Days 30-28Nov 1-3Live

Telephony-native voice anti-spoofing for Gnani.ai

A phone line costs published anti-spoofing models 12 to 16 EER points on codecs they never saw, and a threshold fixed on dev data does not survive the trip.

Data: REAL audio (ASVspoof 2019 LA clips, English proxy), SIMULATED phone channel (real ffmpeg codecs)

The problem

Gnani's Armour365 verifies callers by voice. A cloned voice made with TTS or voice conversion reaches it through an 8 kHz phone line, and codecs such as G.711, AMR-NB and GSM strip many of the cues a countermeasure relies on.

This build asks three questions: how much does the narrowband channel hurt a countermeasure, does a threshold fixed on dev survive it, and can an IVR flow on Gnani Prisma STT and Timbre TTS state an honest operating point.

Every clip is VAD-trimmed and loudness-normalised, then sent through 14 telephony conditions with real ffmpeg codecs. Four conditions (AMR-NB 4.75k, GSM-FR and two two-hop chains) are held out of training.

Key result

Codec augmentation holds the line: held-out-codec EER 34.42% with it vs 41.84% without (same LCNN). Best published model, SSL+AASIST: 9.15% clean, 20.97% on held-out codecs.

Equal error rate, clean vs held-out codecs (lower is better)

Equal error rate, clean vs held-out codecs (lower is better)clean poolheld-out codecsLFCC-GMM 8 kHz, codec-aug25.5933.63LFCC-LCNN 8 kHz, codec-aug27.6434.42LFCC-LCNN, no codec aug (ablation)26.5941.84LFCC-GMM clean 16 kHz (wideband control)24.1044.73AASIST (published SOTA)23.6539.22AASIST-L (published SOTA)21.5737.13SSL+AASIST (published SOTA)9.1520.97XLS-R teacher, fine-tuned at 8 kHz23.3421.63Distilled LCNN student26.7934.6301020304050
Source: REPORT.md section 3. The clean pool is clean 8 kHz (nb_pcm) for the 8 kHz models (LFCC models, teacher, student) and clean 16 kHz for the published SOTA models and the wideband control. SOTA and teacher scored on Modal GPUs; everything else on CPU.

Comparison with baselines

Countermeasures on ASVspoof 2019 LA eval (2,560 utterances, unseen attacks A07 to A19)

ModelEER cleanEER seenEER held-out [95% CI]minDCF held-outFRR@FAR1 held-outdelta-NB
LFCC-GMM 8 kHz, codec-aug25.5930.3833.63 [32.29, 34.81]0.78199.28+8.04
LFCC-LCNN 8 kHz, codec-aug27.6430.7034.42 [33.04, 35.60]0.79992.47+6.78
LFCC-LCNN, no codec aug (ablation)26.5935.1741.84 [40.63, 42.86]0.94398.45+15.25
LFCC-GMM clean 16 kHz (wideband control)24.1041.6344.73 [43.53, 45.97]0.998100.00+20.63
AASIST (published, 16 kHz)23.6537.3039.22 [37.63, 40.20]0.85592.17+15.57
AASIST-L (published, 16 kHz)21.5734.3237.13 [35.89, 38.35]0.86996.78+15.56
SSL+AASIST (published, 16 kHz)9.1518.4520.97 [19.93, 21.96]0.57271.60+11.83
XLS-R teacher, fine-tuned at 8 kHz23.3424.2321.63 [20.03, 22.71]0.3817.35 (FAR 27.0)-1.71
Distilled LCNN student26.7930.5834.63 [33.35, 35.86]0.83691.17+7.84

EER, FRR in %. delta-NB = EER(held-out) minus EER(clean pool), same utterances. FRR@FAR1 uses the threshold fixed on dev at FAR 1%; for the teacher that threshold lets 27% of spoofs through on held-out codecs. SOTA and teacher rows: Modal GPU runs (about $9.42 in total, including a first teacher run that did not learn).

IVR flow on real eval clips at dev-fixed operating points (200 genuine + 200 attack clips per codec)

CodecOperating pointGenuine acceptedAttack accepted
G.711 (seen)far132.5%11.0%
G.711 (seen)min_dcf88.5%62.5%
G.711 (seen)eer84.5%49.0%
AMR-NB 4.75k (held out)far14.5%1.0%
AMR-NB 4.75k (held out)min_dcf75.5%55.5%
AMR-NB 4.75k (held out)eer57.5%33.5%

CM gate only: echo STT, identity gate bypassed. No operating point meets genuine >= 90% with attack <= 5%. Median local latency about 30 ms.

Other findings

What this does not show

How to reproduce

From the repository root (commands from its README; Windows PowerShell unless noted).

.\run.ps1 setup        # venv, CPU torch, requirements, clones AASIST and SSL_Anti-spoofing
.\run.ps1 quick        # under 5 minutes: unit tests, teacher smoke, IVR demo on real clips, tables
.\run.ps1 trials       # IVR flow on real ASVspoof clips at the dev-fixed operating points
.\run.ps1 report       # rebuild artifacts/results.md from release/
.\run.ps1 gpu-pretrained   # Modal: AASIST, AASIST-L, SSL+AASIST on every condition

Source of every number on this page: REPORT.md (state on 2026-10-24) and README.md in the build's repository.