Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift • Hacker News Board

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#131

Post by jwr » Fri, Mar 06, 2026, 4:21 PM UTC

kavith wrote: post #28

jwr wrote: post #10

As a heavy user of MacWhisper (for dictation), I'm looking forward to better speech-to-text models. MacWhisper with Whisper Large v3 Turbo model works fine, but latency adds up quickly, especially if you use online LLMs for post-processing (and it really improves things a lot).

Not sure if this will help but I've set up Handy [1] with Parakeet V2 for STT and gpt-oss-120b on Cerebras [2] for post-processing and I'm happy with the performance of this setup! [1] https://handy.computer/ [2] https://www.cerebras.ai/

But the advantage here is — what? That I don't have to pay for it? I'm glad it works for you, but I'm in a (solo) business setup where I'd much rather pay a (very reasonable) amount of money for a tool that gets things right and smooths out the edges.

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#132

Post by cpill » Fri, Mar 06, 2026, 5:39 PM UTC

taf2 wrote: post #55

vessenes wrote: post #5

This is cool. It makes me want an unsloth quant though! A 7b local model with tool calling would be genuinely useful, although I understand this is not that. UPDATE: I'd skip this for now - it does not allow any kind of interactive conversation - as I learned after downloading 5G of models - it's a proof of concept that takes a wav file in.

I forked and added tool calling by running another llm in parallel to infer when to call tools it works well for me to toggle lights on and off. Code updates here https://github.com/taf2/personaplex

what do you mean "infer"? how does the LLM get anything it of this as input?

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#133

Post by cpill » Fri, Mar 06, 2026, 8:09 PM UTC

AlexeyBelov wrote: post #127

andreadev wrote: post #117

Earlier quoted context omitted.

The framing in this thread is full-duplex vs composable pipeline, but I think the real architecture is both running simultaneously — and this library is already halfway there. The fact that qwen3-asr-swift bundles ASR, TTS, and PersonaPlex in one Swift package means you already have all the pieces. PersonaPlex handles the "mouth" — low-latency backchanneling, natural turn-taking, filler responses at RTF 0.87. Meanwhi…

LLM slop.

Don't be so hard on yourself :P

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#134

Post by cpill » Fri, Mar 06, 2026, 9:19 PM UTC

I think the best example I've seen of the pipeline approach: ASR(STT) -> LLM -> TTS is still from about 10 months ago in HN https://github.com/KoljaB/RealtimeVoiceChat

Just a pity the code is undecipherable :(

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#135

Post by AlexeyBelov » Sat, Mar 07, 2026, 5:01 AM UTC

andreadev wrote: post #128

AlexeyBelov wrote: post #127

Earlier quoted context omitted.

LLM slop.

Which part specifically?

The part where it's in all your comments.

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#136

Post by andreadev » Sat, Mar 07, 2026, 5:31 AM UTC

AlexeyBelov wrote: post #135

andreadev wrote: post #128

Earlier quoted context omitted.

Which part specifically?

The part where it's in all your comments.

You are wrong but I am not going to keep going back and forth.

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#137

Post by Yanko_11 » Sat, Mar 07, 2026, 9:47 AM UTC

[dead]

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#138

Post by Yanko_11 » Sun, Mar 08, 2026, 5:28 PM UTC

[dead]

Re: Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

#139

Post by shorggth » Sun, Mar 15, 2026, 3:44 PM UTC

jwr wrote: post #10

As a heavy user of MacWhisper (for dictation), I'm looking forward to better speech-to-text models. MacWhisper with Whisper Large v3 Turbo model works fine, but latency adds up quickly, especially if you use online LLMs for post-processing (and it really improves things a lot).

In my benchmarks Parakeet v3 actually beats Whisper Turbo on CER (3.8% vs 4.0% on BBC News audio) while being nearly 2x faster (25x vs 14x RT on M1). The accuracy gap may depend on audio type — Parakeet tends to be better on clean speech while Whisper handles noisy/accented audio more robustly.