This is a very interesting idea, but unfortunately I seriously couldn't listen to that robot voice for more than a couple minutes at a time.
I read recently about big advances in AI speech production. I assume that is all still proprietary?
(And note, the voice is not bad compared to what computer speech synthesis used to be. I went back and listened again and was quite surprised by some human-like intonations. But, again, I couldn't listen to an article that would be longer than a minute or two.)
So it seems to be using AWS Polly's "standard" Joanna voice. However, you can also use "neural" voices (basically enhanced with ML) which sound somewhat better. Here's Joanna in neural mode: http://no.gd/voice2.mp3 .. and here's a British female voice which sounds pretty good: http://no.gd/voice1.mp3
Neural voices are 4x the price of standard ones though! Standard is $4 per 1 million characters, neural is $16. https://aws.amazon.com/polly/pricing/
I really wish I could use the Neural voices on this, but at $4 per million characters, I'm just beyond breaking even. Since this is a self-funded side project, I can't run at a loss, and I don't think you can reasonably charge more for the kind of service ListenLater.fm provides, but I'm really excited about the prices coming down in this area.
For what it's worth, Google, Amazon, and basically everywhere else I've looked have the $4/million or $16/million for the higher quality voices.
I'm currently just using Joanna since I can get some cost savings by not having to run Text to Speech on the same article multiple times, so long as everyone gets the same voice. But I'm considering offering a limited selection of voices for non-american listeners soon.
I've played with Polly a fair bit but at very low scale and, to be honest, I hadn't even noticed the price difference till today. It's amazing how vast it is and I can see why you've gone the way you have.
It probably pays to get a foot in the door with this idea now because if the prices do eventually collapse (or, better, high quality open source TTS makes huge advances) you'll be well positioned to take advantage of it.
I'd offer a more expensive version with the neural options available. Maybe offer a monthly subscription so people can give it a try with a smaller commitment. Just because you might not think that it's worth it, doesn't mean other people won't have different value propositions.
I have friends and family who would easily pay $100+/year to use this (compared to the $36/year you're charging now). The British neural option above is much more palatable than the current default, imo.
I’ll keep it in mind. I’ve yet to talk to someone willing to pay $100+/year to get that voice, but if they’re out there, I’d be happy to offer the higher quality voice options.
For anyone interested in this, send me an email at femme@femmeandroid.com saying you’d like to hear when I release the higher quality voices at a higher price.
+1 on this, would happy to pay more for a better voice. Currently it’s similar to what I get with Instapaper, so I don’t know if I would make a switch. But if the voice is better then I would like to use Listenlater!
Not my field of expertise, but seeing that generating the voice is one of the biggest costs, have you think about generating it yourself with something like https://github.com/mozilla/TTS running the process in commodity hardware?
That's good. I honestly wouldn't be able to tell the difference between Neural Jenny and someone speaking in a customer-service tone of voice. It even has the implicit pause before the "or". Still $16/million though.
I currently use https://read2me.online/ which uses the ML TTS voices from one of the big cloud providers....Google (I think). I then have a custom RSS file hosted on an S3 bucket.
I hand crank the whole system when there's something new I want to listen to
I can see this is very useful when I don’t have time to read when driving or don’t want to use my eyes after work. The elephant in the room is how do you cope with articles behind paywall?
So I just added a feature last night as this was blowing up on Reddit that lets you paste text into a form. It’s a little less convenient at the moment. But I’d like to improve the process of sending it arbitrary text.
I like stuff like this but am always ruined by the text-to-speech engines. I just can’t do it.
For people who want a curated selection of high-quality articles read by best of breed narrators, check out Audm [1]. It isn’t going to be be on-demand or cover everything that one would want, but the source material is quite good (The Atlantic, New York Times, ProPublica, the New Yorker, Wired, etc.) and the narration is notable because most of the roster are well-known audiobook readers, which is different from similar services or even the stuff Apple is doing with Apple News+. For me, I’ve found that not only does having a human narrator make a huge difference, but having a good reader is a game changer. I’ve been a subscriber for a few years and it’s one of the most high-value subscriptions I have.
I read a lot — at least four or five books a month and hundreds of pages of blogs, magazines, scientific papers, etc. — but I really do like to listen to books or news when I have the chance. I can read about twice as fast as I can listen, but it isn’t always ideal.
For some reason, I can happily listen to robotic voices for a very long period of time. I have listened to several books that way, when there isn't a good narrated version.
In fact, apart from a few quirks (a pause at the "." of Mr. for instance), the iOS TTS engine sounds almost pleasant to me. It's no where near a good narrator, of course, but good enough if there is no poetic value in the words.
As opposed to the other commenters, I like the voice. I can listen at 2x speed and understand without trouble. It's way better than the voices that come with espeak(-ng), at least. This one doesn't sound robotic at all, in comparison.
I'm glad you feel that way. I added the speed options on the homepage in hopes that others would come to the same realization that the voice is pretty good if you're already listening to audio sped up. I definitely know listening to robotic voices isn't for everyone, and I'd love for the voices to get better, but for now it's working well enough for me to enjoy using it.
This looks like it is made by the same people behind Huffduffer [0]? Which is an absolutely amazing service I've been using for years: Just throw any audio file against it, and it will add it to a personal podcast feed.
For me, reading and listening to the text at the same time doubles the focus.
I use the Alex voice on my MacBook with a bookmarklet script to both read any web page and highlight text as it is being read. Special integrations for reddit and HN to skip reading user names and chrome. I used to have a version of this script work in PDF files as well, but recently that route has been closed. A nice UX feature is to trigger reading by clicking on the desired word, so you can easily pause and resume.
lit, earlier today i needed to do shopping which I hate so i wanted to listen to some of the articles my friends have been sending me and I have not responded because they are too long but of course mobile text to speech extensions inside android where not working well and even the goolge assistant was a pain in the ass to setup and then didn't work.
So yea thanks for making this went for premium right away I assume its good but in general want to support audio based products :)
One suggestion. I ran into a hiccup when I was setting up the Chrome extension. It asked for the API from my account. I copied the API and pasted it into the extension but it didn't seem to work when I tried sending an article. I went back to the extension setting and realized that when I pasted the API number it added it to the default text saying to go to listenlater for the api code. Once I manually deleted this and left the API key in everything worked. Maybe you can update so that when you click into the box to enter API the explanation text auto removes itself.
Neat, thanks for sharing! This looks somewhat similar to Herald (https://heraldpod.com/). Herald supports custom voices for paid users, as well as forwarding newsletters and having them read aloud. What kinds of additional features are on your roadmap?
40 comments
[ 11.3 ms ] story [ 117 ms ] threadI read recently about big advances in AI speech production. I assume that is all still proprietary?
(And note, the voice is not bad compared to what computer speech synthesis used to be. I went back and listened again and was quite surprised by some human-like intonations. But, again, I couldn't listen to an article that would be longer than a minute or two.)
Neural voices are 4x the price of standard ones though! Standard is $4 per 1 million characters, neural is $16. https://aws.amazon.com/polly/pricing/
For what it's worth, Google, Amazon, and basically everywhere else I've looked have the $4/million or $16/million for the higher quality voices.
I'm currently just using Joanna since I can get some cost savings by not having to run Text to Speech on the same article multiple times, so long as everyone gets the same voice. But I'm considering offering a limited selection of voices for non-american listeners soon.
It probably pays to get a foot in the door with this idea now because if the prices do eventually collapse (or, better, high quality open source TTS makes huge advances) you'll be well positioned to take advantage of it.
I have friends and family who would easily pay $100+/year to use this (compared to the $36/year you're charging now). The British neural option above is much more palatable than the current default, imo.
All in all, nice work!
For anyone interested in this, send me an email at femme@femmeandroid.com saying you’d like to hear when I release the higher quality voices at a higher price.
Jenny (neural) Customer service and newscast
I like Polly from an integration perspective but I’ve been searching for some as good as Azure Jenny.
Anyone found something as good or better?
I thought that maybe listenlater.fm would record a high-quality reading from the most sent-in articles a day (do things that don't scale).
However, it apparently uses text-to-speech :/
https://help.getpocket.com/article/1081-listening-to-article...
Edit: Oh, this new service provides value in that it generates a custom podcast feed using the articles, which you then subscribe to.
I currently use https://read2me.online/ which uses the ML TTS voices from one of the big cloud providers....Google (I think). I then have a custom RSS file hosted on an S3 bucket.
I hand crank the whole system when there's something new I want to listen to
Doesn't seem to be working. Putting a link returns "Bad request"; uploading a .txt file returns "Bad referrer/request type".
Here's the error from file upload:
https://postimg.cc/NLfXFspf
Here's the error from using a link:
https://postimg.cc/QHQfDxtN
For people who want a curated selection of high-quality articles read by best of breed narrators, check out Audm [1]. It isn’t going to be be on-demand or cover everything that one would want, but the source material is quite good (The Atlantic, New York Times, ProPublica, the New Yorker, Wired, etc.) and the narration is notable because most of the roster are well-known audiobook readers, which is different from similar services or even the stuff Apple is doing with Apple News+. For me, I’ve found that not only does having a human narrator make a huge difference, but having a good reader is a game changer. I’ve been a subscriber for a few years and it’s one of the most high-value subscriptions I have.
I read a lot — at least four or five books a month and hundreds of pages of blogs, magazines, scientific papers, etc. — but I really do like to listen to books or news when I have the chance. I can read about twice as fast as I can listen, but it isn’t always ideal.
[1]: https://audm.com
In fact, apart from a few quirks (a pause at the "." of Mr. for instance), the iOS TTS engine sounds almost pleasant to me. It's no where near a good narrator, of course, but good enough if there is no poetic value in the words.
[0] https://huffduffer.com/
Thanks for mentioning it. And as I have in my ToS and Privacy Policy, I will not be using your email for anything you haven't explicitly requested.
I use the Alex voice on my MacBook with a bookmarklet script to both read any web page and highlight text as it is being read. Special integrations for reddit and HN to skip reading user names and chrome. I used to have a version of this script work in PDF files as well, but recently that route has been closed. A nice UX feature is to trigger reading by clicking on the desired word, so you can easily pause and resume.
So yea thanks for making this went for premium right away I assume its good but in general want to support audio based products :)