I've used Amazon's text-to-speech API to create voice-over for my video | Hacker News Reader

I've used Amazon's text-to-speech API to create voice-over for my video

youtube.com

8 pointsrayalez6 comments

I have trouble speaking, I don't really like my voice and accent, so I decided to try using Amazon Polly and see if it helps me to create a watchable video.

It ended up working surprisingly well. I have used reveal.js to generate slides, and PhantomJS to automatically render them into images. Then I've edited them together using Kdenlive. The entire process of turning my article into a video took a couple of hours.

I think it turned out pretty cool and I figured I'd share it with you guys. It's imperfect, but it's really impressive how far text-to-speech engines have come.

At this point it still probably makes more sense to spend $10-$20 on fiverr to create a professional voiceover, but I can totally see how in the near future it would be possible to automatically generate high quality videos.

I think I'm onto something here, maybe I'll make a SaaS tool out of this. Very cool stuff.

Fascinating. Is the api simply text-in to audio-out, or can the text be marked up for composable with something like SSML? The ability to insert appropriate paused for pacing, change intonation, etc would be very useful.

Oh wow, they go further than SSML, with visemes and their phoneme maps as well. [0] It's a shame there doesn't seem to be a quality IDE for such things.

[0] http://docs.aws.amazon.com/polly/latest/dg/speechmarks.html

Yup, you can use SSML. I've used it to add pauses and other tweaks to make it sound more natural.

You can read about the available tags here:

http://docs.aws.amazon.com/polly/latest/dg/supported-ssml.ht...

I've already come across a bunch of YouTube videos with artificial voiceover. Personally, I'd rather listen to a person who stutters or he a bad accent, than a machine.

I like it. I've been thinking of something similar. Just a note, I hear in the US people find a female voice more pleasing for automated voices. But it depends on the country. In Germany, I understand a male voice is better.

Reply on news.ycombinator.com