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.

Qwen3 ForcedAligner 0.6B · Browser

Align an existing transcript to mono audio and estimate word/character start and end times using the official non-autoregressive 5000-class timestamp head. Does not transcribe unknown speech.

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

Align an existing transcript to mono audio and estimate word/character start and end times using the official non-autoregressive 5000-class timestamp head. Does not transcribe unknown speech.

License
Apache-2.0
Model released
Jan 29, 2026
Source repository created
Jan 28, 2026
Model weights · Q4
1.0 GB

Best for

  • Correct subtitle word timing against an existing transcript
  • Inspect local transcript/audio alignment with human review

Keep in mind

  • Requires an accurate supplied transcript; wrong or missing words can receive plausible but incorrect timestamps. Not transcription or speech recognition.
  • Adapter accepts up to30seconds of normalized mono16kHz audio and4000Unicode characters. These are conservative application limits, not the upstream five-minute maximum.
  • Upstream accepts11languages, but local quality evidence is an English pilot from one speech recording with clean/noise/shift conditions. No broad multilingual accuracy claim.
  • Timestamp quantization is80ms. Zero-duration spans, overlaps, incorrect word splitting and boundary errors remain possible; review subtitle timing.
  • Reference quality metric measures agreement with saved ElevenLabs Scribe v2 boundaries, not human ground truth.
  • No incremental alignment output is exposed by the pinned nonautoregressive forward/decode API; generateStream is unavailable.
  • No reliable speech-presence or transcript-truth guarantee: silence or mismatched transcripts may produce fabricated timing. Not suitable for identity, forensic or high-impact decisions.
  • Browser support is limited to English transcript splitting and measured Q4 configurations. Languages accepted by the desktop adapter are not browser support.
  • WebGPU may delegate shape/unsupported nodes to CPU; no full-acceleration claim. Q4 graph is about1.05GB and needs additional browser/GPU memory.
  • Original MIT-licensed upstream timestamp monotonicity repair is applied after argmax. Short anomalies snap to neighbors; longer anomalies interpolate, so repaired boundaries require review.

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
https://huggingface.co/Qwen/Qwen3-ForcedAligner-0.6B
Release verification
Verified
Release evidence
Official January29 release announcement explicitly names ForcedAligner0.6B. Official HF-native adaptation released June26 is an artifact adaptation, not a new original checkpoint release.
Source updated
Jan 30, 2026
Browser model files
valoomba/Qwen3-ForcedAligner-0.6B-ONNX
Files updated
Mar 31, 2026
Artifact revision
261c9ed100c1b18a4a1fbc488e05625dc9a4ae5c

Input

AUDIO

Sample result

Qwen3 ForcedAligner 0.6B · Browser
Record a clip, then run inference. This does not send live audio to the model.

Settings

Only English transcript splitting is implemented and tested in this browser adapter. English is the sole accepted selection; this does not translate text or enable other languages.

Output

Explore the tone of your audio

Choose a recording or try the sample to compare vocal emotion scores.

Benchmark evidence

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

Qwen3 ForcedAligner 0.6B · Browser

2 configurations tested
Passed2
Failed0
TestedNov 3, 2026
Runtimeonnxruntime-web 1.24.3

Quality criteria

{ "scope": "English single-source clean/noise/leading-silence pilot, independent saved Scribe v2 boundary reference; not broad multilingual/human-ground-truth accuracy", "exactWordCoverage": true, "finiteNonnegativeMonotonicTimes": true, "endToleranceSeconds": 0.24, "maximumZeroDurationFraction": 0.1, "maximumBoundaryMAESeconds": 0.2, "maximumRealTimeFactor": 1, "performanceGrades": { "excellentMaxRTF": 0.1, "goodMaxRTF": 0.5, "acceptableMaxRTF": 1 }, "frozenFrom": "Native ForcedAligner onboarding gates reused before browser acceptance runs", "selection": "Qwen original2026-01-29 release; requested0.6Baligner, exact communityQ4graph. Transformers.js4.2.0 lacks exact architecture mapping; use its tokenizer/Whisperfeatures with pinned ONNXRuntimeWeb1.24.3 model-local inference. Pre2024Wav2Vec2 not eligible new addition; existing transcription timestamps are not forced alignment." }

PrecisionDeviceResult
Q4GPU · WebGPUAcceleratedPassedVery fast
Q4CPU · WASMPassedModerate
Test setup & sample details
Quality gate
{ "scope": "English single-source clean/noise/leading-silence pilot, independent saved Scribe v2 boundary reference; not broad multilingual/human-ground-truth accuracy", "exactWordCoverage": true, "finiteNonnegativeMonotonicTimes": true, "endToleranceSeconds": 0.24, "maximumZeroDurationFraction": 0.1, "maximumBoundaryMAESeconds": 0.2, "maximumRealTimeFactor": 1, "performanceGrades": { "excellentMaxRTF": 0.1, "goodMaxRTF": 0.5, "acceptableMaxRTF": 1 }, "frozenFrom": "Native ForcedAligner onboarding gates reused before browser acceptance runs", "selection": "Qwen original2026-01-29 release; requested0.6Baligner, exact communityQ4graph. Transformers.js4.2.0 lacks exact architecture mapping; use its tokenizer/Whisperfeatures with pinned ONNXRuntimeWeb1.24.3 model-local inference. Pre2024Wav2Vec2 not eligible new addition; existing transcription timestamps are not forced alignment." }
Tested
2026-11-03T18:16:36.504Z
Environment
{ "browser": "156.0.8078.80", "os": "macOS27.2", "hardware": "Apple M3 Max", "runtime": "ONNXRuntimeWeb1.24.3/Transformers.js4.2.0 preprocessing" }
Runtime
onnxruntime-web 1.24.3
Artifact revision
261c9ed100c1b18a4a1fbc488e05625dc9a4ae5c
Fixture
{
  "file": "model-samples/desktop/qwen3-forced-aligner/clean.wav",
  "url": "https://assets.localai.tools/samples/v1/files/ba1a561773d6b6e55c7b9d326d9aa792f6d3a6f430556b83a78f78a2a4cb25e1/clean.wav",
  "size": 352044,
  "sha1": "ce87ae892606eecb8ef65803ed7c4d1ea13ca437",
  "sha256": "ba1a561773d6b6e55c7b9d326d9aa792f6d3a6f430556b83a78f78a2a4cb25e1"
}
Rerun when
Exact artifact/runtime revision; Audio/transcript/reference checksum or quality gates; Task preprocessing/decoding or backend behavior
Per-case quality & failure details
[
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumBoundaryMAESeconds": 0.05590909090909099,
      "reference": "Saved Scribe v2 agreement, not human ground truth"
    },
    "runtime": {
      "engine": "onnxruntime-web",
      "version": "1.24.3",
      "artifactRevision": "261c9ed100c1b18a4a1fbc488e05625dc9a4ae5c",
      "loadMs": 6374.9000000059605,
      "inferenceMs": 450.5,
      "totalMs": 6825.4000000059605,
      "realTimeFactor": 0.04019090909036723,
      "grade": "excellent",
      "acceleration": "partial",
      "transport": "Local verified immutable snapshot; load excludes network download",
      "browser": "156.0.8078.80"
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumBoundaryMAESeconds": 0.05590909090909099,
      "reference": "Saved Scribe v2 agreement, not human ground truth"
    },
    "runtime": {
      "engine": "onnxruntime-web",
      "version": "1.24.3",
      "artifactRevision": "261c9ed100c1b18a4a1fbc488e05625dc9a4ae5c",
      "loadMs": 5412.800000011921,
      "inferenceMs": 8402.200000017881,
      "totalMs": 13815.000000029802,
      "realTimeFactor": 0.7536000000005183,
      "grade": "acceptable",
      "acceleration": "cpu",
      "transport": "Local verified immutable snapshot; load excludes network download",
      "browser": "156.0.8078.80"
    }
  }
]

Something not quite right?

Tell us what happened with Qwen3 ForcedAligner. 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.