I love using https://convert.to.it/ for all of my converting needs; everything is client-side in the browser with over 400 file extension options to choose from.
“Never ever” is crazy talk. Just put a couple of dollars in a prepaid account and assume it may all get used up. “Standard practices” won’t get you far in life.
I did something very similar to this, for myself, but which takes a web page for an article, cleans up the text to preserve the article only, and converts it using TTS, then adds it to a cloud bucket with an RSS stream, and I use that RSS link as a podcast in Apple Podcast.
I do similar things, just use ElevenLabs for tts - a good selection of voices.
A good protip - tell an agent to rewrite the article slightly for better listening flow. Some subtle changes in sentence structure etc can make listening way more pleasant.
Oh man. I did something similar but didn’t do the last mile with the cloud bucket and RSS stream and that seems key. I never have the energy to manually copy the mp3s over to the phone.
I did something similar and just put the files on an old laptop and used Tailscale. That gives me an https URL and Apple Podcasts downloads the files fine.
How well does it handle pronunciation, especially of things like proper nouns? This has been one of the harder parts I've found. On a financial web site I saw one example where the generated audio pronounced initiates as a noun instead of a verb. It's a subtle difference but can significantly change the meaning if worded wrong.
I was trying to use TTS to regenerate some old audiobooks and getting names pronounced correctly is pretty hard unless the model supports something like SSML.
It was always harder to get it to sound right, I had built Lissin around this thesis itself. You can throw in a PDF or a link and can get a podcast, deepdive or just narration with different voices in Lissin.
For those interested in better voices, you can try and let me know.
If someone sends me AI generatef audio they're reccomending, I from then on out see them differently and not in a good way. I dont even want to read the outputs of an LLM let alone listen to them. Wild.
Let me guess, you can already build such a system yourself quite trivially by getting an FTP account, mounting it locally with curlftpfs and then use SVN or CVS on the mounted filesystem?
I thought this would mean I could use my subscription, as OpenAI/Codex supports this kind of authentication flow, and it's very easy to integrate with that as a service, and users login with the OpenAI authentication flow, or via device codes.
Instead this asks for an API key, which is very different from "Use your OpenAI subscription" and to be honest, I'm not sure how anyone who has a OpenAI subscription could mistake this for such. Has this even been tested once?
Maybe I missed something obvious, I had a five second look and closed it when it seemed to be outright intentionally misleading.
There are two parts to the processing. The processing of the document itself is done on your OpenAI/Codex subscription. Additionally, you have to submit an API key for the text-to-speech processing. There you can use a Gemini key or an OpenAI key.
Thanks! Currently, I'm using Gemini TTS and Kokoro TTS as a free option. Kokoro is fast and I can run it on my own server, but I'm not using it as a standard yet because Gemini supports more languages and has more voices available. But text-to-speech processing is the biggest bottleneck at the moment.
At home this is super easy, use one of the multiple TTS models out there and provide a clean clip of Morgan Freeman's voice. Typically around 10 seconds will do. I just run a python script to daily scan my FreshRSS sources, compile my news and generate a 'podcast' that's about 10 minutes for a morning listen. Currently using F5-TTS, but there are lot's of other models out there as well.
Claude has gotten pretty sensitive to copyright. In addition to the no song lyrics that was posted in another article, I also find it will not generally transcribe podcasts (you have to play with model versions and even effort inside that).
I went the route of handling epub first (really, anything pandoc can convert to epub) on ebookaloud, and it took quite a few iterations to get consistent parsing across a wide corpus of test documents.
My experience is that gathering and testing with that corpus was about 30% of the overall effort. I have not yet gathered the motivation to tackle pdf: it’s more of a layout format than a text format. Not just tables but multi-column layouts and similar oddities leave a pile of imperfect heuristics in their wake.
Multi-lingual support is even worse. I was working with arabic script in pdf years ago and they don’t even capture the encoded text, just rendering instructions due to the high number of ligatures. I don’t know if things have improved since then.
Source: https://github.com/p2r3/convert
For example, these audiobook generators runs fully offline. https://github.com/Finrandojin/alexandria-audiobook https://github.com/senigami/audiobook-studio.pinokio https://github.com/topics/audiobook-generator
That should be everybody!
There are standard practices for delegating responsibility, never ever be pasting API keys into websites.
A good protip - tell an agent to rewrite the article slightly for better listening flow. Some subtle changes in sentence structure etc can make listening way more pleasant.
https://blog.nawaz.org/posts/2024/Apr/reading-articles-via-p...
I use the TTS on Google Cloud. It's effectively free.
I was trying to use TTS to regenerate some old audiobooks and getting names pronounced correctly is pretty hard unless the model supports something like SSML.
For those interested in better voices, you can try and let me know.
https://apps.apple.com/app/apple-store/id6758897526?pt=12810...
For a wrapper around costly LLM providers.
For a result one could get with voicebox for free.
Via MCP more voices are available (all OpenAI and Gemini voices, depending on which api key you use).
https://news.ycombinator.com/item?id=9224
I thought this would mean I could use my subscription, as OpenAI/Codex supports this kind of authentication flow, and it's very easy to integrate with that as a service, and users login with the OpenAI authentication flow, or via device codes.
Instead this asks for an API key, which is very different from "Use your OpenAI subscription" and to be honest, I'm not sure how anyone who has a OpenAI subscription could mistake this for such. Has this even been tested once?
Maybe I missed something obvious, I had a five second look and closed it when it seemed to be outright intentionally misleading.
Much less interesting.
edit: i guess this explains it, its sorta half half
Your subscription will be used for:
Your API key will be used for:Thanks for listing the model you're using, I've not really done any TTS stuff and this sounds like a fun thing to spin up.
I'd love to see an actual result what it can do
My experience is that gathering and testing with that corpus was about 30% of the overall effort. I have not yet gathered the motivation to tackle pdf: it’s more of a layout format than a text format. Not just tables but multi-column layouts and similar oddities leave a pile of imperfect heuristics in their wake.
Multi-lingual support is even worse. I was working with arabic script in pdf years ago and they don’t even capture the encoded text, just rendering instructions due to the high number of ligatures. I don’t know if things have improved since then.