Best Options for Upgrading Speech Capabilities (looking for a PAI speech service for conversations, mobile option, …wake…word? 😬

[emory]

New and improved with less v on my -flags:

Voice Conversations with Your DA

I have a dialed-in ElevenLabs voice for my PAI's TTS announcements, and I've even routed it through macOS speech services - more useful than I expected.

The Core Problem

I learn and retain information by talking about it. Not reading, not listening - talking through it. (This is actually an asset when threat modeling or understanding how teams work, but it's a constraint for knowledge work.)

I have a heavily personalized Claude Mobile I can talk to, but that's not going to cut it. I use an ElevenLabs Agent with similar aura-prompting at my desk, but I want this everywhere - especially away from my desk where I need it most.

What I've Found So Far

[Claude-to-Speech](https://github.com/LAURA-agent/Claude-to-Speech) - A Claude Code plugin

I haven't used many plugins. Honestly, there are too many suites (especially tailored for infosec professionals) - they're big lifts that add 40+ new commands/functions each. I've cloned many and use them as inspiration for agents or skills instead.

This one has some social awareness built in - like not reading diffs aloud at you, which is nice.

What I'm Actually Looking For

"Having a chat with a colleague or friend in the room" - conversational flow, spoken dialogue with my DA and subagents in Claude Code space.

If anyone has used something like this (or better), I'd love to hear about it.

Options I'm Considering

1. Present the ElevenLabs voice agent to my phone - Currently using Gemini for this, need to figure out how it's even working (don't recall creating creds for my Google tenant)

2. Cram my DA's aura prompt into an app using Gemini Live API - Would get acceptable results but not ideal

3. Something better I haven't found yet

The goal: Spoken conversations with Sundae and subagents in Claude Code, with the same context and personality depth I get in text.

---

Has anyone solved this? What's working for you?

---

[smolcompute]

Hi Emory. May I suggest writing your posts in your first language and having Claude translate them to English before posting? You seem to have a lot of interesting ideas to contribute, but I find it difficult to understand you.

[emory]

Unfortunately, my first language is American English. I believe the above is a form of "model collapse" that my DA would have spared you from though so, noted.

This is almost certainly why my PAI determined that assistive technology was going to be her purpose.

[emory]

Sundae took a cruise and trimmed things and arranged it differently for me, hope that helps. I often write to think, and sometimes I forget to do a "garbage collection" task of some sort?

I think when I'm in github I default to a "communicating with LLMs and Sundae" mode lately. I gotta watch out or I'll learn gibberspeak or start whistling a 9600 baud carrier at my Homepod. Thanks for the flag @smolcompute <4

[smolcompute]

I'd also be very interested in making this work. Basically voice mode for PAI.
I'm sure we'll get there eventually.

[danielmiessler]

Please refer to this major change coming to the repo that should address your issues. If it doesn't please resubmit. Thank you for the work!

#222