Hacker News Reader

Comment by petercooper | Hacker News Reader

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/

View on HN

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.

All in all, nice work!

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!

I'd pay extra $5 per month for a better voice.

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?

My favorite neural voice is from azure - https://azure.microsoft.com/en-us/services/cognitive-service...

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?

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.

Reply on news.ycombinator.com