teensy-vad-v8 — the 500k scaling ablation

teensy-vad-v8 asks one question: does 10–14× the parameters and 6.6× the training data beat the 42k-parameter v7 champion? No — and the reasons are the most useful result of the campaign. Four ~500k-class models (3-layer GRU-192, 4-layer causal transformer, 2-hidden MLP "a3", each distilled from Silero on 660 hours of prior-balanced mixtures) were trained with quantization-aware fine-tuning and benchmarked on the family's calibrated real-world protocol.

Results (human-labelled real audio, calibrated protocol)

model	params	KB	TEN F1	TEN AUC	AMI F1	AMI AUC	µs/20ms
v8-GRU-192×3	582k	2,285	0.8503	0.7843	0.9191	0.9227	138.8
v8-transformer d128×4	570k	2,229	0.8726	0.7255	0.9234	0.8969	84.7
v8-a3 MLP (float)	549k	2,157	0.9122	0.8844	0.8949	0.8880	75.0
v8-a3 MLP (int8 QAT)	549k	554	0.9133	0.8949	0.8852	0.8889	281.9*
v7-GRU96 (champion)	42k	169	0.8992	0.8934	0.9153	0.9182	37.9
Silero VAD (1.77M)	1,774k	2,200	0.9398	0.9514	0.7136	0.8938	90.2
WebRTC VAD	~6k	~50	n/a	n/a	0.8419	0.7602	1.8

* the numpy int8 matmul runtime is unoptimized; the int8 artifact itself is 4× smaller than float (554 KB vs 2.1 MB).

Findings

1. Scale did not beat fit. No v8 model dominates the 42k v7 champion. The GRU-192 and transformer do edge it on AMI rooms (F1 0.919–0.923), but their clean near-mic ranking collapsed (TEN AUC 0.73–0.78 vs 0.89).
2. The context-MLP scaled best. a3-int8 matches v7's TEN AUC (0.8949 vs 0.8934) with the family's best TEN F1 (0.9133) — at 13× its size.
3. QAT won. int8 fine-tuning improved TEN ranking over float (0.8844 → 0.8949) and held AMI. QAT is validated at 500k scale.
4. Recurrence > attention at every tested scale for clean-audio ranking.
5. Under-training is the honest caveat: the compute budget allowed only ~1.8 passes over 660 h (the deep models likely need 5+). Scale may still win — but only with proportional compute.

Files

file	use
teensy-v8-gru192.npz	3-layer GRU-192, numpy runtime weights (2.2 MB)
teensy-v8-tt.npz	causal transformer d128×4 (2.2 MB)
teensy-v8-a3.npz	context-MLP a3, float (2.1 MB)
teensy-v8-a3-qat.npz	context-MLP a3, int8 QAT (554 KB)
chart_v8.png	comparison chart

License & data

Weights: CC BY 4.0 — commercial use with attribution (© 2026 Pankaj Doharey / Metacritical, TeensyVAD by VoxLogic). Data: LibriSpeech 960 h train subsets (CC BY 4.0), MUSAN (CC BY 4.0), AMI ambience (CC BY 4.0), Silero teacher (MIT). Code: MIT.

Citation

@software{doharey2026teensyvadv8,
  title  = {TeensyVAD-v8: A 500k-Parameter Scaling Ablation for
            Tiny Voice Activity Detection},
  author = {Doharey, Pankaj},
  year   = {2026},
  url    = {https://huggingface.co/Teensy/teensy-vad-v8}
}