Biologically Inspired Predictive Gating Signed-Dyadic Coding for Mandarin Keyword Spotting

Zijian Zhou; Rui Liu; Rongxiang Nie; Rui Ma; Ziyao Zhang; Xin Du

cs.SDOct 30, 2026

Biologically Inspired Predictive Gating Signed-Dyadic Coding for Mandarin Keyword Spotting

Authors: Zijian Zhou, Rui Liu, Rongxiang Nie, Rui Ma, Ziyao Zhang, Xin Du

Organizations: Sivy Limited, Hangzhou, China · The Cavendish Laboratory, Department of Physics, University of Cambridge, Cambridge, UK

Abstract

In continuous spiking keyword spotting, temporally correlated acoustic frames can repeatedly trigger memory access and neural computation at a dense input interface, even when hidden activity is sparse. Predictive processing and sensory adaptation suggest retaining predictable input locally and responding selectively to deviations. Inspired by this mechanism, we propose Predictive Gating Signed-Dyadic Coding (PG). PG maintains a local prediction for each input channel and transmits an event only when the residual exceeds a learned threshold. Each event carries a sign and a finite power-of-two magnitude, and the receiver reconstructs it with fixed-point shifts and additions. In Mandarin keyword spotting, INPUT-PG reaches 90.00% accuracy while reducing mean SRAM reads by 33.87% and inference cycles by 32.77%. PG provides a hardware interface for research on low-power continuous speech processing and event-based sensory systems.

Figures & tables

Figure 1: Predictive Gating Signed-Dyadic Coding (PG) and its interface with keyword classification. PG maintains a shared prediction, transmits thresholded signed-dyadic residuals, and reconstructs the input before the Deep-SNN. Temporal aggregation produces the keyword scores.

Figure 2: Selective PG placement and boundary semantics. DIRECT has no PG, INPUT-PG places PG before the first linear map, and ALL-PG places PG before all learned maps. Small residuals remain silent, whereas significant mismatches emit signed events; hidden LIF transitions are shown schematically.

Track	Question	Population	Primary outputs	Scope
E1 from scratch, 50 epochs	Task performance	2 architectures, 3 seeds, 720 clips	accuracy, activity proxy	software; rate/TTFS on Deep-SNN
E5 routed	Energy cost of work	Deep-SNN, seed 17, clip 3826	cycles, reads, power, energy	one fixed clip, after routing
Integer inference	Accuracy and workload	Deep-SNN, seed 17, 720 clips	accuracy, cycles, reads	integer software; frozen weights
Small movable	Separate frontier	$40\rightarrow 64\rightarrow 5$ , seed 17	Q9 accuracy, routed energy	separate microarchitecture

Table 1: Evaluation tracks and auxiliary study. Each row lists its population, primary outputs, and scope.

Architecture	Variant	Accuracy (%)	Input/coded events	Hidden spikes	Complete proxy	$\Delta$ proxy
Deep-SNN	DIRECT	$88.84\pm 1.55$	0.0	19,344.3	29,276.8	0.00%
	INPUT-PG (ours)	$88.47\pm 3.70$	2,806.5	20,595.0	23,401.5	$\mathbf{-20.07\%}$
	ALL-PG (ours)	$88.66\pm 0.45$	27,788.7	diagnostic	27,788.7	$-5.08\%$
	RATE-T8	88.61 $\pm$ 1.77	12,867.6	19,290.2	32,157.8	+9.84%
	TTFS-T8	86.94 $\pm$ 1.25	7,610.9	19,502.7	27,113.6	$-$ 7.39%
LIF-TCN	DIRECT	$88.75\pm 2.57$	0.0	26,283.6	36,216.1	0.00%

Table 2: Software results after 50-epoch training over seeds 7, 17, and 27. Accuracy uses validation-selected checkpoints; activity counts are means per clip over the three seeds. RATE-T8 and TTFS-T8 are input-only Deep-SNN baselines. Input/coded event counts follow the definitions in Methods, and the complete proxy is the cross-variant workload measure.

Architecture	Comparison	Mean [95% CI]
Deep-SNN	INPUT-PG (ours)–DIRECT	$-$ 0.37 [ $-$ 5.73, 4.99]
Deep-SNN	ALL-PG (ours) –DIRECT	$-$ 0.19 [ $-$ 3.99, 3.62]
Deep-SNN	RATE-T8–DIRECT	$-$ 0.23 [ $-$ 2.34, 1.88]
Deep-SNN	TTFS-T8–DIRECT	$-$ 1.90 [ $-$ 3.29, $-$ 0.50]
LIF-TCN	INPUT-PG (ours)–DIRECT	$-$ 0.65 [ $-$ 6.73, 5.43]
LIF-TCN	ALL-PG (ours) –DIRECT	$-$ 1.90 [ $-$ 8.27, 4.48]

Table 3: Paired accuracy differences relative to DIRECT. Intervals are 95% paired Student-

t

intervals over three seeds and are reported in percentage points.

Figure 3: Software accuracy and complete communication activity. Results use 50-epoch training over seeds 7, 17 and 27 on 720 test clips; whiskers show sample SD and dashed lines show DIRECT accuracy. RATE-T8 and TTFS-T8 are input-only Deep-SNN baselines. Panels (c-d) show mean ALL-PG events by boundary. Complete activity follows Eqs. ( 8 )–( 9 ) and is reported as a communication proxy.

Variant	Cycles	SRAM reads	Inference dyn. (mW)	Resident ( $\mu$ J)	Transported ( $\mu$ J)	Area (mm 2 )
DIRECT	11,792,327	5,737,243	273.645	33,252.992	33,290.587	43.785
INPUT-PG (ours)	7,810,745	3,746,452	277.010	20,122.617	20,159.901	44.222
ALL-PG (ours)	9,158,879	4,420,519	278.367	24,943.967	24,979.233	44.369
RATE-T8	16,155,227	7,918,693	269.362	44,365.536	44,400.672	43.638
TTFS-T8	10,639,233	5,160,696	270.449	28,369.628	28,405.549	43.704

Table 4: Deep-SNN results after routing for seed 17, frozen clip 3826 (232 frames). Sky130HD TT, 25 ∘ C, 1.8 V, nominal 10 MHz. Energy is incremental energy after subtracting the idle baseline; Resident excludes weight load and Transported includes it. RATE-T8 and TTFS-T8 use the implementations applied only at the input, as defined in the experimental protocol.

Variant	Q9 acc. (%)	Cycles	Resident ( $\mu$ J)	Transported ( $\mu$ J)
DIRECT	83.33	1,374,251	216.633	216.679
RATE-T8	81.53	2,443,449	337.234	337.282
TTFS-T8	82.92	1,102,571	179.086	179.131
Previous-K6 PG (ours)	79.72	703,803	83.651	83.692

Table 5: Movable-SRAM experiment with a small network. Q9 accuracy uses 720 test clips; cycles and energy use one frozen clip in the movable-SRAM architecture.

Method	Accuracy (%)	Input activity/clip	SRAM reads/clip	Inference cycles/clip
DIRECT	89.58	9,932.44	5,054,116.01	10,448,419.24
INPUT-PG (ours)	90.00	2,813.53	3,342,360.28	7,024,907.79
ALL-PG (ours)	89.58	2,932.49	5,016,142.84	10,372,472.91

Table 6: Integer software inference on 720 test clips with frozen Deep-SNN weights from seed 17. Workload columns report means per clip and exclude parameter loading. Input activity counts dense values for DIRECT and emitted input events for PG.

Figure 4: Routed energy estimates and integer software accuracy. Panels (A-E) use Deep-SNN seed 17 and clip 3826 (232 frames) after Sky130HD routing at 1.8 V, 25 ∘ C and nominal 10 MHz. Resident energy is incremental dynamic energy after idle subtraction and excludes weight loading; power is active inference dynamic power. Gate activity is simulated without propagation delays, with the measured timing and DRC status reported for the prototype. Panel (F) gives integer software accuracy on 720 test clips with frozen seed-17 weights.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Variant	Cycles	SRAM reads	Inference power
INPUT-PG (ours)	$-33.76\%$	$-34.70\%$	$+1.23\%$
ALL-PG (ours)	$-22.33\%$	$-22.95\%$	$+1.73\%$
RATE-T8	+37.00%	+38.02%	-1.57%
TTFS-T8	-9.78%	-10.05%	-1.17%

Appendix

Table A1: Deep-SNN changes relative to DIRECT for the experiment using one fixed clip in Table 4 .

Architecture	INPUT-PG (ours)–DIRECT	ALL-PG (ours)–DIRECT
Deep-SNN	$-0.37$	$-0.19$
LIF-TCN	$-0.65$	$-1.90$

Appendix

Table A2: Mean paired accuracy differences for PG placements (percentage points), summarized from Table 3 .

Spike-Based Signal Encoding

See all 24 papers

Adaptive Speech-to-Spike Encoding for Spiking Neural Networks
Jun 17, 2026Taharim Rahman Anon, Jakaria Islam EmonKeyword Spotting Speech Processing
1PI LLC, Sapporo, Hokkaido, Japan
Analog-Friendly Predictive Coding without Activation Derivatives
Sep 26, 2026Francesco InnocentiAnalog Computing Predictive Coding
Brain Network Dynamics Unit University of Oxford, UK
Neuromorphic visual attention for Sign-language recognition on SpiNNaker
May 7, 2026Sarka Liskova, Olha Vedmedenko, Mazdak Fatahi +3Event-Based Vision Visual Attention
Dept. of Cybernetics, Faculty of Electrical Engineering, Czech Technical University in Prague, Czech Republic · Faculty of Information Technology, Czech Technical University in Prague, Czech Republic

Jun 17, 2026J/K move · Enter open · S save

Adaptive Speech-to-Spike Encoding for Spiking Neural Networks

Taharim Rahman Anon, Jakaria Islam Emon

1PI LLC, Sapporo, Hokkaido, Japan

Keyword Spotting Speech Processing Spiking Neural Networks Neuromorphic Computing Spike-Based Signal Encoding

Open paper AskPDF arXiv

The mismatch between continuous acoustic signals and discrete event-driven processing remains a fundamental bottleneck for neuromorphic speech processing. Current systems typically rely on fixed spike encoders, forcing downstream Spiking Neural Networks (SNNs) to compensate for non-adaptive input representations. To address this, we present a learnable residual speech-to-spike encoder jointly trained end-to-end with a Recurrent Leaky Integrate-and-Fire (R-LIF) backbone. We validate this approach on the Google Speech Commands v2 (GSC-v2) benchmark, achieving up to 94.97% accuracy. Notably, the learned encoder remains highly parameter-efficient with a compact 35k-parameter variant that reaches 89.8%, matching or exceeding prior baselines that require an order of magnitude more parameters. Our encoder-focused analysis, including linear probing and gradient-residual inspection, indicates that the encoder does not target faithful signal reconstruction but instead learns task-aligned spike representations that enhance class separability. Finally, we benchmark bio-inspired, hardware-friendly credit assignment by comparing Direct Feedback Alignment (DFA) with surrogate-gradient BPTT under identical architectures and training conditions. We find that DFA reaches 91.5% accuracy, quantifying the performance trade-off of bio-inspired learning rules for modern neuromorphic audio.

Abstract

Spike-Based Signal Encoding

Explore similar work

Adaptive Speech-to-Spike Encoding for Spiking Neural Networks

Analog-Friendly Predictive Coding without Activation Derivatives

Neuromorphic visual attention for Sign-language recognition on SpiNNaker