Upload folder using huggingface_hub

README.md

@@ -1,3 +1,94 @@
 ---
 license: apache-2.0
+base_model: FireRedTeam/FireRedVAD
+pipeline_tag: voice-activity-detection
+tags:
+- voice-activity-detection
+- vad
+- dfsmn
+- audio
+- speech
+- safetensors
+- svod
 ---
+
+# FireRedVAD (non-streaming) — safetensors for Svod
+
+The non-streaming voice-activity-detection checkpoint of
+[FireRedVAD](https://github.com/FireRedTeam/FireRedVAD)
+([FireRedASR2S technical report, arXiv:2603.10420](https://arxiv.org/abs/2603.10420)),
+converted from the original `model.pth.tar` + `cmvn.ark` to a single
+safetensors file for the [Svod](https://github.com/npatsakula/svod) inference
+runtime. Weights are unmodified f32 — only renamed, reshaped, and bundled with
+the CMVN statistics.
+
+FireRedVAD is a feed-forward DFSMN (no recurrence): 8 FSMN layers with
+depthwise lookback/lookahead temporal filters, 588k parameters, operating on
+80-bin kaldi log-mel fbank at 16 kHz (25 ms window / 10 ms hop). It detects
+speech in 100+ languages; upstream reports 97.57 F1 on FLEURS-VAD-102.
+
+## Files
+
+| file | contents |
+|---|---|
+| `firered_vad.safetensors` | 45 model tensors + `cmvn_means`/`cmvn_istd` `[80]` (global CMVN derived from `cmvn.ark`: variance floored at 1e-20, inverse std precomputed) |
+| `golden.safetensors` | parity-test reference for `assets/hello_zh.wav`: `samples` (16 kHz mono, `[-1, 1]` f32), `feat` (pre-CMVN `kaldi-native-fbank` output `[230, 80]`), `probs` (PyTorch `DetectModel` output `[230]`) |
+
+## Architecture / config
+
+`idim=80, R=8, M=1, H=256, P=128, N1=20, S1=1, N2=20, S2=1, odim=1`
+(read from the checkpoint's embedded args).
+
+Tensor schema: `fc1.{weight,bias}` (80→256), `fc2.{weight,bias}` (256→128),
+`fsmn1.{lookback,lookahead}.weight` `[128, 1, 1, 20]`,
+`blocks.{0..6}.{fc1.weight,fc1.bias,fc2.weight,lookback.weight,lookahead.weight}`,
+`dnn.{weight,bias}` (128→256), `out.{weight,bias}` (256→1),
+`cmvn_means`/`cmvn_istd` `[80]`. FSMN filters are reshaped from PyTorch's
+`[P, 1, 20]` to `[P, 1, 1, 20]` for 2-D depthwise convolution.
+
+## Provenance & verification
+
+Produced by [`scripts/convert_firered_vad.py`](https://github.com/npatsakula/svod/blob/main/scripts/convert_firered_vad.py):
+
+```bash
+uv run scripts/convert_firered_vad.py \
+    --checkpoint model.pth.tar \
+    --cmvn cmvn.ark \
+    --selfcheck \
+    --golden hello_zh.wav
+```
+
+Verified against the PyTorch reference (CPU, f32):
+
+- FSMN asymmetric-padding formulation vs reference symmetric-pad-then-slice:
+  bit-identical (`--selfcheck`, max |Δ| = 0).
+- Svod device model on golden features vs `DetectModel` probs: max |Δ| ≈ 1.2e-7.
+- End-to-end (Svod fbank → model) vs reference probs: max |Δ| ≈ 1.8e-7.
+
+## Usage (Svod)
+
+```rust
+use svod_model::firered_vad::{FireRedVad, FireRedVadSplitter};
+
+// Downloads firered_vad.safetensors from this repo:
+let splitter = FireRedVadSplitter::from_hub()?;
+// Or from a local file:
+let model = FireRedVad::from_safetensors(path)?;
+```
+
+`golden.safetensors` is consumed by Svod's parity tests
+(`cargo test -p svod-model firered -- --ignored`).
+
+## License & citation
+
+Apache-2.0, matching the upstream
+[FireRedVAD](https://github.com/FireRedTeam/FireRedVAD) release by Xiaohongshu.
+
+```bibtex
+@article{xu2026fireredasr2s,
+  title={FireRedASR2S: A State-of-the-Art Industrial-Grade All-in-One Automatic Speech Recognition System},
+  author={Xu, Kaituo and Jia, Yan and Huang, Kai and Chen, Junjie and Li, Wenpeng and Liu, Kun and Xie, Feng-Long and Tang, Xu and Hu, Yao},
+  journal={arXiv preprint arXiv:2603.10420},
+  year={2026}
+}
+```

firered_vad.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b2b9577b77d85bac630535021dca88d56d618876070bce54850dcef4d7c09dc
+size 2358420

golden.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:973d2580dc755004383fa5c6f3a5d3c1d0e41dd558a8d2ccebbf1e31971f3d6a
+size 223216