Speech Engine SPA — v9.1 Priority 1–5 Implementation

Restored v8.5 modules • Local STFT spectral subtraction • Entropy VAD • Local CTC beam search • pre-roll flush • biometric gate • ONNX model loader hooks
State: IDLE

Priority 2 / 3 / 5 — Optional Real ONNX Model Loader

ONNX Runtime: checking...

Load optional ONNX models from disk. If no model is loaded, the engine uses local DSP / fallback inference. For full 100-score production mode, load real models for VAD, acoustic CTC, speaker embedding, and p-MLM rescoring.

Silero VAD Model (.onnx)
StatusNot loaded
Acoustic CTC Model (.onnx)
StatusNot loaded
Speaker Embedding Model (.onnx)
StatusNot loaded
p-MLM Rescorer Model (.onnx)
StatusNot loaded

Priority 1 — Synthetic Multi-Element Stream Test Bench

Synthesizer Idle

Priority 1 — Granular System Health Diagnostics

Diagnostics: waiting...
1. Web Audio API
AudioContext class...
Instance creation...
16 kHz target...
Context state...
2. AudioWorklet
audioWorklet property...
Data URI encoder...
addModule()...
Processor instantiation...
3. Web Worker
Worker class...
Worker instantiation...
Messaging pipeline...
Ping / pong latency...
4. WASM / WebGPU
WebAssembly engine...
WASM SIMD 128...
WebGPU API...
GPU adapter...
5. Microphone / AGC
mediaDevices API...
getUserMedia()...
Mic preflight...
Active track...
Software AGC...
6. Biometric Engine
Vector storage...
Cosine engine...
Owner profile...
Gate calibration...

Priority 5 — Biometric Voice-ID Gate

Biometric: unenrolled
Owner Enrollment
No owner profile
Live Similarity
--
Embedding Drift
--
Gate Decision
UNENROLLED / ALLOW

Priority 1 — Progressive Adaptation & Milestones

Engine Idle
Acoustic convergence 0%
Precision gain: +0.0% Session: 00:00
Processed frames
0
Noise variance
Uncalibrated
RIR / spectral lock
0%
Biometric drift
--

Priority 1 — Positive Ambient Audio Indicator

POSITIVE AMBIENT AUDIO
Acoustic state
Optimal Speech Environment
SNR
0.0 dB
Noise floor
0.0 dBFS
DRR score
0%

Master Speech Processing Performance Index (SPI)

Local fallback mode
MASTER SPI
0.0
Accuracy (40%)
0.0 / 100
Acoustics (30%)
0.0 / 100
Reliability (15%)
0.0 / 100
Speed (15%)
0.0 / 100

Real-Time Detection Category Matrix

Decoder: fallback estimator
1. VAD / Voice Activity
0%
2. Phoneme Accuracy
0%
3. Syllable Nucleus
0%
4. Streaming Partials
0%
5. Full Word Certainty
0%

Priority 1 — Undecided & Partial Detection Table

0 partials
Time Hypothesis State Confidence Rescoring Resolution
Waiting for partial speech hypotheses...

System Event Log

0 events

Local CTC / p-MLM Speculative Feed

Spectral Magnitude

Implementation note: This build includes the full Priority 1–5 architecture. Real ONNX model execution depends on user-supplied model files because model weights cannot be embedded portably in a single HTML file. Without ONNX models, the app remains fully functional using local DSP, entropy VAD, spectral subtraction, and a local CTC beam-search fallback estimator.