Turk jobs are called HITs — short for the dystopic-sounding “Human Intelligence Tasks.” After you’ve signed up as a new Requester on Mechanical Turk, you can design a new template from the homepage using one of several samples. Choose the Default Template.

On the Properties screen, we’ll write a short description of the task, define how many people we want to work on it, and how much we’re willing to pay them.

For a five-minute MP3, I think allotting two hours per assignment is ample time, and I expired the entire HIT in 12 hours because I was in a hurry. As for pay rate, you’ll need to determine the “Reward per Assignment” based on the difficulty of the task and what you think is fair. I chose $2.00 per five-minute MP3, or about $0.40/minute. Depending on the difficulty, you might want to try going higher or lower.

I only wanted one worker to attempt each clip, so I changed the “number of assignments per HIT” to 1. (If you want redundant transcripts for each clip, change this to 2 or 3… But be aware your costs will double or triple!)

After entering all this information, here’s what my finished Properties screen looked like:

On the Design Layout screen, you design the template that gets displayed to each worker, using basic variables that will be substituted later. For this template, we make up only one variable named “$url.” You can call it anything you like.

The basics you’ll need are a title, some simple rules, the link to the audio file with a substitution variable, and a text form for the worker to type the transcription into. If you’d like to use my template HTML, here it is. (Make sure you change the path to your own audio files!)

Two things to notice in my example. First, the “${url}” variables will be substituted with values in the “url” column of the spreadsheet we’ll create in the next step. Second, any form element you create will end up in your final output from Mechanical Turk, so don’t worry about the naming. I called mine simply “transcription.” Here’s what the relevant part looks like in the final template:

Please transcribe this five-minute MP3:

<a href=“${url}“&gt;${url}</a>

Enter your transcription below:

<textarea name=”transcription” cols=”80″ rows=”30″></textarea>

For the worker’s convenience, I also added an embedded Flash player for the MP3, but this is entirely optional. When you’re done designing your template, it should look something like this:

On the next screen, make sure it looks the way you like, and click “Preview and Finish” to save the HIT template.

Step 4: Upload the data for your HITs.

Once we’re done designing our template, we can select it to create a new HIT batch. We’ll be creating a simple comma-separated file (.CSV) filled with the data that will be substituted into our template.

On the Publish tab, select the template you just created by clicking the “Select” button:

Now, Amazon generates a sample CSV for you to put the URLs to your MP3s in. Click the link to “Download a sample input file” and open the downloaded CSV in a text editor. If you’ve done everything right, it should look like this:





Replace the “Hit1_url_data” lines with the full URLs to your own MP3 files. For me, this looked something like:


And so on. Save the CSV file, and upload it to Amazon. When you’re done, your uploaded file should appear, with the number of input lines.

Step 5: Publish your HITs.

Select your uploaded input file, and preview the finished batch of HITs. You’ll be able to page through each HIT, seeing exactly what workers will see. Use this opportunity to test that your audio files can be downloaded and heard properly. If it all looks good, click “Next” to confirm and publish your batch. This is what the final screen looks like:

If you don’t have any money in your Amazon Payments account, you’ll be prompted to fund it with a credit card. After you’ve paid, click “Publish HITs” and you’re done!

Your HITs will publish out to the Mechanical Turk workers, who will find and work on your task. Depending on the length and number of your MP3s, expect some work back within an hour.

As they’re working, you can browse and approve the results. The final output is an exported CSV, a spreadsheet of all the finished work that can be opened in Excel for your review.


You’d be insane not to use this for your own transcription projects. Absolutely nothing else comes close in price and speed.

One thought: I suspect it’d get even faster if you split clips into more pieces. I’d bet that splitting into one-minute segments would reduce the time by at least half. I’ll bet you’d be able to command lower rates with smaller MP3s too, since the time commitment would be lower, driving more competition for the tasks. If anyone experiments along these lines, please let me know!


    We used to use CastingWords for our sites (including when they first started out. At that time, the service was pretty much straight using Mechanical Turk with not much in the way of post-editing.

    It worked well a lot of the time, but sometimes you would get phonetically transcribed stuff that was just horribly painful – unless you were into beat poetry, I guess. Andy, I’m guessing you lucked onto the ‘good’ Turkers for this type of thing, and/or they’ve got better at filtering out the terrible ones – this was a good while ago.

    However, CastingWords ended up raising their prices for well-edited stuff – they’re still a lot cheaper than actual transcriptionists, mind you.

    What we ended up doing, since we have a LOT of transcription, is just using two or three trusted game enthusiasts whom we pay a bit more than CastingWords minimum, but enjoy the subject matter we talk about. So it’s our own community-sourced three-person MTurk, I guess.

    I have used MTurk for transcribing articles, instead of OCR-ing, and that seems to work really well.

    One great thing about MTurk is that people on it are used to HORRIBLY rote tasks, like identifying road markings (!), so anything like transcription of an interesting interview is actually exciting for them.

    CastingWords is a great business, and I definitely encourage people focused on quality to go for them.

    For quick, dirty, and cheap transcription, MTurk worked out perfectly for me, but your mileage may vary. There’s really no reason for a turker to deliberately perform badly, since they can get rejected (or blocked) and not get paid for their work… The longer the transcription, the greater the incentive to do it right.

    Oh my word. I recently paid upward of $300 to have somebody transcribe two hours of audio so that I wouldn’t have to. (I hate transcription, too.) I’m bookmarking this for the future — no question.


    I routinely hire people off of craigslist to come to my studio and add keywords and metadata to photos that I’m planning on sending to my stock photography agency, and this seems like a really fantastic way to accomplish that same task. Do you know if it’s possible for Mechanical Turk to write the input of what the workers are typing back into the same file you’re sending them to transcribe? If I can figure out a way to set up AMT so that I can upload a series of images, and get them back a few hours later with embedded keywords, that would be, as they say, incredible.


    Try setting up an MTurk HIT for someone to write a program to do what you want. Talk about self-replication!

    Anyway, depending on how you attach the keywords to the files, there may be a tool like jhead (first hit on Google), which can modify EXIF and IPTC data, to help insert the keywords into the images.

    Hal: Mechanical Turk workers can do whatever you like, as long as it’s relatively simple and doesn’t require special software or identifying themselves.

    I’m amazed how poorly documented this is, but I believe simply adding an HTML file upload (i.e. “<input type=”file” name=”myimage1″>”) will allow a turker to upload a file, and you’ll be able to retrieve it. Somehow. I haven’t tested this yet.

    I just tested two approaches using the new batch HIT builder interface for receiving file uploads, and neither worked. You can receive file uploads individually using the previous version of the Requester Site, but you sacrifice all the power of the HIT builder templating. Oh, well.

    Thanks for your suggestions guys… looks like I have a new project for myself!

    I’ll poke around and report back.

    Interesting review, Andy. Like one of your earlier commenters, I’ve been turned off by expensive local transcription services. However, I also think the market for Mechanical Turk may be limited by the skills sets that are required to set up the jobs — not everyone will be able to quickly get up to speed on Audacity or coding HTML.

    My dream transcription app (described in an essay linked from my name) may be a few years off, but it will really revolutionize the market — and also impact online video and search. I can’t wait …

    Ian Lamont

    The Industry Standard

    For a five-minute MP3, I think allotting two hours per assignment is ample time […] As for pay rate, […] I chose $2.00 per five-minute MP3

    So you effectively paid someone $1/hr?

    Wow — all those folks who are commenting at how great this is don’t get to complain when their jobs are outsourced overseas.

    Maybe you can also find some children in India to get this work done even cheaper yet – for let’s say 45 Cents? Great idea, isn’t it?

    Not to be a prat, but based on what Andy said in the article (that transcription takes roughly 3 times the length of the clip), then a 5 minute clip takes about 15 minutes to transcribe.

    Now, $2 for 15 minutes’ work is $6 on an hourly basis, which is minimum wage out in my state, last I checked.

    Just double-checked and the post actually said upwards of 8 times, rather than three, which brings the cost per hour closer to $3, which does seem a bit low.

    This only words for generic conversations. If you are talking about something technical or specialized, it won’t work.

    The issues is simply being able to recognize/spell the specialized terms. This could be as simple as neighborhood names and people’s names. Or, it could be the jargon in a particular field.

    This is part of why medical transcription costs what it does, with its specialized vocabulary.

    @Tod: What an inane comment. Firstly, allotting someone two hours to transcribe a five minute MP3 is ample time. Many people could do it in fifteen or twenty minutes. In addition, as Andy writes, many of the people who use Turk are from the US.

    Mechanical Turk workers are anonymous, so if it’s truly confidential, you’re out of luck. But I imagine that would be true of any outsourced transcription.

    I suppose you could mitigate risk by using a tool like mp3splt to split the MP3 into teeny 30-second segments, but it’d take more work to assemble the final transcript.

    Outsourcing to another country? On the contrary, about 76% of turkers are here in the United States, and over half have bachelor’s degrees.

    I think you missed my point, which wasn’t that turkers are out-of-country, but that the pursuit of cheap labour is exactly why programming jobs and other services are being sent to Mumbai rather than the Bay Area. You can’t on the one hand celebrate a “cheap easy” service and at the same time decry companies that are outsourcing jobs for the precisely the same reason.

    People aren’t being forced into slave labor here.

    Of course not. But they’re also not being paid minimum wage.

    Think about it — If I couldn’t have found someone to transcribe my interview for less than $20, I would’ve just done it myself. This isn’t making professional transcription obsolete… Amateur transcription is opening up a new market for people that wouldn’t have used the services otherwise, and people are getting paid for it.

    Besides, I think programming is quite a bit different than transcription which, as demonstrated, can be done by completely untrained labor in spare time.

    I’m sure there are professional transcribers who would argue that this is amateur hour and it threatens their jobs, but that sounds dangerously close to the NO!SPEC group that’s fighting tooth-and-nail against crowdsourced design.

    There will always be a place for professional transcribers, just as there will always be a place for professional designers. But if an amateur can do it for dirt-cheap with reasonable quality, then there’s a clear and unique market for that too.

    I think programming is quite a bit different than transcription which, as demonstrated, can be done by completely untrained labor in spare time.

    I’m not sure there is any real principled difference. As I said earlier, the impulse to pay someone $1/hr to do transcription is not really any different than the impulse to pay someone $20,000/yr to do programming in India. Training really isn’t the issue here — if one is OK with paying someone very little to do this job, I don’t see how one can then complain when more skilled jobs are also sent to lower wage workers.

    I really don’t intend to be a jerk about this, and I’m glad that you’re satisfied with the quality of the work done. But for me it just seems wrong to pay someone so little per hour, regardless of whether there is a “clear and unique market for that”.


    You can upload your audio interviews to and get them transcribed for about $15 an hour with very good to excellent accuracy, depending on how technical, with 48-96 hour turnaround.

    I’ve been using the service for over a year and have recommended it to many friends. I’m very satisfied and I’m not affiliated.

    I would recommend doing your first transcription, or at least 15 minutes of it, yourself. You learn a lot. Like the fact that a conversation that you think you completely understand as a listener actually has quite a bit that you can’t understand if you’re actually trying to transcribe it. And that is true even if you yourself are taking part in the recorded conversation: You’ll find that *you cannot even understand your own voice*! With this experience you’ll cut transcribers a bit of slack.

    With the transcription task, isn’t it easier to just slow down the audio and adjust the pitch than to keep rewinding (and taking your fingers off the keys)?

    Also, “Now, $2 for 15 minutes’ work is $6 on an hourly basis”. Are you sure?

    Glad to see people using the new interface of Mechanical Turk. With the programming barrier down, it is just a matter of time to see clever tasks uploaded on MTurk that will really use the power of the tool.

    To those complaining about minimum wage: Mechanical Turk is not a place to distribute professional assignments. (Sites like and perform this task quite well.) It is more like freelancing because people can work whenever they want, for however long they want. A normal employee does not get this.

    MTurk is a tool that really brings the following benefits:

    (a) allows easy parallel execution of the task across tens and hundreds of workers,

    (b) allows splitting of tremendously tedious assignments into small, easy to perform micro-tasks, so that people can work on these things without getting bored, and

    (c) allows multiple people to work on the same task, thus enabling easier cross-checking of the (noisy) results.

    Andy, keep posting such articles about Mechanical Turk!

    Thank you for this. I feel like an idiot for not thinking of Mechanical Turk for this hour-long audio transcription I’ve been doing myself.

    Andy – You had better specify that Amazon Mechanical Turk does not support requesters outside of the U.S.A. 🙁

    Very disappointing, but I thought it might be best to inform any of your international readers about this before they expend any time on it.

    Another good reason for shorter clips: automatic timestamping. If you seed 1 minute clips, you’ve automatically got one-minute timestamps without explicitly paying extra for that.

    quote from Andy:

    Amateur transcription is opening up a new market for people that wouldn’t have used the services otherwise, and people are getting paid for it.

    Is “amateur job” or “amateur market” the new justification for circumventing minimum wage requirements?

    It’s an interesting question. Legally, minimum wage doesn’t apply because workers are classed as independent contractors. But let’s look at the ethical question instead… Is it unfair?

    Should all existing worker protection laws apply to someone who’s working for, say, ten minutes during their lunch break at their real job? Or a college kid goofing off in his dorm for spare change?

    Turkers have complete control over the HITs they decide to work, how long they work, and where they do they work. If they don’t like working 2 cent HITs, they can choose to only do $1.00 HITs.

    The people using Mechanical Turk are empowered to do whatever they like at any time. I just don’t think the same guidelines around minimum wage apply.

    I needed ongoing transcription for my PeepCode video projects and found a number of skilled transcribers at oDesk.

    I’m paying about $12/hour, with a 4:1 ratio of transcription time to audio length. The quality is great and I would never have done it otherwise.

    Professional transcribers have a special foot pedal that enables them to stop the audio or listen to an instant replay. It would be painful to do it without that.

    Andy: people working anywhere are empowered to do whatever they like at any time, including quit. There is nothing special about Mechanical Turk workers. By your logic, if minimum wage laws shouldn’t (ethically) apply to Mechanical Turk workers then they shouldn’t apply to anyone and we should let the job market determine the wages it will support.

    Michael: That’s not true. Independent contractors are far more empowered than employees, able to choose how long they work, the tasks they choose to work, and the amount they charge for their labor.

    Independent contractors (like Mechanical Turkers) are self-employed, so aren’t subject to employee protections like minimum wage. Do you think the government should prohibit the self-employed from working for less than the minimum wage?

    I run a small transcription outsourcing business in the UK .. and I do have CVs coming to me from people who will work for next to nothing (i.e. $3USD/hour)

    I have done some volunteer work in Cambodia (i.e. sweatshops) the people working there consider themselves very lucky (as opposed to toiling in fields during droughts) and often support their family on their wages.

    They are happy (enmpowered?) to be exploited cause they get some money, and the business is happy to exploit anyone who is in a desperate position.

    This is the place of “social responsibility”…

    business / people who do not exploit people just cause they can…i.e. the busimess /person will stand by their standards, values, ethics even if it means not making as much money. The underlying premise of the ’empowered worker’ (i.e open market economics) is that they have a choice. In some cases, that choise does not exist.

    We have a recent survey conducted on audio transcription pay scales (based on over 500 transcriptions wages) that we base Transcript Diva pay on…. .. even though we could take advantage of the more desperate job seekers… we don ‘t.. Have and look at this pay scale and compare it . I think it might surprise a few of you!

    Thanks for that Alan.

    Actually the pay rates appear to be ranging from $1.80 per hour to £3.60 per hour for the people actually doing the work (see here for a review of casting words transcription payment structure).

    …. compared to average US and London transcription service wages being paid this is a pretty serious (cue sound of person coughing) ‘saving’ Amazon is making.

    I know this is a cheap service and therefore good, but would you feel good about supporting your supermarket if they were employing people for $1.80 per hour?

    Isnt there is something wrong with this?

    I just checked out mturk by spending a whole hour checking lists of company addresses, for which I got paid the grand total of $1. Only the truly desperate will work for that kind of money, and as an mturk requester, I’d feel a bit uneasy about exploiting people’s financial desperation.

    There are some fun assignments that pay reasonable well, like getting $5 to create an example animation on a new DIY animation website, but I see mturk strictly as something for students with some spare time who would probably do the tasks even if they weren’t paid for it.

    The bottom line is, no matter how well-intended mturk is, most of the requesters are businesses trying to get a job done as cheaply as possible, with a total disregard for the people they employ; it’s the nineteenth century in digital guise.

    Sounds great, but of course the downside is that you’re giving potentially secret information to someone you don’t even know the name of (and whatever agreement they’ve agreed to is worthless). This might be fine for unimportant transcripts, but for transcripts of interviews with famous people or of buisness phone calls or meetings, it would probably be unwise to use it.

    That said, it’s nice to know people are out there willing to be paid very little to do boring things.

    I actually am right now slogging through transcribing a 52-minute podcast for $6. At my current rate it will take 8 hours unless I speed it up. I’ve actually been considering going around town offering my services for $5/hr or $3/hr, I’m in Pittsburgh PA and can’t find a job. I’m educated as well as skilled in many trades, and I know there’s regular money to be made all around me, but opening the spigot is proving impossible. I just got back on MTurk, and remember now that I need to do a few HITs before I realize that it’s not for me. There’s an interesting threshold being tapped here, where someone is willing to work for so little given certain parameters (such as the instant gratification of completing a 2 cent HIT). I’m not paying for internet right now, but I am paying for my electricity, and this won’t cut it! :]

    I think my fellow transcribers have struck a chord here. People are on Mturk for two reasons; fun and experience, or sheer desperation. I have 2 years of in-house transcription experience, but there are literally 10 transcriptionists fighting for 1 contract. The local job market is horrendous, to the point that you can’t even trust your resume with the receptionist because he or she is scared that you will replace them. I currently have 3 contracts, but none of them has consistent work. I recently transcribed a portion of audio on Mturk for peanuts, only because I needed a specific, recent transcription example. Honestly, I am not too sure that the people requesting these transcriptions are getting quality work. Each and every document I return has been thoroughly proofed, researched and spell- checked. Have any of these transcripts been compared against the audio for consistency? Or are you looking at a 10 page document and saying “Wow, I got all this work done for next to nothing!”

    The author said himself “I hate transcribing”, yet he was willing to send that work to someone else. He had 2 hours of work done for $15.40; so still over minimum wage, but who can feed their family on that? I’m trying not to be sensitive, but I have the distinct feeling that people are missing the point here. I dont feel good about saving money if I don’t tip a waiter. Sure, it isn’t required by law, but wouldn’t you feel bad? I understand that it is not an exact comparison, but the underlining sentiment is the same. I personally hope that things begin to pick up, but I fear the worst is ahead.

    We posted a 20 minute audio file… right from an mp3 recorder to my website. Then just put up a link to the mp3 and asked for the text to be formated in the format:



    (for man: ; woman )

    Took all of about 5 minutes to post up to HIT. We got back (overnight) amazing results for $10.

    I suggest taking no time to divy up your audio, etc. Spend that time reviewing what the guy (or gal) did.

    I’m a stay at home mom, and work on Mechanical Turk in my spare time. The transcription projects are my favorite, although I don’t care for Casting Words. As a former paralegal, I take great pride in the accuracy of my transcriptions. It’s true that the pay isn’t stellar, but for a busy mom, it’s nice to be able to choose only the tasks that I have time for. I often get paid $4.00 a day for one podcast or lecture, and that is fine by me! It’s not supplemental income, but it’s play money! (or right now, Christmas money.)

    You went to a lot of work. Very clever, but I have a transciption service for which I charge the same price: $30/hour. My service is more personalized and you don’t have to spend two hours building a web page. 😀

    I just came across this site, and for those of you who are talking about it not being worth it and being like outsourcing to other countries, I have a few words.

    First of all, since it is amateure work generally, you can’t expect to pay as much.

    Second thing, since it is more like contract work, rather than an actual full on job, rates aren’t expected to be as high. Since the people doing it are anonymous, you don’t sit there and have resumes sent in and interview the people working for you, so there is no need to pay much for such quick easy work that someone can do whenever they have time and feel like doing some easy work.

    Thirdly, for someone like me who is visually impaired and on limited income via SSI, I have never found anything work from home no experience that even compares to this. I’ve joined a few GPT sites or whatever, but the pay is miniscule compared to this. I could earn with this site in one day what I earn on myLot in a month, and I have no experience, so for me it is worth it, especially since finding good work, with the economy, and the always existing discrimination has been next to impossible for me. If I can pull in even $15 per day for a half hour file and approximately 3 hours of work, or even a little more or less, as I haven’t actually tried it yet, but if it really works as the review and other commenters describe, then it’d be more than worth it. $15 a day for even 4 days is an extra $60 per week, $240 per month, which for me is good money to add to my low benefits. It’s not enough to take me off SSI, but it’s enough to supplement it without them taking any away from me. So, that all being said, I would say that this type of gig is more than worth it at least for someone in my situation. If it really works as described, I for one would be eternally grateful for the opportunity.

    Okay, sorry, long rant over, but I felt the need to comment here because I don’t think it seems as bad as it’s made out to be unless it’s scam, but being that it’s through Amazon, I’d wager it’s not. Now, off to go try it. Might pop back here sometime and update.

    I started out as a turker but have turned into a requester (someone who pays for mturk work). The quality of responses varies widely, especially when workers have to write in freeform English. I need people who can competently write in modern English, so I did a little test. I uploaded the four Gospels from the King James Bible and asked the workers to rewrite individual verses in their own words. Click on my name to see the final results.

    If you are going to do a lot of the same task repeatedly over time, it is best to qualify workers and only accept the ones who do quality work. It is easy to create qualifications through the command line interface.

    Those who are quick to cite minimum wage as an ethical standard might take note that minimum wage is an ethical compromise itself: the perceived virtue of improving minimum welfare is accomplished through force. As such, the fruits of it are usually not what is desired or expected (e.g. automation, facilitating outsourced and migrant work, doing without, etc.). Amanda’s comments are a great example of the kind of low wage market that has been needed but disallowed through a wealth sharing scheme based on force.

    When some morality is regulated, a natural result is for people to substitute the regulated morality for their own otherwise higher standard.

    I work for a transcription company, and I think the reason why professional transcriptists get paid decent wages is reliability. Clients pay premium for reliable services, and amatuers or hobbiests even if talented are not dedicated and consistant so don’t qualify for the high wages. However, turk lessons the risk by spreading the job across many workers which muddles the quality issue slightly

    Though you have found a great price for transcription, I would be amazed if the quality was good. With regards to audio transcription, you will get what you pay for. Of course there are times when you have very clear audio and may be able to get a good quality transcript in return, but why chance that? There are so many good quality transcription providers out there that still offer competitive pricing.

    It’s funny how most of the people objecting to amateur transcription services are the people that work for professional transcription services, like Melody and Steve.


    I am not objecting to it at all. All I was saying is that, almost always, you will get what you pay for. I have witnessed many situations where clients ended up having to pay twice because of the poor quality they received the first time. However, I am all about saving when and where you can. Of course, I would love for everybody to come to my website for their transcription needs, but if they are happy somewhere else, then that is also wonderful. I am sorry if I came across in a bad way with my post. Have a great day.

    You pay peanuts, you get monkeys!

    But, on the other hand if you want quality (as opposed to quantity) then you have to pay a little extra for a reliable and experienced transcriptionist.

    I don’t think the gap between amateur and professional transcriptions is as wide as you’d think. There are definitely more errors in the amateur transcriptions, but because it’s so cheap, you can pay for a second pass through a second worker to clean up the original, and it’d still cost a fraction of a professional.

    Very interesting concept, which I’m testing out as we speak. But from a usability and management standpoint, I can already say that the MTurk requester site has a lot of room for improvement. As far as I can see, there’s no precise way to even tell if one of your own HITs has been accepted by a worker. I have two batches labeled as “In Progress,” but it’s very unclear if that means they’ve just been published or if someone is actually working on them. The only reason I believe someone is working on them is because I no longer spot them on the HIT list page, whereas I did for a time after I published.

    Aside from not spelling out if a HIT has been accepted, the Manage page also doesn’t tell me WHEN the HIT was accepted and how much more time the worker has left. It would also be nice to know how many workers started the HIT and then ultimately rejected it, so that I might better optimize my HIT config in the future, but this information is nowhere to be found on the Manage page, either. I did get an automated MTurk status email that had some of this information, but why is it nowhere on the site itself?

    Lastly, how will I know when my HIT has been completed or I have some results to view? Will I be emailed? Will I have to blindly keep checking the site myself on a regular basis? I have no idea, and again this is not spelled out anywhere. There are almost no account settings for requesters to configure, and the documentation is laughably sparse.

    I realize it’s in beta, but the site certainly needs a lot of work from usability standpoint. Throwing up a little more documentation would be nice, at the very least.

    I like this idea that Andy has initiated using Mturk for transcribing. I would like to complete a general transcription (one speaker very clear completed using express dictation) and see how Mturk does. Please inform me how I add the mp3 file on the Mturk request page. My initial mp3 will be 47 seconds duration.

    Keep in mind that I am keeping this as simple as I can. If it works then I would be sending possibly 5, 10 and 15 minutes audio clips to MTurk.

    Thanks for your help. I will let the group know how it came out.


    Jeff: You can’t upload MP3s to Mechanical Turk. You need to put the MP3s somewhere publicly-accessible on the web, and then link to them in your request.

    It seems the system is already flawed then if its not simple just to upload a 47 second MP3 sound file for transcription.

    Mel: It’s still quicker to get started with Mechanical Turk than your company, which requires filling out a 27-question quote request, waiting while you “endeavour to email a quotation within 24 hours,” and then requesting and filling out a “Project Booking Form.”

    If you can’t figure out how to upload an MP3 to your own webspace or a service like RapidShare or Megaupload, then this method isn’t for you.

    Andy: I think not! Its a lot quicker than having to mess around with a 35 min file, splitting it up into small separate clips and then uploading them all separately.

    With our system you upload your whole 35 min file (which will take around 3 mins)It then gets transcribed by one transcriptionist and a complete transcript is returned (which, as its been transcribed by a professional will need very little, if at all, editing)….Done!

    With your method you have to ‘prepare your audio’ by splitting your files into several individual sound files,’Design your HIT template’ then ‘upload the data for your HITS’ and then finally ‘Publish your HITS’.Your method will return several documents which have been transcribed by several different people, all with different transcription styles, which you then have to spend time cutting and pasting into one transcript. Seems a lot more faffing around than just filling out a simple Project Booking Form and uploading one single file to me!

    As we’re a professional company its only prudent for us to ask our customers to complete a Project Booking Form (which will then cover any other work undertaken for them) so at least we know who they are and where they’re based for invoicing purposes.

    It’s a totally different experience and, personally, I’m willing to sacrifice quality for lower price and more control. With Mechanical Turk, I can get started immediately and know exactly how much it’s going to cost me.

    With a professional service like yours, it’s a black hole. I don’t know how long it’s going to take you to respond to me or how much it costs, since that information isn’t disclosed anywhere.

    I started my task in the middle of the night, it was done in three hours without waiting for a quote or approval, and my 36-minute recording was transcribed while I slept, in less than three hours, for about $0.40 per minute of audio.

    Were the results as clean as what your company could provide, given enough time and money? Absolutely not. It had some errors, and was returned in several rows of a spreadsheet instead of a single Word doc. But the quality was definitely good enough for me and saved me hours of work.

    Andy: Its hardly a ‘black hole’. I guess we don’t readily display our prices for the same reason you probably dont freely display your own hourly rates anywhere Andy (and I did go check). Its also purposefully done to start some initial dialogue with a prospect visiting the site to understand exactly what they want – it could be transcribing a 3 day conference on nanotechnology to dictating a short piece of general correspondence (the whole point of getting prospects to fill in a simple quote form when they visit our site).

    We charge different ‘price per recorded minute’ rates which reflect the extra work & time involved in identifying say 10 separate voices in a 60 minute focus group compared to just a single voice slowly dictating a couple of paragraphs. Where the material involves technical terms or industry-specific language we also go to some lengths to research on the internet the terms used to ensure we get the transcript as near perfect as we can.

    Using different time zones is also an excellent use of time – i.e. while you’re fast asleep in Portland (dreaming of that perfect transcript which will be waiting in your inbox first thing in the morning) then its actually our working day over here in the UK.

    I guess what it boils down to Andy is ‘different horses for different courses’.

    One thing’s for sure…you’ve ensured an excellent Google rating now. When someone searches for ‘audio transcription’ – you’re about 3rd/4th on the page. Well done!

    Dear Mr. Waxy,

    I tried your html text. It seems that the embeddable google mp3 player no longer works. Could you kindly confirm. Thank you.

    Those of you who think it’s slick to exploit Americans who are trying to feed their families by working on AMT are the reason this country is in the crapper to begin with.

    Well done you for not being forced to work for hours on a hit that pays pennies to a dollar or two. If only I could be so lucky.

    It was my understanding that the exploitation of human beings was morally reprehensible. In fact, if I’m not mistaken that’s why so many Americans are dying in a useless war as you sit here and pat one another on the back for saving money at the expense of another’s dignity and sweat.

    You all sicken me. You don’t deserve to be in this country. Hell, you don’t deserve to be on this planet.

    You should be ashamed of yourselves!

    Works fine for me great!

    Andy wrote:

    “If you can’t figure out how to upload an MP3 to your own webspace or a service like RapidShare or Megaupload, then this method isn’t for you.”

    If you’re looking for a free filehost,

    Check out

    upload up to 1GB, No registration, No adds, 100% Free

    I transcribe for Casting Words and have done other transcriptions on M-Turk. CW is definitely worth the time and I make more than minimum wage assuming there is good work up and admittedly sometimes there is not.

    I am a skilled, organized and fast typist. I have been with them long enough to know the job and the style that they use.

    The reason I do this work is exactly what has been mentioned. I am totally free to work as much or little as I wish. I can choose my jobs. If I get into something and change my mind I can return it.

    The system is set up on a point system so the better paid jobs are going to be better transcribers. All transcripts are “graded” by one or more workers, improved if needed and edited. Then the staff at CW must also review them before being returned.

    If people are getting back transcripts with many errors, they probably are sending in poor quality audio. Believe me we get a lot of that! I do a lot of editing and I can guarantee that what I send out of clear, normal American speach will be 99% correct.

    Andy wrote:

    “If you can’t figure out how to upload an MP3 to your own webspace or a service like RapidShare or Megaupload, then this method isn’t for you.”

    Have you checked out upload mirrors?

    Please note this service is very easy to use, uploads are faster then most other sortlike services, they support up to 12 file hosts and are 100% add free!

    I write the Health Business Blog and have been using Casting Words for my podcasts for a couple of years. In the past few months they’ve become totally unreliable –not sure why but they basically stopped transcribing my podcasts.

    In my experience the CW transcripts are excellent. I usually spend about 15-30 minutes editing a typical 20 minute podcast once I receive the transcript. The results are great.

    I’m going to try using mturk myself and see how it goes and may also give podclerk a try after reading the comments here.

    I just use Soundflower and MacSpeech Dictate. Pipe the audio of the file into Dictate and transcribe in realtime. Faster and cheaper than your method. 🙂

    Podclerk’s rates are now the same as CastingWords — $0.75/minute for standard turnaround. It’s not clear from the podclerk website whether they provide any of the review/QA services that CW does.

    Like Oliver, I have worked for Casting Words (and actually did some of the Gamasutra HITs).

    Please forgive my shameless self-promotion, but check out my website the next time you need accurate, confidential, ready-to-go transcription at very competitive rates.

    If you really want a hands-off transcription, I’d recommend doing another pass for considerably less, asking one or two turkers to review and clean up the original transcription. The funky characters are smartquotes, ellipses, and other special characters generated by turkers working in Microsoft Word and pasting in the results into the form field. You might want to ask people not to use Microsoft Word, in the future. (Notepad doesn’t make the same mistake.)

    Transcription services Transcript Divas – currently use an online program that converts word docs into HTML tags for clients wanting to post directly onto their websites.

    This add on to the standard transcription service is available at an extra 10p/minute. This is reproofed at the end of this – but as yet – we have not had any issues with the system.

    We had initially considered using MTurk as the backend for our transcription service. But found it difficult to tailor to the transcription process we had in mind.

    The system we use now has multiple stages. We split up the files into smaller chunks which are then picked up by our transcribers. Each transcript is then reviewed, speaker initials and timestamps are added and then they are finally collated.

    We’ve gotten pretty decent results with our system so far with some very satisfied customers.

    More about our process at

    Castingwords was great at first, but now I’m not so sure. One of our clients told us recently that Castingwords was charging more than $2 per audio minute for decent quality clean audio. I think that’s way too much even for a company this established.

    We tried some of our files on MTurk straight off, but well, disappointing results. I mean we’d rather take more time and transcribe those files in-house rather than spending double amount of time proofreading them.

    Well, our rate is $0.90 per minute for decent quality audio. TAT is 3 days for one hour of audio.

    This could be a life saver for one of our client’s who has a tight budget and needs a very sketchy transcription to get the gist of the text.

    I’m just wondering whether the HITs are for English speakers only? I’m looking for a Korean transcriptionist. Does anyone know whether Mechanical Turk Workers consist of non-English speakers too?


    I have logged in Mechanical turks just recently. I want to work as a transcriber in Mecahnical turk but how do I go about it if I don’t have an account where I can be paid money. Is western union money transfer possible? tell me how to go about it. I will greatly appreciate it.

    If all work from Castingwords is subcontracted and transcribed almost 99% through Mturk, I wonder if there is any data confidentiality agreement that they’d be willing to sign.

    Dear Andy,

    I like MTurk, but I was wondering if you could tell me other sites just like that. I guess I’m wondering: Is there an Indian or Chinese or PHilippine version of MTurk, (where the Turk worker is not restricted to American)?


    A direct email back, aside from posting your answer on your blog, would be much appreciated!


    I am in the midst of writing a book and I have recorded my thoughts as a way of getting things down as a foundation for expanding. Some times things are a little broken, where I change thoughts mid stream so to speak. What things should I bear in mind while seeking a transcription?

    Any suggestions as to what service to use? Mturk or others?

    Any help would be appreciated.




    You have probably already approached transcription companies regarding your book project, but a few suggestions that may help you, and others in the same boat, would be that you have a clear idea of how much editing you want the transcriptionist to be responsible for. They can do a verbatim transcript where every word you say will be transcribed, including um, you know, and if you have any verbal tics such as starting each sentence with “And….”

    A modified or clean verbatim will remove only the previously mentioned words, but if you start an idea and trail off after several words, that will be included in the transcript. You would be responsible for editing any of those types of things out yourself, or leaving them in if they helped make a point.

    You may also specify that the transcriptionist create a completely readable version — basically what would amount to a ready-to-publish document, although with a book you should probably have it professionally edited. Trail offs will only be included if they add content or meaning. Sentence structure and grammar and word tense will be corrected as needed, but only if it can be done without changing exactly what was said.

    Again, the amount of editing you trust the transcriptionist to do is up to you. I think you are safest going with a modified verbatim and cleaning up the final transcript yourself to make sure that nothing important is missing.

    Sounds all fantastic – I am currently facing the horrible horrible task of transcribing interview data for my PhD thesis. Alas! Mechanical Turks is only available to US residents and I am in Australia. Anyone willing to help me out here? I have a paypal and an Amazon acct.


    Hi, for those who are looking to earn more money online you need to look into a superb new guide to creating your own Info Products. Most people start with affiliate internet marketing, but this technique is so much easier and more profitable (and I find it much more rewarding as well, creating my own products). The best thing is, rather than making an attempt to influence folks to buy someone else’s product, this information shows you how to generate profits by selling people precisely what they want! It’s a lot simpler, and way more profitable! Click here to find out more … but act fast, as they are closing the doors on the 16th August to ensure that the strategies don’t get over-saturated

    VERY useful post Andy, thank you.

    I was pulling my hair out trying to work out how someone else had, seemingly managed to have an MTurk upload an annotated Word file when all I could see was .csv options.

    Thanks again.


    While I have no problem with trying to finish a job timely and for somewhat less money, AMERICANS deserve to make a living too. I am so tired of people trying to find something for nothing. No wonder America is short on jobs and unemployment is horrible. Everyone is taking and giving their work to India, Pakistan, China, or others. When are Americans going to wake up and see that quality is important, that paying a bit more for a product that is American made is best for America and the people who live here. MTs go through tremendous schooling and continuing education to perform their jobs correctly and we deserve to make a living too. Paying $15 something for three hours’ worth of work isn’t even minimum wage in America and skilled people shouldn’t have to work for sweatshop wages. Wake up America. Our economy will NEVER jump back if only CEOs and greedy companies keep sending jobs overseas when they are wealthy enough. It’s maddening when someone has built a business for 16 years and has another 20 years of experience in the business. As a small business owner I have bills and certainly cannot stay in business making $5/hr. I have started only purchasing and supporting American-made goods or American-based companies.

    I work as a Turker. People should upload to MTurk directly instead of services like CastingWords.

    CastingWords doesn’t pay for over a week. And if you’re working MTurk for an average of $2/hour like me… you need money NOW. I’m a U.S. resident. I’d love if people made their Requests “US only”, and excluded foreigners.

    So people – please submit your audio files to MTurk… I don’t want to leave my mom’s basement and get a real job. Thanks.

    Mechanical Turk is nice because, instead of hiring starving college students and making them work for class credit (aka zero money)… I can now pay them $5/hr, and they can type with one hand while they masturbate with the other. Sounds like a win-win.

    This thing is reminiscent to the ‘open source’ translation methods where people volunteer themselves to help on projects and gain experience. Except, they get paid a bit of cash. Not a bad deal.

    Thanks for the nice how-to article.

    This was very helpful. I needed a 20 minute video transcribed in a pinch and had never used Mechanical Turk because something about it had me unsure if it was a good use of time.

    your post was well-written, well thought out and so step-by-step that it actually felt like fun to follow your instructions.

    Thanks again!



    “For the worker’s convenience, I also added an embedded Flash player for the MP3, but this is entirely optional.”

    How do you embed a Flash Player? I’m pulling my hair out over here.

    After recommending this article to people for years, I finally tried it myself tonight. Thanks for saving me time and money.

    Thanks for the how-to. I was looking for an easier faster way to transcribe (I use a service already but I don’t have 3 days to wait). I just uploaded 24 files at 5 minutes each and I’m excited to see what happens!

    I recently became a Turker. Having been laid-off from a decent job in 2007, for the past three years I have been working a long-term temp job doing medical assembly for little more than the minimum wage for the geographical area I work in, because that is the only available work out there (I drive about 65 miles each way to/from work.)

    I need extra money as I recently went through bankruptcy and am currently having my house foreclosed. And because of my unusual hours at work, it would be very difficult to find a part-time job at a brick-and-mortar business.

    Although Mechanical Turk is extremely low paying (I made about $2.00 yesterday for 4 hours of work,) it does allow me and others to work from home on the projects we accept. I have been looking at the idea of getting into transcription on MT, as that seems to pay better than most of the other HITs, not to mention that I do well with grammer, spelling and punctuation, and type around 50+ WPM (not super fast, but not slow either.)

    Certainly we do not get paid minimum wage, but then again I know other people working as independent contractors in other industries who also are not making minimum wage right now, but they also keep doing it because of the current economic climate.

    So all I am asking here is please keep in mind that for many people, it is low paying HITs on Mechanical Turk or nothing. Of course we would rather be making a lot more money doing this or even something else, but atleast for now, with the job market the way it is, this is what we have to rely on.

    You don’t type it up yourself these days, you talk it up.

    I used to use Dragon Naturally speaking years ago but it was rubbish.

    I was persuaded to try it again recently and am pleased to say it’s come on leaps and bounds.

    Try it.

    In fact play your audio to it and see how it copes.

    Hey, Andy. I had been using MTurk before in most of my thesis works but I would rather keep using this site called Fiverr now. I’m not sure if you’re familiar with this but it’s pretty straightforward. There’s this transcriber, I don’t know if she’s a girl or what but she has been giving me awesome transcripts so far. And she’s fine with $5 for a 20 minute transcript, yet this transcriber has the most number of good feedback.

    I’m not really in this business, but looking over the information and those early comments about “CastingWords”…

    Looking at the list of MTurks, most of the transcriptions’ creators are Casting Words as the “Requester”.

    So, it’s worth mentioning, that you’re not hiring CastingWords as a middle-man to using this service, since clearly they see the value in using an open competitive service like this.

    Thanks again for this, Andy. It’s saved my bacon numerous times. The funny thing is, the more I use it, the more I find myself paying per 5-minute mp3. Some combination of appreciation and guilt, I suppose.

    Thanks so much for the posts and ideas, Andy (and everyone else with other productive examples)!

    The point here, people, pay a little and spend an hour proof-reading/editing the transcriptions yourself…

    You are still saving a heap of time… and money.

    I work on Mechanical Turk all the time. I don’t see a problem with the pay rate. Any time you work as a piece worker, you tend to make less than minimum wage; just ask any tailor. But the fact that you can do several tasks (depending on what they are) in less than an hour makes the low pay worth it, because you can pick up spare change in your free time. It’s not supposed to be a primary source of income.

    And if you’re worried about quality control, you can filter out the less experienced people be adding qualification minimums.

    Thank you so much for this! This was an excellent, easy to follow and comprehensive tutorial. I am very appreciative of your efforts in putting this together!!

    Thanks for the step-by-step write up. I’ve been putting transcribing some audio for quite awhile. I already have an MTurk account set up. now to just give it a go.

    Hey, thanks for a great tutorial. I use Amazon s3 buckets regularly, but haven’t used the mechanical turk services option.

    Embeding the flashplayer for the worker is a nice touch too!

    I suppose, anything that makes things easier for the outsourcer.

    Thanks @Andy! This just saved me an hour or two. I just posted 5 3-minute videos that I need to transcribe at $2.00/HIT. I actually might be able to watch the Super Bowl now!

    Cam (@camcollins)

    Hi guys,

    There are workers out here who do not charge a fortune for quality transcription, it’s knowing where to find them that’s the problem! I have worked on MTurk in the past. but have gradually been able to build up a few clients of my own. I am still searching for more, so if you know of anybody who is looking for a dedicated, thorough, dependable transcriber who is willing to stay up through the night to meet deadlines at no extra cost, then please get them to email me at [email protected], where they can get quality transcription services for 30p per audio minute. Quick delivery and payment through paypal.

    I cant upload my audio’s after following all your recommended steps for the use of Amazontech…

    Ive designed the template, thats how far Ive gotten. Saved and ready to publish. When I select Publis, and the project, it takes me back to my work–no sign or tabs where I can upload any files.

    Now, my files are wav. files on my desktop, can I simply upload from my location?


    Invaluable tutorial. Saved me much hassle, will definitely use again. Thanks for the writeup.

    AWESOME!!! Except some of the Workers were clearly using voice recognition software, and provided ridiculously unacceptable transcripts. Thanks.

    Very nice – I’m working through a fairly large text classification project right now. Hadn’t thought about MT until I read this article.

    One issue we’re going to have to deal with is classification of text by “meaning” – basically having a human do a first pass at a tagging schema… (which we will then use as a training sample).

    Fantastic! Just did 38m for $16, and quickly did a read through to fix minor errors – only had one 5 minute chunk that had more than a few typos. I think I’ll try 3m for $1 and see if I get any responses, the 5m for $2 got snatched up immediately.


    I tried the link to podclerk posted by Dan and it leads to a one page wordpress site. You might want to check that out.

    I really don’t agree with much of the discussion on these comments. You are all correct and wrong at the same time. Of course it isn’t fair that someone has to work for so little money, but compare that to some countries in the world and the people there would love to earn $6 ph. However, if you are going to contract a professional transcription company or transcriber you have to pay the dues for the quality of the work. My company receive CVs from qualified typists every day seven days a week and we would only hire a couple from every hundred because of the quality they deliver and what our clients expect.

    You pay for what you get, basically.

    I mentioned something liek this in a previous comment I posted on here, but was disappointed to see it isn’t appearing now. Maybe someone could look into that please?

    If 99% of typists out there can’t transcribe well (transcribing is not just typing or audio typing) then software or piece workers aren’t going to be able to do it either.

    I’m a senior, used to do interviews and my own transcriptions via micro tapes. Getting back into the game I find this series of comments extremely helpful, and confirming for the direction I want to go. Thank heavens for community think! And thank you Waxy for your work to this end.

    PS I am a autobiographical ghostwriter/teacher for regular folks who have lived amazing lives. I love my work but can’t do the transcribing anymore without killing myself. Back and leg problems.

    The MTurk site is paying people unfair wages. Not everyone enjoys the work or likes the rate of pay, but they are underemployed and trying to survive.

    It would be nice if people could consider…would you work for $1-3 and hour?? Not likely. It disgusts me how others take advantage of the misfortune of the underemployed (with a college degree and student loans to pay back! Yes, with responsibilities and families too, and trying to do the right thing). Hard working folks that are finding any means possible to pay the bills minus taking a government handout.

    Put yourself in their shoes and stop paying SWEATSHOP rate of pay! It is not right…………Plus, all you cheap and unethical people could just use Adobe software to convert the audio to text…..figure it out. Ugh……..

    I’m new to Mechanical Turk and have noticed these jobs floating around. They are mostly a waste of time because they pay is very low and the audio source is usually messed up in some way. There will almost always be static, or the speaker will have a strong accent, or it was simply recorded in the bottom of a well using a potato (the recording device of choice circa 1951.)

    Most of the serious turkers seem to avoid jobs like these and do the small hits where you transcribe text or take little quizzes. I think the goal for most people is to make around $3 an hour. So yes, we’re working for slave wages. The point of turking (for most of us, anyway) is to make a little money on our lunch break, or in between classes. Tasks that only requires a few minutes at a time.

    That’s the other reason nobody wants to do these audio jobs; they’re a huge time sink. To add insult to injury, the results tend to be highly open to interpretation since the quality is usually low, so you get a lot of rejections (where they pay you nothing but still get the benefit of whatever work you’ve submitted) or they dock your pay significantly like the comment above said.

    It’s just too high risk, too time intensive, and too little pay.

    Yeah, I can’t blame you for getting frustrated. I always cut my audio clips into segments of five minutes or less and offer a fair rate, but many people don’t. It’s insane what people think they can get for pennies.

    This post and the majority of responses make me sick to my stomach. You admit it takes 8 times the length of the audio file to complete, yet have no qualms paying $2, “I chose $2.00 per five-minute MP3” $2.00 for 40 minutes of work!?! Then you brag about it being fast and simple, completed while you slept…how do you sleep, knowing you are taking advantage of others to save a few dollars? This is the same thinking that created sweat shops, and to deny it is only an attempt to justify something you KNOW is wrong. Just because something can be done, doesn’t mean it should be done.

    I spend more time deleting comments in this entry than any other in the 10+ years of running this blog. Most of them are spam from voice transcription services, hoping to squeeze some Google juice out of this post, since it ranks so highly for terms like “audio transcription” on search engines. I delete most of these.

    And then there are complaints, usually anonymous, like yours. Yes, Mechanical Turk pays poorly. No, I don’t think it’s wrong, for all the reasons I’ve stated in the comments above.

    People are primarily using MTurk in their free time. This is the equivalent of finding loose change for them, while doing otherwise mindless tasks.

    MTurk doesn’t need to pay minimum wage because it’s free from time, place, background, and bosses. You can do a task in two minutes between classes, or while smoking pot in your underwear. You can do 2 minutes a week or an hour a day.

    One of the great things about Mechanical Turk is that workers pick and choose exactly what tasks they work on, and if they pay lower than you’d take, they’re free to skip it and move onto the next one.

    If you’re unhappy with the pricing of a particular task, you should do the same.

    And with that, after five years of discussion, I’m closing out these comments. If you have anything useful to add, feel free to email me.

Comments are closed.