Updated content policies: Audio and Video Standards

[RyanJaress]

Hello Instructors! We are pleased to announce two exciting policy updates straight from our Trust & Safety Team. These changes have been driven by valuable feedback from our esteemed instructors like you. Get ready to embrace more flexibility in your course creation process! 🚀

First, we are now accepting courses created with text-to-speech software. We have consistently heard from instructors about their interest in using these programs, and recent advancements in the technology meet our content standards. Usage of text-to-speech tools in your course creation and editing can offer enhanced accessibility and flexibility for you as well as your learners.

While we’re excited about the new opportunities this technology offers, usage of text-to-speech will need to follow a few guidelines. For example, if the audio sounds robotic, contains noticeable mispronunciations, or is otherwise distracting or in violation of the Audio or other Quality Standards, it will not be accepted for publication. As always, courses that perform well in the marketplace may be considered for curation into Udemy Business as well. You can learn more about UB curation here.

Second, we are designating a Course Setup Lecture, as a space for instructors to share required course resources which would otherwise be restricted to the Bonus Lecture. We are committed to keeping course material free of promotional content but are optimistic that this additional lecture type will help learners get discounts for relevant and course-essential external tools and resources such as exam vouchers and software discounts.

To learn more about these updates, please see the Audio and Video Standards and Course Setup Lecture policy pages, which have updated our policies effective today.

[FrankKane]

It would be nice if the policy expressly prohibited sampling the voices of other instructors for creating their AI speech.

[ThomasMitchell]

This is sad news.  

[Veasna, M.]

This is a good news! Especially for non-english speaking instructors. 

Thank you for giving more space for us on Udemy!

[NargisWalji057]

I used text to speech in my course '

Teach preschooler/Kindergartner Math -with Montessori Method"

and was asked to change that. i just did that .

nargis

[GauravSingh194]

This is sad news

[MirkoNasato]

I have a question about the Course Setup Lecture @RyanJaress.

Do I understand the "What shouldn’t I put in a Course Setup Lecture?" section correctly that we're still not allowed to say something like "this course on Algebra requires knowledge of Arithmetic. If you don't know Arithmetic yet please take this other course first (link)." (Where Algebra and Arithmetic are obviously just an example.)

If so that's not very useful unfortunately, at least for my own courses.

[ANAC]

I am delighted with this news. As a teacher with a progressive neurological disease that affects my speech, I am able to use high quality voice cloning. You really wouldn't know the difference - even my partner can't tell the difference.

Being unable to speak fully has prevented me from creating new courses as I can't afford to pay voice over actors. 

So this move creates accessibility and i am pleased that the standards are high in preventing robotic, monotonous speech.

[IrfanSharif]

@ANAC which tool are you using for such voice cloning?

[ANAC]

I was trialling Speechki but they went under. I don't find Descript so great, but some people find it works well. I flip between softwares to find the best, but I'm always on the lookout for the best result. However, it's not 100% use. It's a fall back only and I record my own voice on good days.

[durrani276]

This can be abused by the public by making complete courses using chatgpt written script and using such softwares.

I think this is a bad step towards the platform. 

I am hopeful this gets reverse soon as it will affect hardworking instructors working days on creating a single course. 

[ANAC]

But what about the good teachers, producing original content who develop a speech issue? They should not be discriminated against and they still are working days and weeks on creating their courses.

It is disability discrimination to exclude their teachings because their speech is affected through no fault of their own. 

Should Stephen Hawking have been banned from teaching students because he had a synthetic voice? NO. He changed world thinking on physics and the universe whilst using artificial speech for most of his life.

As for chatgpt written scripts, that's a different debate and shouldn't be confused with accessibility.

[ChrisBankes120]

@durrani276 

Out of curiosity I asked ChatGPT how one can differentiate your course against a course created by someone "by the public by making complete courses using chatgpt written script and using such softwares."

ChatGPT told me in detail what to do.

The problem is the students who enrol but DON'T watch (about 82% of my 212,000 students have never watched anything) . If they're not going to watch they'll just enrol on the course that appears in their email etc and if there are 100 'chatgpt courses' plus yours (even if yours is differentiated etc) then you'll probably lose out.

But then - and this is just an example - in 45 minutes (or 2 minutes fast forwarding) with no prerequisites I can '..Be able to practice professionally as a Child Psychologist...Be Fully Qualified As A Child Psychologist...Fully Accredited Professional Child Psychology Diploma...'  then authenticity is perhaps not the key criteria here...

[ANAC]

Interesting indeed and yes, evven with fully human made, created, spoken and presented, so many don't finish the course! Lots to ponder thanks

[durrani276]

@ChrisBankes120 Yes very much true. But the fact is that I can be anyone. I can also even clone celebrities voices and create courses in their names.

For example I can be MKBHD with tons of courses on youtube and content creation.

I can be anyone and then this becomes naturally unfair. Although udemy might allow it but morally its so unfair. Udemy should take proper action against the abuse of this previlege. Otherwise for special instructors like ANAC exceptions can be made.

[MarinaT]

Hi @MirkoNasato, 

Great question! I recommend verifying this with our policy team to ensure compliance with our guidelines. You can reach out to them via email at policy@udemy.com.

Thank you so much! 

[ANAC]

It can get quite scary but I am sure Udemy will have thorough checks. for example. On Amazon KDP and Youtube, you now have to declare if your work was created with AI. It's not prohibited but you have to be honest. I'm happy to say that sometimes on my own youtubes etc, I use a voiceover clone, upfront. It's only been in a couple of places, but it could get worse. Yet it takes just as long to plan the courses and in fact longer to create, if you have something impeding your recording ability. Editing time for example is exhausting. 

It will be interesting to see where it goes but I would have no issue with a declaration on submission of the work being ai in part, full or human.

[MarinaT]

I also wanted to address the question raised by @FrankKane here and provide an update to ensure clarity, particularly for instructors who may have concerns about voice plagiarism:

We wanted to ensure you that our teams will be on the lookout for any impersonation of both content and voice. There are additional questions and screenings that Trust & Safety would employ when there is an apparent voice mismatch to the instructor.  

Any usage of text-to-speech programs to attempt to impersonate an instructor would not be permitted and result in a ban. 

[Saisulaiman]

I welcome this development.

I personally like the acceptance of the use of text-speech software in course creation. The world keeps on changing so we've to embrace changes.  As a retired professional teacher,  I discover that even the online learning opportunity we're providing on Udemy and other platforms  is still new and unpopular among some conservative learners, they can't stop us from getting the benefits and conveniences that online learning entails.

My question is, can I combine my voice recording and the text-to-speech audio in a single course? I mean combining organic voice  (OV) with an artificial voice (AV), or simply, using OVAV  in appropriate places and ways within a course.

For example, I can use OV in my promotional or self-introduction video to let learners have  a feel of my personality characteristics, and AV in the instructional videos where objectivity, factual information, illustrations and use of diverse teaching methods, will significantly make a course useful and appealing to learners.

You can also contribute to this issue by giving more examples on using  OVAV  in appropriate places and ways within a course.  Thank you.

[ANAC]

Out of interest and completely just for fun, I created the videos for a mini course ( it was never published here or anywhere, but I wanted to play).

For each video, I introduced the learning points in own voice. The video lecture itself, used a realistic voice from elenvenlabs. It sounded perfect! then I summarised at the end in my own voice. The really interesting part was that for each subsequent video lesson, I chose a different VO from elevenlabs and alternated between male and female VO's. In total I had used four different VO's, plus me.

I then shared my series of videos in their learning sequence, with some friends and they all agreed it kept their interest and retention better. 

It was a small personal experiment but gave me a lot to think about. 

For today's online students, so many are hooked on the quick TikTok  culture of videos and some just like to read the subtitles only, or listen to audio in the car on the way to work. Keeping learners engaged is changing and experimenting was fun, but also informative. 

It's just food for thought and wouldn't work for many types of courses, but it was interesting to do.

[Saisulaiman]

Yes, combining organic voice (OV)  and  artificial voice (AV) wouldn't work for some types of courses. Courses teaching skills through demonstrations,  for example, require the voice and video of the instructor making explanations.

I like the experiment you've made.

[ANAC]

Totally agree. It works best where the video lecture is slides with a VO. I agree demonstrations need the presenter.

[Ali Project Manager]

As a new member of the community I found that the Course Setup Lecture is useful. Thank you

[IanBradley63]

Massively worried by this, very easy not just to abuse but basically copy someone's content straight off through the amazing work Udemy does with auto transcription. I'd hope to have more oversight and restrictions with these. It would be very easy to create a bot to create content that is just stolen work.

I understand that being able to teach in other languages is a plus, a non-native speaker can now use an AI learning model and then a text to speech to make course content, but this is only going to lead to subpar products. I am working with people to translate my more popular content into Spanish and Korean. That takes time, money and other experts to do, if I use AI then it won't be as good, won't translate the nuances and culture that the different languages need

[junderwood]

Like some others here, I also had a medical issue that caused me to have an ongoing speech disability. I've struggled mightily to produce acceptable audio and found that even when students liked my content, many would give me a poor rating due to my speech being slurred at times. This is a great help for those of us who have the knowledge and desire to teach, but physical limitations. I agree with those who are worried about this because I can also see how this could be abused. I do think it is a good thing to allow those who need a little help to be able to avail themselves of current technology. For those who worry, would you feel better if those who use text-to-speech applications submitted medical proof of them having a speech disability due to a medical condition?

[Cambota]

That's good news. I was expectend Udemy to allow Text-to-speech content. It will allow us to expand our content to another langua and reach more stundents.
Thank you.

[junderwood]

Another question: Is Udemy set up to review the quality of text-to-speech so developers can be assured their choice of product meets its standards prior to converting an entire course? Thank you.

[Jessica Thao Nguyen]

These are fantastic updates!

[ManthanPatel2024]

This is very bad, It will just lose the human touch to sharing knowledge and expertise as any random person without any knowledge and expertise can create courses with text-to-speech software and by writing script from ChatGPT!!!!

Soon, I guess we will hear reversal decision from udemy platform, hope so!!!

[EduEngTeam]

Where can I place the course setup lecture?

Plus is it allowed to share external links in it or only udemy related courses coupons?

Is there an example for such lecture? 

Plus what difference does it make from the setup lecture?

[PeterCSimon]

What are some pre-approved text-to-speech services. For example, obviously ElevenLabs is an option. What are others? Thank you

[MarinaT]

Hi @junderwood, 

Great and important question! We understand that you don't want to dedicate your valuable time to a course only to find out it doesn't meet our audio standards. Generally speaking, our quality audio standards require text-to-speech to be clear and easy to understand, and it must properly pronounce any acronyms and technical jargon in your course.

If you are uncertain whether your text-to-speech software meets our audio standards,

I recommend taking advantage of our test video service. You can submit a short video, and our team will check if the audio quality will pass our review process.

For more specific questions, please contact our policy team directly at policy@udemy.com.

[junderwood]

Wonderful! I've used those resources before, but wasn't sure if they were set up for this. Great! I've tried out several text-to-speech apps since hearing about this and wow! There are some amazing ones that have been developed since the last time I was looking at them. That was just a few years ago, right after I got out of the hospital and my "slur" or, as some said, my "lisp" was most noticeable at that time. Thank  you so much for your suggestions. I'm excited!

[VakoShvili]

It's an experiment Udemy should conduct. Maybe. It can go either way. Might end up being very beneficial to all parties, Udemy, instructors, and students:

- wider reach through multi-languages

- faster course creation and course updates for instructors. We might not need to record every single lectures. Record only the talking heads and let the AI voice do the voiceover parts.

- more instructors on the platform. Those who felt to be shy on camera or felt their voice never sounded good enough.

On the other hand. The platform might end up being flooded with AI-voice courses. Students might develop a distaste for it. And associate Udemy with just a "bunch of AI courses". Do you enjoy watching movies with AI voices? Or YouTube videos with AI voices? Probably not. Sure, if the result is indistinguishable then probably we won't be bothered. But our time and money is valuable and we constantly try to filter out the never-ending stream of content we have access to. We're trying to find the gems. Knowing something is artificial, is an enough reason for us to filter that thing out.

Copyright issues are going to be an uphill battle. People are already stealing and leaking out courses on YouTube and torrent sites. Now, they'll be able to steal the course content, ask ChatGPT to paraphrase the entire content, mix things up, mash it up with several other courses on the same topic, and bam! You've got a completely new original course, impossible to make copyright claims to. Put a sexy AI voiceover on it, with AI talking heads when needed, and you've got a course that has a chance to outsell the original course.

Someone should definitely conduct this experiment. Might end up beneficial for humanity. I dunno. But is Udemy really in the position to be the one doing it? I'd leave the experiment to YouTube and Netflix. If it works there, then bring it in.

Probably more step-by-step approach would have been better. Like giving this privilege to instructors with disabilities. And for rest of us, capping AI-generated voice to like 20% to 40% of the course.