3 ms·

Show HN: Split-Test Different Voice AI Providers (Playground)

7 points by bephrem321 2y ago

I'm currently working on a "home for voice AI developers" called Vocalized. Here is a link to the playground where you can split-test different voice AI providers (I'm still working on making it mobile-responsive so try it on desktop!). I want the playground to be an complete documentation of every offering that exists in the space (whether relevant or not) so each can be compared. The intent for the site is to be a lightweight site you can bookmark & come back to to poke around if interested. The site also has a company directory that I'm working on. I wanted to offload my browser bookmarks into a navigable site others interested could read & contribute to. other notes: - when GPT-4o for voice is available I will add it - I'm looking for contributors to the company directory - I'm open to open-sourcing the site if there's interest

tmshapland 2y ago
The playground is really cool! I also really like the directory of Voice AI companies on your site here: https://vocalized.dev/apriora https://vocalized.dev/apriora. I've really enjoyed being a part of the Voice AI community. There are a lot of good people with a lot of brain power working in this space in a collaborative way. I don't know if we fit on your projects page, but I'd love if you considered adding our product to your projects page. We're building a context-aware semantic cache. The hard part with semantic caching is getting it to work in conversation because the same question means two different things at different points in the convo. We're focused on that problem. A lot of Voice AI companies use our context-aware semantic cache to reduce their latency. https://canonical.chat/blog/how_to_build_context_aware_semantic_cache https://canonical.chat/blog/how_to_build_context_aware_seman...
- bephrem321 2y ago
  will add it today! thanks for sharing
  - tmshapland 2y ago
    thank you! I'm here on LinkedIn. Would love to connect! https://www.linkedin.com/in/tom-shapland-b4494212/ https://www.linkedin.com/in/tom-shapland-b4494212/
waldrews 2y ago
Could you talk about your choices for speech thresholds/silence detection? GPT-4o supposedly can speak with interruptions, but for the rest, presumably you have to either do VAD or some ad-hoc volume based rules to detect start and stop, unless you're using browser WebSpeech API on the listening side?

7 points by bephrem321 2y ago

tmshapland 2y ago

The playground is really cool! I also really like the directory of Voice AI companies on your site here: https://vocalized.dev/apriora https://vocalized.dev/apriora. I've really enjoyed being a part of the Voice AI community. There are a lot of good people with a lot of brain power working in this space in a collaborative way. I don't know if we fit on your projects page, but I'd love if you considered adding our product to your projects page. We're building a context-aware semantic cache. The hard part with semantic caching is getting it to work in conversation because the same question means two different things at different points in the convo. We're focused on that problem. A lot of Voice AI companies use our context-aware semantic cache to reduce their latency. https://canonical.chat/blog/how_to_build_context_aware_semantic_cache https://canonical.chat/blog/how_to_build_context_aware_seman...

bephrem321 2y ago
will add it today! thanks for sharing
- tmshapland 2y ago
  thank you! I'm here on LinkedIn. Would love to connect! https://www.linkedin.com/in/tom-shapland-b4494212/ https://www.linkedin.com/in/tom-shapland-b4494212/

waldrews 2y ago

Could you talk about your choices for speech thresholds/silence detection? GPT-4o supposedly can speak with interruptions, but for the rest, presumably you have to either do VAD or some ad-hoc volume based rules to detect start and stop, unless you're using browser WebSpeech API on the listening side?