> Really cool project. This will definitely help me read more papers. Can you share more on how the backend is parsing and converting text to speech?
Thank you kindly! I'm really glad to hear this - this was exactly what I hoped for when developing this.
Happy to share more details : right now we use an ensemble of methods to parse uploaded PDFs. A chunk of this involves using GROBID (https://grobid.readthedocs.io/en/latest/), a machine learning library aimed at parsing academic papers. Funnily enough, GROBID is itself a cascade of sequence labeling models trained on document parsing.
The text to speech portion is driven by OpenAI's text-to-speech models, which in my experience seem to deliver the market leading audio quality.
The summarization is driven by GPT-3.5-turbo
As such, the platform does focus quite a bit on making good sounding audiobooks from academic PDFs.
Some of the updates on the roadmap will include improved handling around tabular, graph, and figure content along with mathematical and scientific equations.
Its likely that a multi-modal LLM could do a reasonable job at describing this content in spoken form.
> UI issue: Login by google icon covers the password box when trying to create a new account on my phone.
My apologies about this! I'll get this fixed asap
> Do you have any formal channel for feature request? I'll pay for this app.
Very much appreciate this - could you reach out support [at] trurecord.com?
I would love to touch base about feature requests you have in mind - I'm really keen to deliver a great experience for users like yourself and am eager to learn about what you'd find helpful.
Thank you again for your message and look forward to getting in touch
It very much can; however, the iOS app has a limit of 50 pages and an hourly limit of 5 uploads in our 'free tier'.
I didn't want to rush to monetize the iOS app so I could really learn from users like yourself, and subsequently work hard to really make the app great to use.
Currently, to sign up for a subscription you can go to the 'Subscription' setting on our web app : http://oration.app/accounts/login/
Subscribing for an account will bypass the upload page limit and hourly limit.
> would it be possible to specify pages of a large pdf?
This is definitely something that I'm aiming to ship soon - I'm trying to both deliver something where a user can have a simple upload experience and an enjoyable Audiobook output, but also provide some more fine-grained handles (like specifying what pages to use, among other things)
> I'll get back to you with a list after using it for a few weeks.
Definitely eagerly looking forward to this!
Please don't hesitate to email us at support [at] TruRecord.com (there is also a support e-mail link within the app). We'd be more than happy to also meet with yourself over Zoom to learn from our experience and work towards delivering great functionality.
Thank you kindly! I'm really glad to hear this - this was exactly what I hoped for when developing this. Happy to share more details : right now we use an ensemble of methods to parse uploaded PDFs. A chunk of this involves using GROBID (https://grobid.readthedocs.io/en/latest/), a machine learning library aimed at parsing academic papers. Funnily enough, GROBID is itself a cascade of sequence labeling models trained on document parsing. The text to speech portion is driven by OpenAI's text-to-speech models, which in my experience seem to deliver the market leading audio quality. The summarization is driven by GPT-3.5-turbo As such, the platform does focus quite a bit on making good sounding audiobooks from academic PDFs. Some of the updates on the roadmap will include improved handling around tabular, graph, and figure content along with mathematical and scientific equations. Its likely that a multi-modal LLM could do a reasonable job at describing this content in spoken form.
> UI issue: Login by google icon covers the password box when trying to create a new account on my phone.
My apologies about this! I'll get this fixed asap
> Do you have any formal channel for feature request? I'll pay for this app.
Very much appreciate this - could you reach out support [at] trurecord.com? I would love to touch base about feature requests you have in mind - I'm really keen to deliver a great experience for users like yourself and am eager to learn about what you'd find helpful.
Thank you again for your message and look forward to getting in touch