Independent testing will determine whether parakeet.wgsl makes accurate, dependency-free browser-local ASR practical across ordinary WebGPU-capable devices at its claimed speed.

state: expiredheat: lowuncertainty: highknownscott: mediumlocal-inference webgpu speech-recognitionnarcotic-shNVIDIA

What is this?

parakeet.wgsl is presented as a browser-local speech-recognition implementation of NVIDIA’s Parakeet TDT 0.6B V2, built with raw WebGPU/WGSL and SIMD WASM rather than a larger inference framework. The supplied results support the broader premises that Parakeet models can be very fast and that WebGPU can run compute-intensive deep-learning kernels portably in browsers. However, they provide no independent device-level tests of parakeet.wgsl itself, so its claimed accuracy, speed, dependency-free operation, and compatibility across ordinary WebGPU hardware remain unverified here.

Why it matters to Scott

Scott already holds both sides of this case: Capability Audit requires representative, vendor-neutral testing rather than demo claims, while Hardware-aware local inference and gamepc document active work with local speech inference. The browser/WebGPU deployment path could materially extend his CUDA/Whisper-centred stack to client devices, but until independent cross-device benchmarks exist it remains an unverified implementation in a territory already tracked by the local-inference and local-speech validation radar pages.

ip:concept.capability-auditdev:concept.hardware-aware-local-inferencedev:project.gamepcdev:technology.whisperradar:concept.local-inferenceradar:concept.ai-benchmarksradar:audio-cpp-0-4-local-speech-validationradar:nemo-speech-cpp-local-stack

queries asked of Scott's wikis

browser-local inference architecture
WebGPU inference portability and performance
dependency-free AI runtimes
on-device speech recognition
local inference privacy and economics
cross-device AI benchmark methodology

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (3) — ⭐ canonical anchor

source	object	author	score	comments
🟠 reddit	parakeet.wgsl – Fast, accurate ASR in the browser, via raw WebGPU and SIMD WASM LocalLLaMA	hamza_q_	1	0
🟧 echo.github ⭐	The initial commit introduces parakeet.wgsl as “High-performance inference of NVIDIA's Parakeet TDT 0.6B V2 ... in the browser,” using “raw	Hamza Qayyum (hamzaq2000)	—	—
🟠 reddit	parakeet.wgsl – Fast, accurate ASR in the browser, via raw WebGPU & SIMD WASM LocalLLaMA	hamza_q_	31	10

Interpretation history

2026-08-10T06:28:34Z
No independent benchmarks, cross-device tests, or implementation adoption appeared within the episode’s active horizon; the discussion has stalled at launch claims and anecdotal praise. Close this episode unless reproducible ordinary-device evidence creates a fresh validation delta.
2026-08-08T06:26:36Z
Refreshed discussion remains anecdotal praise without device details, accuracy measurements, or reproducible cross-device tests of parakeet.wgsl. It adds no independent line of evidence, so the case remains an open but cool validation question.
2026-08-07T23:30:44Z
A commenter adds anecdotal phone-speed and accuracy testimony, but may be describing the underlying Parakeet model rather than parakeet.wgsl and supplies no device, browser, measurements, or reproducible test. The case therefore still lacks independent validation of this WebGPU implementation across ordinary hardware.
2026-08-07T20:29:03Z
The added item is same-author redistribution of the initial release, not independent validation; the case still hinges on representative accuracy and cross-device performance tests. Minor engagement adds no substantive corroboration, so the episode can cool while remaining open and testable.
2026-08-07T18:22:12Z
evidence attached: reddit.post.1vi77dr — Provides a first-party implementation and claimed browser-local performance directly relevant to the open validation case.
2026-08-07T17:29:41Z
grounded: known/medium — Scott already holds both sides of this case: Capability Audit requires representative, vendor-neutral testing rather than demo claims, while Hardware-aware loca
2026-08-07T17:26:45Z
origin walked (codex/luna, conf 0.99): anchor reddit.post.1vi6q49 -> echo.github.56a46d2dbe by Hamza Qayyum (hamzaq2000)
2026-08-07T17:25:34Z
case created — A usable implementation and source package create a concrete local-inference episode, though its exceptional speed claim remains unvalidated.

Decision trace

08-10 16:28expireNo independent benchmarks, cross-device tests, or implementation adoption appeared within the episode’s active horizon; the discussion has stalled at launch claims and anecdotal praise. Close this epi
08-10 16:28alert_silentThe staleness trigger contains no new consequential evidence, so there is nothing to surface; a future independent accuracy and cross-device benchmark should reopen the subject as a new delta.
08-10 16:28alert_routeThe staleness trigger contains no new consequential evidence, so there is nothing to surface; a future independent accuracy and cross-device benchmark should reopen the subject as a new delta.
08-09 01:21sensor_dirtyengagement_update
08-08 16:26repriceRefreshed discussion remains anecdotal praise without device details, accuracy measurements, or reproducible cross-device tests of parakeet.wgsl. It adds no independent line of evidence, so the case r
08-08 16:26alert_silentThe new delta is repetitive commentary rather than a confirmed benchmark, implementation adoption, or product deployment; it can wait for independent accuracy and ordinary-device performance evidence.
08-08 16:26alert_routeThe new delta is repetitive commentary rather than a confirmed benchmark, implementation adoption, or product deployment; it can wait for independent accuracy and ordinary-device performance evidence.
08-08 16:21sensor_dirtycomment_update
08-08 13:21sensor_dirtyengagement_update
08-08 10:21sensor_dirtyengagement_update
08-08 09:30repriceA commenter adds anecdotal phone-speed and accuracy testimony, but may be describing the underlying Parakeet model rather than parakeet.wgsl and supplies no device, browser, measurements, or reproduci
08-08 09:30alert_silentThe refreshed comment is weak, ambiguous testimony rather than a confirmed benchmark or deployment of parakeet.wgsl; it can wait for reproducible cross-device results or first-party product evidence.
08-08 09:30alert_routeThe refreshed comment is weak, ambiguous testimony rather than a confirmed benchmark or deployment of parakeet.wgsl; it can wait for reproducible cross-device results or first-party product evidence.
08-08 08:21sensor_dirtycomment_update
08-08 06:29repriceThe added item is same-author redistribution of the initial release, not independent validation; the case still hinges on representative accuracy and cross-device performance tests. Minor engagement a
08-08 06:29alert_silentNo new consequential delta occurred: the implementation and author-reported M5 result were already known, while independent benchmarks, accuracy measurements, and ordinary-device compatibility evidenc
08-08 06:29alert_routeNo new consequential delta occurred: the implementation and author-reported M5 result were already known, while independent benchmarks, accuracy measurements, and ordinary-device compatibility evidenc
08-08 04:22alert_silentThis is a same-author repost of the already-seen initial parakeet.wgsl release, adding no independent benchmarks, cross-device results, accuracy measurements, or material implementation change. Retain
08-08 04:22alert_routeThis is a same-author repost of the already-seen initial parakeet.wgsl release, adding no independent benchmarks, cross-device results, accuracy measurements, or material implementation change. Retain
08-08 04:22attachProvides a first-party implementation and claimed browser-local performance directly relevant to the open validation case.
08-08 04:21propose_attachProvides a first-party implementation and claimed browser-local performance directly relevant to the open validation case.
08-08 03:40alert_shadowSource code and an npm package now make this a testable implementation relevant to Scott’s local-speech stack. The release is established, but its claimed accuracy, broad device compatibility, and 180
08-08 03:40alert_routeSource code and an npm package now make this a testable implementation relevant to Scott’s local-speech stack. The release is established, but its claimed accuracy, broad device compatibility, and 180
08-08 03:29groundScott already holds both sides of this case: Capability Audit requires representative, vendor-neutral testing rather than demo claims, while Hardware-aware local inference and gamepc document active w
08-08 03:26promote_anchororigin walk conf 0.99
08-08 03:25createA usable implementation and source package create a concrete local-inference episode, though its exceptional speed claim remains unvalidated.