A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

NVIDIA Parakeet TDT 0.6B v3

Transcribe speech locally in a browser using NVIDIA’s multilingual 0.6B FastConformer TDT model. Returns the complete parakeet.js transcript, word-level and token timestamps, confidence scores and metrics.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Transcribe speech locally in a browser using NVIDIA’s multilingual 0.6B FastConformer TDT model. Returns the complete parakeet.js transcript, word-level and token timestamps, confidence scores and metrics.

Created by
NVIDIA
License
CC-BY-4.0
Model released
Aug 14, 2025
Source repository created
Aug 4, 2025
Model weights · FP16
1.3 GB

Best for

  • Local transcription and caption prototypes
  • Speech search and dictation in the original model’s 25 European languages

Keep in mind

  • Recognition and automatic language detection can fail on accents, noisy or quiet audio, music, overlapping speakers and out-of-domain vocabulary. Only the checked-in English fixtures have measured quality; upstream language coverage is not per-language validation.
  • Requires normalized mono 16 kHz Float32 PCM, at least 400 samples (25 ms). Long recordings are constrained by available browser memory and processing time. This adapter does not implement streaming, diarization, translation or speaker identification.
  • TDT word times are estimated acoustic boundaries; confidence values are uncalibrated. Review transcripts before use in medical, legal or other consequential settings.
  • Large model downloads and compilation need substantial storage and memory. Measurements apply only to the recorded browser/hardware. WebGPU candidates use a WASM decoder and may partition encoder work onto WASM; they are not fully accelerated.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
nvidia/parakeet-tdt-0.6b-v3
Release verification
Verified
Release evidence
Exact checkpoint model card Release Date section states Huggingface 08/14/2025; repository creation is recorded separately.
Source updated
Aug 5, 2026
Browser model files
ysdede/parakeet-tdt-0.6b-v3-onnx
Files updated
Oct 25, 2026
Artifact revision
e5356bf6e24bdb7a269536b0e9e7869bc034b893

Input

AUDIO

Sample result

NVIDIA Parakeet TDT 0.6B v3
Record a clip, then run inference. This does not send live audio to the model.

Settings

Include native word/token times. Enabled includes estimated word/token boundaries; disabled omits token times and leaves the native word timing fields at zero.
Advanced settings 2
Scale token logits before computing confidence scores. Lower values sharpen confidence scores and higher values flatten them. The recognized words and timestamps are unchanged.
Include native uncalibrated token and word confidence estimates. Enabled includes confidence estimates; disabled omits detailed scores. Scores are not probabilities of correctness.

Output

Your transcript or answer will appear here

Choose the input task, then add a WAV clip or text, or try a named sample.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

NVIDIA Parakeet TDT 0.6B v3

6 configurations tested
Passed2
Failed4
TestedNov 3, 2026
Runtimeparakeet.js 1.4.4

Quality criteria

No explicit quality gate recorded. Review per-case evidence; full certification is pending.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedVery fast
INT8CPU · WASMPassedFast
INT8GPU · WebGPUFailed—
FP16CPU · WASMFailed—
FP32CPU · WASMFailed—
FP32GPU · WebGPUFailed—
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-03T12:55:44.724Z
Environment
{"os":"Darwin","release":"27.2.0","arch":"arm64","cpu":"Apple M3 Max","memoryBytes":103079215104,"browser":"156.0.8078.80"}
Runtime
parakeet.js 1.4.4
Artifact revision
e5356bf6e24bdb7a269536b0e9e7869bc034b893
Fixture
{
  "file": "model-samples/vosk-jfk.flac",
  "url": "https://assets.localai.tools/assets/benchmarks/vosk/63a4b1e4c1dc655ac70961ffbf518acd249df237e5a0152faae9a4a836949715/vosk-jfk.flac",
  "size": 1152693,
  "sha1": "537ddf75dff344f8d31c1cd6ebee23ee8362d921",
  "durationSeconds": 11,
  "sampleRate": 16000
}
Rerun when
model artifact revision change; runtime version change; fixture or quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true
    },
    "runtime": {
      "loadMs": 4460.199999988079,
      "inferenceMs": 15651.899999976158,
      "totalMs": 20367.19999998808,
      "realTimeFactor": 0.3865406343183241,
      "grade": "good",
      "execution": "WASM encoder and decoder",
      "fullyAccelerated": false
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "passed": false
    },
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 21459.29999998212,
      "grade": "failed",
      "execution": "Hybrid: encoder WebGPU with WASM partitioning permitted; decoder WASM",
      "fullyAccelerated": false
    },
    "failure": "Error: Speech quality gate failed"
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": false
    },
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 5642.700000017881,
      "grade": "failed",
      "execution": "WASM encoder and decoder",
      "fullyAccelerated": false
    },
    "failure": "Error: [WorkerError:runtime_error] init: Can't create a session. ERROR_CODE: 6, ERROR_MESSAGE: std::bad_alloc"
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true
    },
    "runtime": {
      "loadMs": 6605.0999999940395,
      "inferenceMs": 1423.9000000059605,
      "totalMs": 8280.09999999404,
      "realTimeFactor": 0.035164753749321424,
      "grade": "excellent",
      "execution": "Hybrid: encoder WebGPU with WASM partitioning permitted; decoder WASM",
      "fullyAccelerated": false
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": false
    },
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 3741.7000000178814,
      "grade": "failed",
      "execution": "WASM encoder and decoder",
      "fullyAccelerated": false
    },
    "failure": "Error: [WorkerError:runtime_error] init: The requested file could not be read, typically due to permission problems that have occurred after a reference to a file was acquired."
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "fail",
    "quality": {
      "passed": false
    },
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 3113.899999976158,
      "grade": "failed",
      "execution": "Hybrid: encoder WebGPU with WASM partitioning permitted; decoder WASM",
      "fullyAccelerated": false
    },
    "failure": "Error: [WorkerError:runtime_error] init: The requested file could not be read, typically due to permission problems that have occurred after a reference to a file was acquired."
  }
]

Something not quite right?

Tell us what happened with Parakeet TDT 3. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.