narrationbox

Are you still working on it?

Haven't read the full report yet, just a quick question. Are your numbers for cold start without pre fill or is it after warmed cache?

> Typical production voice agent is an ensemble of three models: STT, an LLM, and TTS.

To use a claudism, I would like to push back on this. The industry is very much moving towards one-model-does-all end to end trained similar to LLMs and VLMs. Mostly for latency reasons and partially because the results for the end to end trained models are just so much better than those using three pieces architectures.

I think most of the value prop is in automatic evals, not routing specifically. A better pitch for you would be "the LM Arena of voice models" rather than comparing yourself to openrouter because the value add is rather questionable. For TTS specifically, the current SOTA for production systems are all using prompt based voice gen i.e. instead of having 10 different Tacotron models trained on 10 different models, these days it's all a single large model and the "style" is a prompt in the system prompt. The input is usually something like

  <System prompt>
  Speak in a deep smooth voice similar to a documentary narrator
  </System prompt>
  <Text to Narrate>
  Speko is the ultimate evaluation platform for voice agents. We do automatic  evals.
  </Text to Narrate>

It's the same for voice cloning too, you just pass the reference speech as an input file for all generations. A lot of systems don't have any separate style vector extraction step or model-specific fine-tuning anymore.

So something like OpenRouter for voices offer questionable value given that stakeholders usually make this sort of decisions once at the start of the project. On the other hand if you can offer automatic evals and figure out which prompts give the most similar results across different voice providers, that would offer a lot more value. It would be nice to be able to switch from e.g. Grok voice agents to ChatGPT voice agents knowing that the output style won't change too much. There are many companies now with evals as a core business model: LM Arena, Artificial Analysis, Prompt foo (before they got acquired and pivoted to security only) so many take a look at them.

Source: we have been building TTS systems for over a decade too https://narrationbox.com

Yes, the SOTA is currently much more advanced.

Give us a try, I think we are what you are looking for

https://narrationbox.com

A lot of our customers use us [0] for that, it works pretty well if executed properly. The voiceovers work best as inserts into an existing podcast. If you see the articles of major news orgs like NYT, they often have a (usually) machine narrated voiceover.

[0] https://narrationbox.com

Yeah, neural codecs are pretty amazing. The most incredible part is that they can do compression well across the temporal domain, something which has been non-trivial.

Their sentence segmentation heuristics were not configured correctly. It's not an inherent limitation of the technology itself.

The newer transformer based generators are a bit better in this regard (since they can maintain a longer context window, not just in short tiny snippets).

Mel + multispeaker vocoder is very much a classic (tacotron era) TTS approach

Since it does the signal processing in the Fourier domain, does this suffer from audio artefacts e.g. hissing in the output? Torch's inverse STFT uses Griffin-Lim which is probabilistic and if you don't train it sufficiently, you may sometimes get noise in the output.

https://pytorch.org/docs/stable/generated/torch.istft.html#t...

An alternative would be to use a vocoder network (or just target a neural speech codec like SoundStream).

> Recognizes "human" and recognizes "desk". I sit on desk. Does AI mark it as a desk or as a chair?

Not an issue if the image segmentation is advanced enough. You can train the model to understand "human sitting". It may not generalize to other animals sitting but human action recognition is perfectly possible right now.

Your average mobile processor doesn't have anywhere near enough processing power to run a state of the art text to speech network in real-time. Most text to speech on mobile hardware are stream from the cloud.

For the high end stuff no, but many of the lower tier jobs are under threat.

We used to be in this field too (https://kloudtrader.com/narwhal). It is a very crowded market and monetisation is tricky.

You can throw this together pretty quickly using one of the AutoML APIs.

So if I am launching a Spotify/Audible-style music/audiobook streaming platform, how will the pricing work out? Do we pay you instead of the original author for any user uploaded content? If the original author chooses to upload their content onto our platform, do we scan it with your API and explicitly whitelist it and pay them directly?

Your company looks very cool btw. What's your email?

What about TransferWise ewallets?

Do you have any public pricing or startup plans?

Does the accounting system support Canada and other commonwealth countries? Or is this US only?

Some ETFs also allow easy investment in stocks of foreign countries without requiring additional brokerages on behalf of the end user. I presume this does not offer that functionality?

I will look into it, Wiki2SSML looks very handy.

It looks great, have you considered adding a visual editor?

We have one for our systems: https://narrationbox.com

Are there any free plans or discounts for HN users?

We never expected audio to be this popular either when we built Narration Box :)

Lua is a great language, but the fragmentation between versions makes it difficult to be adopted outside of niche embedded areas. On a side note, we maintain a Lua newsletter but haven't had time to update it. This article looks perfect for a new edition.

https://luadigest.netlify.app

You can try locking down the app, it's not ideal but it is better than nothing:

https://news.ycombinator.com/item?id=25664130

A while back I wrote about this

https://medium.com/@kloudtrader/reducing-whatsapp-digital-fo...

Not sure if it still applies to the latest version of Android and WhatsApp but it might help. However it only mitigates certain real-time tracking and contact discovery, not to mention switching profiles is somewhat of a hassle.

This looks great! Congrats on launching. We used to be in this space too. The most tricky part is performance, especially if you are backtesting against an algorithm instead of manual trading. Having to wait for the results on 30 years of trading data can get rather annoying at times. If you do implement support for algorithmic trading, it might be helpful to rewrite the core in WebAssembly.

Interesting, it seems their products are geo-locked. The Canadian site still shows "Request invite".

Is this like Stripe Issuing where it is only available to a small number of companies? A lot of newer Stripe products seems to be available to large enterprises only. We applied for that waitlist multiple times but never heard back. What are the revenue or scale requirements for your targeted customers? Is it suitable for fintech e-wallets/banks?