I’ve been working on Polly Podcast for a few months and I’m excited to share it with the community. I made this so that I could listen to articles while at work or during my commute.
Once I got into building the platform I found this incredible voice synthesis community in NYC, which led me to Recess Labs[0], which is where I am now while I work on building my own voice model based on the recent Tacotron 2 paper[1].
Would love feedback and I’m happy to answer any questions or talk about voice-tech.
[0] http://www.recesslabs.com/ [1] https://research.googleblog.com/2017/12/tacotron-2-generatin...
1. Have you heard of Narro (narro.co)? It does the same thing, but it’s been around longer and has many integrations.
2. Can you share the tech stack currently powering your voice synthesis? Since it’s called PollyPodcast, are you currently using AWS Polly for TTS?
3. Is this something you intend to monetize?
4. It seems like you offer unlimited readings for free. Do you have an abuse prevention strategy? This could become a hole in your wallet if you’re paying for a charged-per-character TTS service.
5. Creating custom voices is a great differentiator. What are some problems you have to solve to create them and how is this effort progressing? Do you have quality and performance goals?
1. Narro is really, really, great.
2. The app itself is on Rails with AWS Polly transcribing most of the requests. x% of the transcriptions are routed to a Flask API that runs a neural TTS engine + with token-based SSML rolling out slowly for some heavy-users.
3. I have no plans to monetize Polly Podcast.
4. Unlimited readings.
5. I'm working now with large publishers to figure out what the costs look like to generate new synthetic voices at scale. Hope to release this publicly as soon as I get the go-ahead.
If you have any questions feel free to reach out in email =]
Pretty cool product.
