frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Tried to benchmark Google's new on-device dictation model and basically couldn't

https://www.getonit.ai/eloquent-review
1•telenardo•1h ago

Comments

telenardo•1h ago
I tried to benchmark Google’s new on-device dictation app (Eloquent) and basically couldn’t. It drops about half of my dictations.

Background: Google shipped a new fully‑local dictation app yesterday with proprietary new models, so I was excited to benchmark it against the leading open models (Qwen3‑ASR, NVIDIA Parakeet V3, etc).

I have a harness that drives a dictation app by playing an audio file through a virtual input device and captures the app’s pasted output, so I can compare different apps on the same clips. I also have ~1,500 manually corrected clips from my daily engineering work.

What happened: I couldn’t get a clean eval, because ~half of dictations come back missing a large number of words. A clip of with ~20+ words routinely returns just 5-10 words. I assumed my harness was broken, so I used the app manually, speaking slowly and clearly into the mic. Same thing: roughly half the time, I only get a small fraction of what I actually said.

When Eloquent did return a complete transcript (15 of 50 tests), its accuracy was actually competitive ~24% WER vs ~21% for Qwen3-ASR on the same clips. The problem isn't the recognition. It's that for most dictations, you don't get your words back at all!

My theory: The transcriber is a chat‑style AI model, and chat models sometimes reply about your audio instead of transcribing it.

To test this, I ran Gemma 3n (Google's open model from the same family) directly on the same clips bypassing the Eloquent app. On 11 / 44 attempts it responded something like “I’m sorry, I can’t transcribe this,” instead of producing a transcript. Gemma had the same ~60 % word error rate as Eloquent. My guess is that Eloquent’s model has the same issue, the app just hides it.

Has anyone been able to get good results with this app? Or are others seeing this issue?

Disclosure: I build a competitive local dictation app, so not a neutral party!

702 Ultimatum: Warrant Requirement or Bust

https://www.eff.org/deeplinks/2026/06/702-ultimatum-warrant-requirement-or-bust
2•kevinwang•2m ago•0 comments

Loop-Harness

https://github.com/lSAAGl/loop-harness
2•LordIsBack•5m ago•0 comments

Show HN: Obsidian Image Upload Toolkit – upload images to 10 cloud providers

https://github.com/addozhang/obsidian-image-upload-toolkit
2•addozhang•9m ago•0 comments

Recovering attention during heavy study efforts

https://socketstudy.com/sparks/recovering-attention/
2•wingrove•14m ago•0 comments

Nexus Q Revival

https://mikevoyt.github.io/nexusq-revival/
2•tmp10423288442•15m ago•0 comments

Closing the Loop: One Impressive AI Coding Agent Session for Y-Combinator

https://vmysla.substack.com/p/closing-the-loop-one-impressive-ai
2•vmysla•16m ago•0 comments

A smarter approach to designing metamaterials

https://engineering.berkeley.edu/news/2025/07/a-smarter-approach-to-designing-metamaterials/
2•airstrike•17m ago•0 comments

Beneath The Enshittification, Something Amazing Is Growing

https://www.techdirt.com/2026/06/10/beneath-the-enshittification-something-amazing-is-growing/
4•hn_acker•17m ago•0 comments

Unix GC Remastered

https://mohandacherir.github.io/Qdiv7/posts/unix_new_gc/
2•mananaysiempre•19m ago•0 comments

LaserWriter Seeds

https://inventingthefuture.ghost.io/laserwriter-seeds/
2•frizlab•20m ago•0 comments

A Way to Challenge the Groupthink of Scholarly Journals

https://www.wsj.com/opinion/a-way-to-challenge-the-groupthink-of-scholarly-journals-8e59b215
2•noworld•27m ago•1 comments

Plinko Input – type a code by dropping balls

https://plinkoinput.com/
2•felixguilherme•36m ago•0 comments

The theory taking the rich by storm: China funds data center haters

https://text.npr.org/nx-s1-5844328
5•1659447091•38m ago•0 comments

Show HN: I let an AI C-suite run my company – starter kit from the inside

https://thepromptnova.gumroad.com/l/bfixc
3•clarezoe•40m ago•0 comments

Show HN: Llmbuffer – Python library for cache-optimized LLM conversation history

https://github.com/scottpurdy/llmbuffer
4•scottmp10•41m ago•0 comments

Show HN: Vatnode – EU VAT validation REST API with national registry fallback

https://vatnode.dev
4•rogulia•42m ago•0 comments

/dmg – a Claude Code skill for persistent memory and session sync

https://github.com/responsiblparty/claude-dmg-skill
4•responsiblparty•43m ago•1 comments

Nuts – pip/NPM for Java with first-class workspaces and JDK provisioning (9y+)

https://github.com/thevpc/nuts
3•thevpc•44m ago•0 comments

Show HN: Magenta Real-Time Music Generation on iPhone, Without the GPU

https://github.com/mattmireles/magenta-realtime-2-iphone
5•MediaSquirrel•45m ago•0 comments

Australia's Social Media Ban Is Floundering. Can It Still Help Younger Kids?

https://www.nytimes.com/2026/06/10/world/australia/australia-social-media-ban-under-16.html
4•uxhacker•48m ago•0 comments

Propel: Breaking the Solver Bottleneck in Task-Generator RL

https://vmax.ai/team/propel
3•AMavorParker•49m ago•0 comments

Widespread attacks on Iran have begun [video][50 mins]

https://www.youtube.com/watch?v=2KIzXrtKlbg
5•Bender•50m ago•2 comments

Patch for critical vulnerability in p2pool (Monero) to be released on 2026-06-13

https://github.com/SChernykh/p2pool/releases/tag/pre-release-v4.16
3•sxde•51m ago•1 comments

Did a Chatbot Write a Prize-Winning Story? Does It Matter?

https://www.newyorker.com/books/page-turner/did-a-chatbot-write-a-prize-winning-story-does-it-matter
4•petethomas•53m ago•0 comments

Could Switzerland Become the First Country to Cap Its Population?

https://www.newyorker.com/magazine/2026/06/15/could-switzerland-become-the-first-country-to-limit...
4•petethomas•53m ago•0 comments

Oracle beats on earnings, but stock drops on plans to raise another $20B

https://www.cnbc.com/2026/06/10/oracle-orcl-q4-earnings-report-2026.html
5•root-parent•53m ago•1 comments

Next 100 Days: XBOX Reset

https://news.xbox.com/en-us/2026/06/10/next-100-days-xbox-reset/
3•piotrgrabowski•54m ago•0 comments

Feedback Alignment in Self-Distillation

https://arxiv.org/abs/2606.11173
2•MediaSquirrel•54m ago•0 comments

US President says 'I love the inflation'

https://www.cnbc.com/2026/06/10/trump-inflation-cpi-iran-oil.html
35•root-parent•55m ago•9 comments

Trump baffles Wall Street with top dealmaker praise for Citi

https://www.ft.com/content/346fbc7b-3627-49e1-8f1e-af0cefef4000
4•petethomas•56m ago•0 comments