tonythethompson/Qwen3-ForcedAligner-0.6B-ONNX

Qwen3-ForcedAligner-0.6B (Q4 ONNX) — Trackdub Mirror

Q4 ONNX export of Qwen/Qwen3-ForcedAligner-0.6B, sourced from aloomba/Qwen3-ForcedAligner-0.6B-ONNX.

Hosted here for stable, checksum-verified distribution by Trackdub (local-first AI dubbing workstation).

Model

• Architecture: Encoder (24L, d=1024) + 28L decoder body, non-autoregressive
• Output: Word-level timestamps via slot injection; 5000-class softmax × 80ms resolution
• Max audio duration: 5 minutes
• Inference: Single forward pass
• Languages: 11 (English, Chinese, Cantonese, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish)

Files

File	Size	SHA256
onnx/model_q4.onnx	~1000 MB	59b528896d70b34e57838e160d16d5f7cfc02d86c7c6ad46cdc57c25c15497b7
config.json	—	—
okenizer.json	—	—
ocab.json	—	—
merges.txt	—	—
preprocessor_config.json	—	—
okenizer_config.json	—	—
special_tokens_map.json	—	—
dded_tokens.json	—	—

Note: model_q4.onnx is self-contained (no external data sidecar required).

Usage in Trackdub

Used as the primary forced alignment provider in Trackdub's lip-sync stage (ngine_family: onnx-qwen-forced-aligner). Not intended for standalone use from this mirror.

License

Apache 2.0 — see original model card.