@elizaos/plugin-local-inference 2.0.0-beta.1 → 2.0.3-beta.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +21 -0
- package/README.md +157 -0
- package/dist/actions/generate-media.d.ts +59 -0
- package/dist/actions/generate-media.d.ts.map +1 -0
- package/dist/actions/identify-speaker.d.ts +23 -0
- package/dist/actions/identify-speaker.d.ts.map +1 -0
- package/dist/actions/transcription-control.d.ts +29 -0
- package/dist/actions/transcription-control.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/environment.d.ts +12 -0
- package/dist/adapters/capacitor-llama/environment.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/index.browser.d.ts +9 -0
- package/dist/adapters/capacitor-llama/index.browser.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/index.d.ts +18 -0
- package/dist/adapters/capacitor-llama/index.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/loader.d.ts +35 -0
- package/dist/adapters/capacitor-llama/loader.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/native-voice-capture.d.ts +70 -0
- package/dist/adapters/capacitor-llama/native-voice-capture.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/structured-output.d.ts +62 -0
- package/dist/adapters/capacitor-llama/structured-output.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/text-streaming.d.ts +24 -0
- package/dist/adapters/capacitor-llama/text-streaming.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/types.d.ts +338 -0
- package/dist/adapters/capacitor-llama/types.d.ts.map +1 -0
- package/dist/adapters/capacitor-llama/voice-turn.d.ts +86 -0
- package/dist/adapters/capacitor-llama/voice-turn.d.ts.map +1 -0
- package/dist/backends/apple-foundation.d.ts +56 -0
- package/dist/backends/apple-foundation.d.ts.map +1 -0
- package/dist/index.d.ts +8 -37
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +38979 -430
- package/dist/index.js.map +217 -0
- package/dist/local-inference-routes.d.ts +47 -0
- package/dist/local-inference-routes.d.ts.map +1 -0
- package/dist/provider.d.ts +21 -0
- package/dist/provider.d.ts.map +1 -0
- package/dist/routes/compat-helpers.d.ts +18 -0
- package/dist/routes/compat-helpers.d.ts.map +1 -0
- package/dist/routes/family-member-route.d.ts +62 -0
- package/dist/routes/family-member-route.d.ts.map +1 -0
- package/dist/routes/index.d.ts +20 -0
- package/dist/routes/index.d.ts.map +1 -0
- package/dist/routes/index.js +42040 -0
- package/dist/routes/index.js.map +236 -0
- package/dist/routes/live-diarization-route.d.ts +33 -0
- package/dist/routes/live-diarization-route.d.ts.map +1 -0
- package/dist/routes/local-inference-asr-route.d.ts +4 -0
- package/dist/routes/local-inference-asr-route.d.ts.map +1 -0
- package/dist/routes/local-inference-asr-transcribe.d.ts +20 -0
- package/dist/routes/local-inference-asr-transcribe.d.ts.map +1 -0
- package/dist/routes/local-inference-compat-routes.d.ts +16 -0
- package/dist/routes/local-inference-compat-routes.d.ts.map +1 -0
- package/dist/routes/local-inference-tts-route.d.ts +7 -0
- package/dist/routes/local-inference-tts-route.d.ts.map +1 -0
- package/dist/routes/native-pcm-turn-route.d.ts +3 -0
- package/dist/routes/native-pcm-turn-route.d.ts.map +1 -0
- package/dist/routes/transcript-audio-store.d.ts +15 -0
- package/dist/routes/transcript-audio-store.d.ts.map +1 -0
- package/dist/routes/transcripts-routes.d.ts +44 -0
- package/dist/routes/transcripts-routes.d.ts.map +1 -0
- package/dist/routes/voice-first-run-routes.d.ts +62 -0
- package/dist/routes/voice-first-run-routes.d.ts.map +1 -0
- package/dist/routes/voice-models-routes.d.ts +62 -0
- package/dist/routes/voice-models-routes.d.ts.map +1 -0
- package/dist/routes/voice-profile-plugin-routes.d.ts +19 -0
- package/dist/routes/voice-profile-plugin-routes.d.ts.map +1 -0
- package/dist/routes/voice-profiles-management-routes.d.ts +52 -0
- package/dist/routes/voice-profiles-management-routes.d.ts.map +1 -0
- package/dist/routes/voice-speaker-profile-routes.d.ts +57 -0
- package/dist/routes/voice-speaker-profile-routes.d.ts.map +1 -0
- package/dist/runtime/embedding-manager-support.d.ts +77 -0
- package/dist/runtime/embedding-manager-support.d.ts.map +1 -0
- package/dist/runtime/embedding-presets.d.ts +16 -0
- package/dist/runtime/embedding-presets.d.ts.map +1 -0
- package/dist/runtime/embedding-warmup-policy.d.ts +14 -0
- package/dist/runtime/embedding-warmup-policy.d.ts.map +1 -0
- package/dist/runtime/ensure-local-inference-handler.d.ts +70 -0
- package/dist/runtime/ensure-local-inference-handler.d.ts.map +1 -0
- package/dist/runtime/index.d.ts +15 -0
- package/dist/runtime/index.d.ts.map +1 -0
- package/dist/runtime/index.js +38768 -0
- package/dist/runtime/index.js.map +217 -0
- package/dist/runtime/mobile-local-inference-gate.d.ts +63 -0
- package/dist/runtime/mobile-local-inference-gate.d.ts.map +1 -0
- package/dist/runtime/voice-entity-binding.d.ts +113 -0
- package/dist/runtime/voice-entity-binding.d.ts.map +1 -0
- package/dist/services/active-model.d.ts +310 -0
- package/dist/services/active-model.d.ts.map +1 -0
- package/dist/services/asr-provenance.d.ts +5 -0
- package/dist/services/asr-provenance.d.ts.map +1 -0
- package/dist/services/assignments.d.ts +84 -0
- package/dist/services/assignments.d.ts.map +1 -0
- package/dist/services/backend-selector.d.ts +55 -0
- package/dist/services/backend-selector.d.ts.map +1 -0
- package/dist/services/backend.d.ts +440 -0
- package/dist/services/backend.d.ts.map +1 -0
- package/dist/services/bionic-host-loader.d.ts +67 -0
- package/dist/services/bionic-host-loader.d.ts.map +1 -0
- package/dist/services/bundled-models.d.ts +34 -0
- package/dist/services/bundled-models.d.ts.map +1 -0
- package/dist/services/cache-bridge.d.ts +206 -0
- package/dist/services/cache-bridge.d.ts.map +1 -0
- package/dist/services/catalog.d.ts +10 -0
- package/dist/services/catalog.d.ts.map +1 -0
- package/dist/services/checkpoint-client.d.ts +109 -0
- package/dist/services/checkpoint-client.d.ts.map +1 -0
- package/dist/services/checkpoint-manager.d.ts +217 -0
- package/dist/services/checkpoint-manager.d.ts.map +1 -0
- package/dist/services/cloud-fallback.d.ts +102 -0
- package/dist/services/cloud-fallback.d.ts.map +1 -0
- package/dist/services/context-fit.d.ts +36 -0
- package/dist/services/context-fit.d.ts.map +1 -0
- package/dist/services/conversation-registry.d.ts +142 -0
- package/dist/services/conversation-registry.d.ts.map +1 -0
- package/dist/services/desktop-fused-ffi-backend-runtime.d.ts +111 -0
- package/dist/services/desktop-fused-ffi-backend-runtime.d.ts.map +1 -0
- package/dist/services/device-bridge.d.ts +188 -0
- package/dist/services/device-bridge.d.ts.map +1 -0
- package/dist/services/device-resource-metrics.d.ts +149 -0
- package/dist/services/device-resource-metrics.d.ts.map +1 -0
- package/dist/services/device-tier.d.ts +133 -0
- package/dist/services/device-tier.d.ts.map +1 -0
- package/dist/services/downloader.d.ts +94 -0
- package/dist/services/downloader.d.ts.map +1 -0
- package/dist/services/engine.d.ts +579 -0
- package/dist/services/engine.d.ts.map +1 -0
- package/dist/services/ensure-local-artifacts.d.ts +82 -0
- package/dist/services/ensure-local-artifacts.d.ts.map +1 -0
- package/dist/services/external-scanner.d.ts +17 -0
- package/dist/services/external-scanner.d.ts.map +1 -0
- package/dist/services/ffi-llm-mock.d.ts +90 -0
- package/dist/services/ffi-llm-mock.d.ts.map +1 -0
- package/dist/services/ffi-llm-streaming-abi.d.ts +318 -0
- package/dist/services/ffi-llm-streaming-abi.d.ts.map +1 -0
- package/dist/services/ffi-streaming-backend.d.ts +201 -0
- package/dist/services/ffi-streaming-backend.d.ts.map +1 -0
- package/dist/services/ffi-streaming-runner.d.ts +146 -0
- package/dist/services/ffi-streaming-runner.d.ts.map +1 -0
- package/dist/services/gpu-autotune.d.ts +150 -0
- package/dist/services/gpu-autotune.d.ts.map +1 -0
- package/dist/services/gpu-detect.d.ts +56 -0
- package/dist/services/gpu-detect.d.ts.map +1 -0
- package/dist/services/handler-registry.d.ts +72 -0
- package/dist/services/handler-registry.d.ts.map +1 -0
- package/dist/services/hardware.d.ts +63 -0
- package/dist/services/hardware.d.ts.map +1 -0
- package/dist/services/image-description-runtime.d.ts +14 -0
- package/dist/services/image-description-runtime.d.ts.map +1 -0
- package/dist/services/imagegen/aosp-unavailable.d.ts +134 -0
- package/dist/services/imagegen/aosp-unavailable.d.ts.map +1 -0
- package/dist/services/imagegen/backend-selector.d.ts +118 -0
- package/dist/services/imagegen/backend-selector.d.ts.map +1 -0
- package/dist/services/imagegen/coreml-unavailable.d.ts +105 -0
- package/dist/services/imagegen/coreml-unavailable.d.ts.map +1 -0
- package/dist/services/imagegen/errors.d.ts +16 -0
- package/dist/services/imagegen/errors.d.ts.map +1 -0
- package/dist/services/imagegen/index.d.ts +58 -0
- package/dist/services/imagegen/index.d.ts.map +1 -0
- package/dist/services/imagegen/mflux.d.ts +74 -0
- package/dist/services/imagegen/mflux.d.ts.map +1 -0
- package/dist/services/imagegen/sd-cpp.d.ts +181 -0
- package/dist/services/imagegen/sd-cpp.d.ts.map +1 -0
- package/dist/services/imagegen/tensorrt-unavailable.d.ts +83 -0
- package/dist/services/imagegen/tensorrt-unavailable.d.ts.map +1 -0
- package/dist/services/imagegen/types.d.ts +181 -0
- package/dist/services/imagegen/types.d.ts.map +1 -0
- package/dist/services/index.d.ts +31 -0
- package/dist/services/index.d.ts.map +1 -0
- package/dist/services/index.js +39453 -0
- package/dist/services/index.js.map +227 -0
- package/dist/services/inference-capabilities.d.ts +132 -0
- package/dist/services/inference-capabilities.d.ts.map +1 -0
- package/dist/services/inference-telemetry.d.ts +59 -0
- package/dist/services/inference-telemetry.d.ts.map +1 -0
- package/dist/services/ios-llama-streaming.d.ts +119 -0
- package/dist/services/ios-llama-streaming.d.ts.map +1 -0
- package/dist/services/kv-spill.d.ts +189 -0
- package/dist/services/kv-spill.d.ts.map +1 -0
- package/dist/services/latency-trace.d.ts +346 -0
- package/dist/services/latency-trace.d.ts.map +1 -0
- package/dist/services/lib-target.d.ts +55 -0
- package/dist/services/lib-target.d.ts.map +1 -0
- package/dist/services/live-signals.d.ts +86 -0
- package/dist/services/live-signals.d.ts.map +1 -0
- package/dist/services/llama-server-metrics.d.ts +114 -0
- package/dist/services/llama-server-metrics.d.ts.map +1 -0
- package/dist/services/llm-streaming-binding.d.ts +96 -0
- package/dist/services/llm-streaming-binding.d.ts.map +1 -0
- package/dist/services/load-args.d.ts +82 -0
- package/dist/services/load-args.d.ts.map +1 -0
- package/dist/services/manifest/index.d.ts +4 -0
- package/dist/services/manifest/index.d.ts.map +1 -0
- package/dist/services/manifest/schema.d.ts +903 -0
- package/dist/services/manifest/schema.d.ts.map +1 -0
- package/dist/services/manifest/types.d.ts +32 -0
- package/dist/services/manifest/types.d.ts.map +1 -0
- package/dist/services/manifest/validator.d.ts +66 -0
- package/dist/services/manifest/validator.d.ts.map +1 -0
- package/dist/services/memory-arbiter.d.ts +348 -0
- package/dist/services/memory-arbiter.d.ts.map +1 -0
- package/dist/services/memory-benchmark.d.ts +76 -0
- package/dist/services/memory-benchmark.d.ts.map +1 -0
- package/dist/services/memory-monitor.d.ts +128 -0
- package/dist/services/memory-monitor.d.ts.map +1 -0
- package/dist/services/memory-pressure.d.ts +130 -0
- package/dist/services/memory-pressure.d.ts.map +1 -0
- package/dist/services/mtp-doctor.d.ts +13 -0
- package/dist/services/mtp-doctor.d.ts.map +1 -0
- package/dist/services/network-policy.d.ts +127 -0
- package/dist/services/network-policy.d.ts.map +1 -0
- package/dist/services/paths.d.ts +6 -0
- package/dist/services/paths.d.ts.map +1 -0
- package/dist/services/planner-skeleton.d.ts +124 -0
- package/dist/services/planner-skeleton.d.ts.map +1 -0
- package/dist/services/providers.d.ts +38 -0
- package/dist/services/providers.d.ts.map +1 -0
- package/dist/services/ram-budget.d.ts +110 -0
- package/dist/services/ram-budget.d.ts.map +1 -0
- package/dist/services/readiness.d.ts +9 -0
- package/dist/services/readiness.d.ts.map +1 -0
- package/dist/services/recommendation.d.ts +111 -0
- package/dist/services/recommendation.d.ts.map +1 -0
- package/dist/services/registry.d.ts +33 -0
- package/dist/services/registry.d.ts.map +1 -0
- package/dist/services/router-handler.d.ts +92 -0
- package/dist/services/router-handler.d.ts.map +1 -0
- package/dist/services/routing-policy.d.ts +92 -0
- package/dist/services/routing-policy.d.ts.map +1 -0
- package/dist/services/routing-preferences.d.ts +8 -0
- package/dist/services/routing-preferences.d.ts.map +1 -0
- package/dist/services/runtime-target.d.ts +98 -0
- package/dist/services/runtime-target.d.ts.map +1 -0
- package/dist/services/service.d.ts +128 -0
- package/dist/services/service.d.ts.map +1 -0
- package/dist/services/session-pool.d.ts +72 -0
- package/dist/services/session-pool.d.ts.map +1 -0
- package/dist/services/structured-output/deterministic-repair.d.ts +23 -0
- package/dist/services/structured-output/deterministic-repair.d.ts.map +1 -0
- package/dist/services/structured-output/index.d.ts +2 -0
- package/dist/services/structured-output/index.d.ts.map +1 -0
- package/dist/services/structured-output.d.ts +311 -0
- package/dist/services/structured-output.d.ts.map +1 -0
- package/dist/services/system-memory.d.ts +33 -0
- package/dist/services/system-memory.d.ts.map +1 -0
- package/dist/services/types.d.ts +19 -0
- package/dist/services/types.d.ts.map +1 -0
- package/dist/services/verify-on-device.d.ts +34 -0
- package/dist/services/verify-on-device.d.ts.map +1 -0
- package/dist/services/verify.d.ts +8 -0
- package/dist/services/verify.d.ts.map +1 -0
- package/dist/services/vision/aosp-unavailable.d.ts +115 -0
- package/dist/services/vision/aosp-unavailable.d.ts.map +1 -0
- package/dist/services/vision/capacitor-llama.d.ts +99 -0
- package/dist/services/vision/capacitor-llama.d.ts.map +1 -0
- package/dist/services/vision/cloud-fallback.d.ts +47 -0
- package/dist/services/vision/cloud-fallback.d.ts.map +1 -0
- package/dist/services/vision/hash.d.ts +71 -0
- package/dist/services/vision/hash.d.ts.map +1 -0
- package/dist/services/vision/index.d.ts +95 -0
- package/dist/services/vision/index.d.ts.map +1 -0
- package/dist/services/vision/llama-server.d.ts +73 -0
- package/dist/services/vision/llama-server.d.ts.map +1 -0
- package/dist/services/vision/types.d.ts +162 -0
- package/dist/services/vision/types.d.ts.map +1 -0
- package/dist/services/vision/vast-fallback.d.ts +18 -0
- package/dist/services/vision/vast-fallback.d.ts.map +1 -0
- package/dist/services/vision-embedding-cache.d.ts +98 -0
- package/dist/services/vision-embedding-cache.d.ts.map +1 -0
- package/dist/services/voice/__test-helpers__/fake-ffi.d.ts +27 -0
- package/dist/services/voice/__test-helpers__/fake-ffi.d.ts.map +1 -0
- package/dist/services/voice/__test-helpers__/synthetic-speech.d.ts +66 -0
- package/dist/services/voice/__test-helpers__/synthetic-speech.d.ts.map +1 -0
- package/dist/services/voice/acoustic-speaker-attribution.d.ts +61 -0
- package/dist/services/voice/acoustic-speaker-attribution.d.ts.map +1 -0
- package/dist/services/voice/audio-frame-consumer.d.ts +294 -0
- package/dist/services/voice/audio-frame-consumer.d.ts.map +1 -0
- package/dist/services/voice/barge-in.d.ts +112 -0
- package/dist/services/voice/barge-in.d.ts.map +1 -0
- package/dist/services/voice/cancellation-coordinator.d.ts +127 -0
- package/dist/services/voice/cancellation-coordinator.d.ts.map +1 -0
- package/dist/services/voice/checkpoint-manager.d.ts +199 -0
- package/dist/services/voice/checkpoint-manager.d.ts.map +1 -0
- package/dist/services/voice/checkpoint-policy.d.ts +178 -0
- package/dist/services/voice/checkpoint-policy.d.ts.map +1 -0
- package/dist/services/voice/corpus-augment.d.ts +111 -0
- package/dist/services/voice/corpus-augment.d.ts.map +1 -0
- package/dist/services/voice/corpus-generator.d.ts +134 -0
- package/dist/services/voice/corpus-generator.d.ts.map +1 -0
- package/dist/services/voice/diarization-error-rate.d.ts +40 -0
- package/dist/services/voice/diarization-error-rate.d.ts.map +1 -0
- package/dist/services/voice/e2e-harness.d.ts +297 -0
- package/dist/services/voice/e2e-harness.d.ts.map +1 -0
- package/dist/services/voice/eager-context-builder.d.ts +170 -0
- package/dist/services/voice/eager-context-builder.d.ts.map +1 -0
- package/dist/services/voice/echo-delay.d.ts +67 -0
- package/dist/services/voice/echo-delay.d.ts.map +1 -0
- package/dist/services/voice/echo-metrics.d.ts +7 -0
- package/dist/services/voice/echo-metrics.d.ts.map +1 -0
- package/dist/services/voice/echo-reference-buffer.d.ts +65 -0
- package/dist/services/voice/echo-reference-buffer.d.ts.map +1 -0
- package/dist/services/voice/eliza1-eot-scorer.d.ts +124 -0
- package/dist/services/voice/eliza1-eot-scorer.d.ts.map +1 -0
- package/dist/services/voice/embedding-server.d.ts +37 -0
- package/dist/services/voice/embedding-server.d.ts.map +1 -0
- package/dist/services/voice/embedding.d.ts +132 -0
- package/dist/services/voice/embedding.d.ts.map +1 -0
- package/dist/services/voice/emotion-attribution.d.ts +68 -0
- package/dist/services/voice/emotion-attribution.d.ts.map +1 -0
- package/dist/services/voice/engine-bridge.d.ts +762 -0
- package/dist/services/voice/engine-bridge.d.ts.map +1 -0
- package/dist/services/voice/eot-classifier-ggml.d.ts +179 -0
- package/dist/services/voice/eot-classifier-ggml.d.ts.map +1 -0
- package/dist/services/voice/eot-classifier.d.ts +211 -0
- package/dist/services/voice/eot-classifier.d.ts.map +1 -0
- package/dist/services/voice/errors.d.ts +20 -0
- package/dist/services/voice/errors.d.ts.map +1 -0
- package/dist/services/voice/expressive-tags.d.ts +158 -0
- package/dist/services/voice/expressive-tags.d.ts.map +1 -0
- package/dist/services/voice/ffi-bindings.d.ts +696 -0
- package/dist/services/voice/ffi-bindings.d.ts.map +1 -0
- package/dist/services/voice/first-line-cache.d.ts +181 -0
- package/dist/services/voice/first-line-cache.d.ts.map +1 -0
- package/dist/services/voice/fused-eot-scorer.d.ts +51 -0
- package/dist/services/voice/fused-eot-scorer.d.ts.map +1 -0
- package/dist/services/voice/index.d.ts +96 -0
- package/dist/services/voice/index.d.ts.map +1 -0
- package/dist/services/voice/kokoro/index.d.ts +24 -0
- package/dist/services/voice/kokoro/index.d.ts.map +1 -0
- package/dist/services/voice/kokoro/kokoro-backend.d.ts +87 -0
- package/dist/services/voice/kokoro/kokoro-backend.d.ts.map +1 -0
- package/dist/services/voice/kokoro/kokoro-engine-discovery.d.ts +58 -0
- package/dist/services/voice/kokoro/kokoro-engine-discovery.d.ts.map +1 -0
- package/dist/services/voice/kokoro/kokoro-ffi-runtime.d.ts +75 -0
- package/dist/services/voice/kokoro/kokoro-ffi-runtime.d.ts.map +1 -0
- package/dist/services/voice/kokoro/kokoro-runtime.d.ts +100 -0
- package/dist/services/voice/kokoro/kokoro-runtime.d.ts.map +1 -0
- package/dist/services/voice/kokoro/phoneme-stream.d.ts +51 -0
- package/dist/services/voice/kokoro/phoneme-stream.d.ts.map +1 -0
- package/dist/services/voice/kokoro/phonemizer.d.ts +50 -0
- package/dist/services/voice/kokoro/phonemizer.d.ts.map +1 -0
- package/dist/services/voice/kokoro/pick-runtime.d.ts +61 -0
- package/dist/services/voice/kokoro/pick-runtime.d.ts.map +1 -0
- package/dist/services/voice/kokoro/runtime-selection.d.ts +31 -0
- package/dist/services/voice/kokoro/runtime-selection.d.ts.map +1 -0
- package/dist/services/voice/kokoro/types.d.ts +82 -0
- package/dist/services/voice/kokoro/types.d.ts.map +1 -0
- package/dist/services/voice/kokoro/voice-presets.d.ts +23 -0
- package/dist/services/voice/kokoro/voice-presets.d.ts.map +1 -0
- package/dist/services/voice/kokoro/voices.d.ts +30 -0
- package/dist/services/voice/kokoro/voices.d.ts.map +1 -0
- package/dist/services/voice/lifecycle.d.ts +135 -0
- package/dist/services/voice/lifecycle.d.ts.map +1 -0
- package/dist/services/voice/live-diarization-session.d.ts +196 -0
- package/dist/services/voice/live-diarization-session.d.ts.map +1 -0
- package/dist/services/voice/metric-math.d.ts +10 -0
- package/dist/services/voice/metric-math.d.ts.map +1 -0
- package/dist/services/voice/mic-source.d.ts +136 -0
- package/dist/services/voice/mic-source.d.ts.map +1 -0
- package/dist/services/voice/nlms-echo-canceller.d.ts +137 -0
- package/dist/services/voice/nlms-echo-canceller.d.ts.map +1 -0
- package/dist/services/voice/optimistic-policy.d.ts +109 -0
- package/dist/services/voice/optimistic-policy.d.ts.map +1 -0
- package/dist/services/voice/optimistic-rollback.d.ts +151 -0
- package/dist/services/voice/optimistic-rollback.d.ts.map +1 -0
- package/dist/services/voice/partial-stabilizer.d.ts +73 -0
- package/dist/services/voice/partial-stabilizer.d.ts.map +1 -0
- package/dist/services/voice/phoneme-tokenizer.d.ts +49 -0
- package/dist/services/voice/phoneme-tokenizer.d.ts.map +1 -0
- package/dist/services/voice/phrase-cache.d.ts +76 -0
- package/dist/services/voice/phrase-cache.d.ts.map +1 -0
- package/dist/services/voice/phrase-chunker.d.ts +62 -0
- package/dist/services/voice/phrase-chunker.d.ts.map +1 -0
- package/dist/services/voice/pipeline-impls.d.ts +151 -0
- package/dist/services/voice/pipeline-impls.d.ts.map +1 -0
- package/dist/services/voice/pipeline.d.ts +216 -0
- package/dist/services/voice/pipeline.d.ts.map +1 -0
- package/dist/services/voice/prefill-client.d.ts +123 -0
- package/dist/services/voice/prefill-client.d.ts.map +1 -0
- package/dist/services/voice/prefix-preserving-queue.d.ts +113 -0
- package/dist/services/voice/prefix-preserving-queue.d.ts.map +1 -0
- package/dist/services/voice/profile-store.d.ts +248 -0
- package/dist/services/voice/profile-store.d.ts.map +1 -0
- package/dist/services/voice/ring-buffer.d.ts +40 -0
- package/dist/services/voice/ring-buffer.d.ts.map +1 -0
- package/dist/services/voice/rollback-queue.d.ts +24 -0
- package/dist/services/voice/rollback-queue.d.ts.map +1 -0
- package/dist/services/voice/samantha-preset-placeholder.d.ts +67 -0
- package/dist/services/voice/samantha-preset-placeholder.d.ts.map +1 -0
- package/dist/services/voice/samantha-preset-regenerator.d.ts +87 -0
- package/dist/services/voice/samantha-preset-regenerator.d.ts.map +1 -0
- package/dist/services/voice/scheduler.d.ts +146 -0
- package/dist/services/voice/scheduler.d.ts.map +1 -0
- package/dist/services/voice/self-voice-imprint.d.ts +33 -0
- package/dist/services/voice/self-voice-imprint.d.ts.map +1 -0
- package/dist/services/voice/shared-resources.d.ts +204 -0
- package/dist/services/voice/shared-resources.d.ts.map +1 -0
- package/dist/services/voice/speaker/attribution-pipeline.d.ts +74 -0
- package/dist/services/voice/speaker/attribution-pipeline.d.ts.map +1 -0
- package/dist/services/voice/speaker/diarizer-fused.d.ts +59 -0
- package/dist/services/voice/speaker/diarizer-fused.d.ts.map +1 -0
- package/dist/services/voice/speaker/diarizer.d.ts +75 -0
- package/dist/services/voice/speaker/diarizer.d.ts.map +1 -0
- package/dist/services/voice/speaker/encoder-fused.d.ts +60 -0
- package/dist/services/voice/speaker/encoder-fused.d.ts.map +1 -0
- package/dist/services/voice/speaker/encoder-ggml.d.ts +33 -0
- package/dist/services/voice/speaker/encoder-ggml.d.ts.map +1 -0
- package/dist/services/voice/speaker/encoder.d.ts +37 -0
- package/dist/services/voice/speaker/encoder.d.ts.map +1 -0
- package/dist/services/voice/speaker-imprint.d.ts +83 -0
- package/dist/services/voice/speaker-imprint.d.ts.map +1 -0
- package/dist/services/voice/speaker-preset-cache.d.ts +77 -0
- package/dist/services/voice/speaker-preset-cache.d.ts.map +1 -0
- package/dist/services/voice/streaming-asr/streaming-pipeline-adapter.d.ts +160 -0
- package/dist/services/voice/streaming-asr/streaming-pipeline-adapter.d.ts.map +1 -0
- package/dist/services/voice/system-audio-sink.d.ts +73 -0
- package/dist/services/voice/system-audio-sink.d.ts.map +1 -0
- package/dist/services/voice/transcriber.d.ts +244 -0
- package/dist/services/voice/transcriber.d.ts.map +1 -0
- package/dist/services/voice/transcript-knowledge.d.ts +37 -0
- package/dist/services/voice/transcript-knowledge.d.ts.map +1 -0
- package/dist/services/voice/transcript-service.d.ts +60 -0
- package/dist/services/voice/transcript-service.d.ts.map +1 -0
- package/dist/services/voice/transcript-store.d.ts +64 -0
- package/dist/services/voice/transcript-store.d.ts.map +1 -0
- package/dist/services/voice/turn-controller.d.ts +183 -0
- package/dist/services/voice/turn-controller.d.ts.map +1 -0
- package/dist/services/voice/types.d.ts +643 -0
- package/dist/services/voice/types.d.ts.map +1 -0
- package/dist/services/voice/vad.d.ts +283 -0
- package/dist/services/voice/vad.d.ts.map +1 -0
- package/dist/services/voice/voice-budget.d.ts +241 -0
- package/dist/services/voice/voice-budget.d.ts.map +1 -0
- package/dist/services/voice/voice-emotion-classifier.d.ts +95 -0
- package/dist/services/voice/voice-emotion-classifier.d.ts.map +1 -0
- package/dist/services/voice/voice-preload-predictor.d.ts +76 -0
- package/dist/services/voice/voice-preload-predictor.d.ts.map +1 -0
- package/dist/services/voice/voice-preset-format.d.ts +158 -0
- package/dist/services/voice/voice-preset-format.d.ts.map +1 -0
- package/dist/services/voice/voice-profile-artifact.d.ts +116 -0
- package/dist/services/voice/voice-profile-artifact.d.ts.map +1 -0
- package/dist/services/voice/voice-profile-routes.d.ts +83 -0
- package/dist/services/voice/voice-profile-routes.d.ts.map +1 -0
- package/dist/services/voice/voice-scenario.d.ts +131 -0
- package/dist/services/voice/voice-scenario.d.ts.map +1 -0
- package/dist/services/voice/voice-state-machine.d.ts +364 -0
- package/dist/services/voice/voice-state-machine.d.ts.map +1 -0
- package/dist/services/voice/voice-workbench-report.d.ts +117 -0
- package/dist/services/voice/voice-workbench-report.d.ts.map +1 -0
- package/dist/services/voice/wake-word-ggml.d.ts +100 -0
- package/dist/services/voice/wake-word-ggml.d.ts.map +1 -0
- package/dist/services/voice/wake-word.d.ts +255 -0
- package/dist/services/voice/wake-word.d.ts.map +1 -0
- package/dist/services/voice/wav-codec.d.ts +11 -0
- package/dist/services/voice/wav-codec.d.ts.map +1 -0
- package/dist/services/voice/workbench-entrypoint.d.ts +42 -0
- package/dist/services/voice/workbench-entrypoint.d.ts.map +1 -0
- package/dist/services/voice/workbench-headless-runner.d.ts +102 -0
- package/dist/services/voice/workbench-headless-runner.d.ts.map +1 -0
- package/dist/services/voice/workbench-logic-services.d.ts +36 -0
- package/dist/services/voice/workbench-logic-services.d.ts.map +1 -0
- package/dist/services/voice/workbench-real-services.d.ts +17 -0
- package/dist/services/voice/workbench-real-services.d.ts.map +1 -0
- package/dist/services/voice/workbench-scenarios.d.ts +24 -0
- package/dist/services/voice/workbench-scenarios.d.ts.map +1 -0
- package/dist/services/voice/wrap-with-first-line-cache.d.ts +70 -0
- package/dist/services/voice/wrap-with-first-line-cache.d.ts.map +1 -0
- package/dist/services/voice-model-updater.d.ts +240 -0
- package/dist/services/voice-model-updater.d.ts.map +1 -0
- package/dist/services/voice-prewarm.d.ts +3 -0
- package/dist/services/voice-prewarm.d.ts.map +1 -0
- package/dist/voice-workbench.d.ts +18 -0
- package/dist/voice-workbench.d.ts.map +1 -0
- package/dist/voice-workbench.js +5259 -0
- package/dist/voice-workbench.js.map +34 -0
- package/package.json +101 -15
- package/registry-entry.json +137 -0
- package/src/actions/generate-media.ts +647 -0
- package/src/actions/identify-speaker.ts +171 -0
- package/src/actions/transcription-control.test.ts +100 -0
- package/src/actions/transcription-control.ts +127 -0
- package/src/adapters/capacitor-llama/__tests__/compat-behavior.test.ts +218 -0
- package/src/adapters/capacitor-llama/__tests__/index.test.ts +68 -0
- package/src/adapters/capacitor-llama/__tests__/structured-output.test.ts +215 -0
- package/src/adapters/capacitor-llama/__tests__/text-streaming.test.ts +174 -0
- package/src/adapters/capacitor-llama/__tests__/voice-turn.test.ts +293 -0
- package/src/adapters/capacitor-llama/environment.ts +71 -0
- package/src/adapters/capacitor-llama/index.browser.ts +83 -0
- package/src/adapters/capacitor-llama/index.ts +831 -0
- package/src/adapters/capacitor-llama/loader.ts +109 -0
- package/src/adapters/capacitor-llama/native-voice-capture.ts +140 -0
- package/src/adapters/capacitor-llama/structured-output.ts +165 -0
- package/src/adapters/capacitor-llama/text-streaming.ts +227 -0
- package/src/adapters/capacitor-llama/types.ts +374 -0
- package/src/adapters/capacitor-llama/voice-turn.ts +178 -0
- package/src/backends/apple-foundation.ts +127 -0
- package/src/index.ts +62 -0
- package/src/local-inference-routes.test.ts +390 -0
- package/src/local-inference-routes.ts +1625 -0
- package/src/provider.ts +1111 -0
- package/src/routes/compat-helpers.ts +275 -0
- package/src/routes/family-member-route.ts +353 -0
- package/src/routes/index.ts +61 -0
- package/src/routes/live-diarization-route.test.ts +347 -0
- package/src/routes/live-diarization-route.ts +198 -0
- package/src/routes/local-inference-asr-route.test.ts +246 -0
- package/src/routes/local-inference-asr-route.ts +166 -0
- package/src/routes/local-inference-asr-transcribe.test.ts +118 -0
- package/src/routes/local-inference-asr-transcribe.ts +97 -0
- package/src/routes/local-inference-compat-routes.test.ts +485 -0
- package/src/routes/local-inference-compat-routes.ts +775 -0
- package/src/routes/local-inference-tts-route.test.ts +179 -0
- package/src/routes/local-inference-tts-route.ts +230 -0
- package/src/routes/native-pcm-turn-route.test.ts +136 -0
- package/src/routes/native-pcm-turn-route.ts +121 -0
- package/src/routes/transcript-audio-store.ts +27 -0
- package/src/routes/transcripts-routes.test.ts +195 -0
- package/src/routes/transcripts-routes.ts +191 -0
- package/src/routes/voice-first-run-routes.ts +524 -0
- package/src/routes/voice-models-routes.ts +554 -0
- package/src/routes/voice-profile-plugin-routes.ts +138 -0
- package/src/routes/voice-profiles-management-routes.ts +476 -0
- package/src/routes/voice-speaker-profile-routes.ts +199 -0
- package/src/runtime/aosp-llama-loader-selection.test.ts +80 -0
- package/src/runtime/bionic-wire-encoding.test.ts +147 -0
- package/src/runtime/capacitor-llama.d.ts +25 -0
- package/src/runtime/embedding-manager-support.ts +497 -0
- package/src/runtime/embedding-presets.ts +81 -0
- package/src/runtime/embedding-warmup-policy.test.ts +53 -0
- package/src/runtime/embedding-warmup-policy.ts +48 -0
- package/src/runtime/ensure-local-inference-handler.test.ts +726 -0
- package/src/runtime/ensure-local-inference-handler.ts +1640 -0
- package/src/runtime/index.ts +36 -0
- package/src/runtime/mobile-local-inference-gate.test.ts +152 -0
- package/src/runtime/mobile-local-inference-gate.ts +99 -0
- package/src/runtime/voice-entity-binding.transcript.test.ts +98 -0
- package/src/runtime/voice-entity-binding.ts +368 -0
- package/src/runtime/voice-speaker-entity-contract.test.ts +149 -0
- package/src/services/README.md +71 -0
- package/src/services/__tests__/backend-selector.precedence.test.ts +333 -0
- package/src/services/__tests__/backend-selector.test.ts +101 -0
- package/src/services/__tests__/checkpoint-manager.test.ts +376 -0
- package/src/services/__tests__/gpu-autotune.test.ts +400 -0
- package/src/services/__tests__/llm-streaming-binding.test.ts +85 -0
- package/src/services/__tests__/planner-grammar.test.ts +372 -0
- package/src/services/__tests__/runtime-target.test.ts +176 -0
- package/src/services/active-model-context-fit.test.ts +125 -0
- package/src/services/active-model-switch-rollback.test.ts +183 -0
- package/src/services/active-model.ts +1416 -0
- package/src/services/asr-provenance.ts +68 -0
- package/src/services/assignment-validation.test.ts +118 -0
- package/src/services/assignments.test.ts +106 -0
- package/src/services/assignments.ts +278 -0
- package/src/services/backend-selector.ts +95 -0
- package/src/services/backend.test.ts +84 -0
- package/src/services/backend.ts +791 -0
- package/src/services/bionic-host-loader.test.ts +226 -0
- package/src/services/bionic-host-loader.ts +252 -0
- package/src/services/bundled-models.ts +129 -0
- package/src/services/cache-bridge.test.ts +516 -0
- package/src/services/cache-bridge.ts +423 -0
- package/src/services/catalog.test.ts +259 -0
- package/src/services/catalog.ts +33 -0
- package/src/services/checkpoint-client.ts +258 -0
- package/src/services/checkpoint-manager.ts +474 -0
- package/src/services/cloud-fallback.ts +230 -0
- package/src/services/context-fit.test.ts +121 -0
- package/src/services/context-fit.ts +113 -0
- package/src/services/conversation-registry.test.ts +235 -0
- package/src/services/conversation-registry.ts +264 -0
- package/src/services/desktop-fused-ffi-backend-runtime.ts +431 -0
- package/src/services/device-bridge.ts +1237 -0
- package/src/services/device-resource-metrics.test.ts +98 -0
- package/src/services/device-resource-metrics.ts +346 -0
- package/src/services/device-tier.test.ts +458 -0
- package/src/services/device-tier.ts +502 -0
- package/src/services/downloader.test.ts +888 -0
- package/src/services/downloader.ts +1039 -0
- package/src/services/engine-direct-bundle.test.ts +90 -0
- package/src/services/engine-streaming.test.ts +80 -0
- package/src/services/engine.ts +2096 -0
- package/src/services/ensure-local-artifacts.integration.test.ts +273 -0
- package/src/services/ensure-local-artifacts.test.ts +368 -0
- package/src/services/ensure-local-artifacts.ts +351 -0
- package/src/services/external-scanner.ts +312 -0
- package/src/services/ffi-llm-mock.ts +354 -0
- package/src/services/ffi-llm-streaming-abi.ts +445 -0
- package/src/services/ffi-streaming-backend.ts +418 -0
- package/src/services/ffi-streaming-runner.test.ts +220 -0
- package/src/services/ffi-streaming-runner.ts +407 -0
- package/src/services/ffi-unload-ordering.test.ts +166 -0
- package/src/services/fused-eliza1-no-regression.test.ts +144 -0
- package/src/services/gpu-autotune.ts +534 -0
- package/src/services/gpu-detect.ts +139 -0
- package/src/services/handler-registry.ts +240 -0
- package/src/services/hardware.test.ts +236 -0
- package/src/services/hardware.ts +438 -0
- package/src/services/image-description-runtime.test.ts +61 -0
- package/src/services/image-description-runtime.ts +118 -0
- package/src/services/imagegen/aosp-unavailable.ts +229 -0
- package/src/services/imagegen/backend-selector.test.ts +190 -0
- package/src/services/imagegen/backend-selector.ts +277 -0
- package/src/services/imagegen/coreml-unavailable.ts +237 -0
- package/src/services/imagegen/errors.ts +40 -0
- package/src/services/imagegen/index.ts +144 -0
- package/src/services/imagegen/mflux.ts +313 -0
- package/src/services/imagegen/sd-cpp.ts +715 -0
- package/src/services/imagegen/tensorrt-unavailable.ts +295 -0
- package/src/services/imagegen/types.ts +193 -0
- package/src/services/index.ts +229 -0
- package/src/services/inference-capabilities.test.ts +75 -0
- package/src/services/inference-capabilities.ts +204 -0
- package/src/services/inference-telemetry.ts +143 -0
- package/src/services/ios-llama-streaming.ts +248 -0
- package/src/services/kv-spill.test.ts +222 -0
- package/src/services/kv-spill.ts +357 -0
- package/src/services/latency-trace.test.ts +266 -0
- package/src/services/latency-trace.ts +844 -0
- package/src/services/lib-target.test.ts +145 -0
- package/src/services/lib-target.ts +102 -0
- package/src/services/live-signals.test.ts +132 -0
- package/src/services/live-signals.ts +177 -0
- package/src/services/llama-server-metrics.test.ts +168 -0
- package/src/services/llama-server-metrics.ts +304 -0
- package/src/services/llm-streaming-binding.ts +136 -0
- package/src/services/load-args.ts +81 -0
- package/src/services/manifest/eliza-1.manifest.v1.json +790 -0
- package/src/services/manifest/index.ts +72 -0
- package/src/services/manifest/manifest.test.ts +791 -0
- package/src/services/manifest/schema.ts +761 -0
- package/src/services/manifest/types.ts +61 -0
- package/src/services/manifest/validator.ts +633 -0
- package/src/services/memory-arbiter.test.ts +558 -0
- package/src/services/memory-arbiter.ts +991 -0
- package/src/services/memory-benchmark.test.ts +91 -0
- package/src/services/memory-benchmark.ts +354 -0
- package/src/services/memory-monitor.test.ts +232 -0
- package/src/services/memory-monitor.ts +309 -0
- package/src/services/memory-pressure.ts +414 -0
- package/src/services/mtp-doctor.ts +86 -0
- package/src/services/network-policy.ts +346 -0
- package/src/services/paths.ts +25 -0
- package/src/services/planner-skeleton.ts +175 -0
- package/src/services/providers.ts +507 -0
- package/src/services/ram-budget-cache.test.ts +164 -0
- package/src/services/ram-budget.ts +309 -0
- package/src/services/readiness.test.ts +87 -0
- package/src/services/readiness.ts +238 -0
- package/src/services/recommendation.test.ts +216 -0
- package/src/services/recommendation.ts +671 -0
- package/src/services/registry.ts +157 -0
- package/src/services/required-kernels-gate.test.ts +64 -0
- package/src/services/router-handler.test.ts +45 -0
- package/src/services/router-handler.ts +426 -0
- package/src/services/routing-policy.test.ts +352 -0
- package/src/services/routing-policy.ts +367 -0
- package/src/services/routing-preferences.ts +17 -0
- package/src/services/runtime-target.ts +154 -0
- package/src/services/service.test.ts +223 -0
- package/src/services/service.ts +750 -0
- package/src/services/session-pool.ts +153 -0
- package/src/services/structured-output/deterministic-repair.test.ts +169 -0
- package/src/services/structured-output/deterministic-repair.ts +443 -0
- package/src/services/structured-output/index.ts +4 -0
- package/src/services/structured-output.test.ts +483 -0
- package/src/services/structured-output.ts +712 -0
- package/src/services/system-memory.test.ts +47 -0
- package/src/services/system-memory.ts +67 -0
- package/src/services/transcription-priority.test.ts +211 -0
- package/src/services/types.ts +59 -0
- package/src/services/verify-on-device.test.ts +87 -0
- package/src/services/verify-on-device.ts +127 -0
- package/src/services/verify.ts +13 -0
- package/src/services/vision/aosp-unavailable.ts +163 -0
- package/src/services/vision/capacitor-llama.ts +255 -0
- package/src/services/vision/cloud-fallback.test.ts +243 -0
- package/src/services/vision/cloud-fallback.ts +268 -0
- package/src/services/vision/fallback-chain.test.ts +86 -0
- package/src/services/vision/hash.ts +157 -0
- package/src/services/vision/index.ts +251 -0
- package/src/services/vision/llama-server.ts +177 -0
- package/src/services/vision/types.ts +163 -0
- package/src/services/vision/vast-fallback.ts +127 -0
- package/src/services/vision-embedding-cache.ts +189 -0
- package/src/services/voice/VOICE_WORKBENCH.md +133 -0
- package/src/services/voice/__fixtures__/voice-workbench-logic-baseline.json +180 -0
- package/src/services/voice/__test-helpers__/fake-ffi.ts +94 -0
- package/src/services/voice/__test-helpers__/synthetic-speech.ts +194 -0
- package/src/services/voice/__tests__/checkpoint-manager.test.ts +241 -0
- package/src/services/voice/__tests__/checkpoint-policy.test.ts +270 -0
- package/src/services/voice/__tests__/eager-context-builder.test.ts +257 -0
- package/src/services/voice/__tests__/eliza1-eot-scorer.test.ts +288 -0
- package/src/services/voice/__tests__/eot-classifier.test.ts +431 -0
- package/src/services/voice/__tests__/optimistic-rollback.test.ts +312 -0
- package/src/services/voice/__tests__/prefill-client.test.ts +266 -0
- package/src/services/voice/__tests__/prefix-preserving-queue.test.ts +208 -0
- package/src/services/voice/__tests__/streaming-asr.test.ts +450 -0
- package/src/services/voice/__tests__/streaming-transcriber.test.ts +339 -0
- package/src/services/voice/__tests__/turn-detector-resolver.test.ts +195 -0
- package/src/services/voice/__tests__/voice-state-machine-prefill.test.ts +275 -0
- package/src/services/voice/__tests__/voice-state-machine.test.ts +354 -0
- package/src/services/voice/acoustic-speaker-attribution.test.ts +165 -0
- package/src/services/voice/acoustic-speaker-attribution.ts +336 -0
- package/src/services/voice/asr-timed.real.test.ts +139 -0
- package/src/services/voice/audio-frame-consumer.test.ts +669 -0
- package/src/services/voice/audio-frame-consumer.ts +651 -0
- package/src/services/voice/barge-in.test.ts +244 -0
- package/src/services/voice/barge-in.ts +335 -0
- package/src/services/voice/cancellation-coordinator.test.ts +196 -0
- package/src/services/voice/cancellation-coordinator.ts +269 -0
- package/src/services/voice/checkpoint-manager.ts +401 -0
- package/src/services/voice/checkpoint-policy.ts +336 -0
- package/src/services/voice/composite-eot-classifier.test.ts +59 -0
- package/src/services/voice/corpus-augment.test.ts +276 -0
- package/src/services/voice/corpus-augment.ts +451 -0
- package/src/services/voice/corpus-generator.test.ts +201 -0
- package/src/services/voice/corpus-generator.ts +413 -0
- package/src/services/voice/diarization-error-rate.greedy.test.ts +140 -0
- package/src/services/voice/diarization-error-rate.test.ts +100 -0
- package/src/services/voice/diarization-error-rate.ts +249 -0
- package/src/services/voice/e2e-harness.der.test.ts +94 -0
- package/src/services/voice/e2e-harness.respond-eot-entity.test.ts +277 -0
- package/src/services/voice/e2e-harness.security-echo.test.ts +103 -0
- package/src/services/voice/e2e-harness.test.ts +182 -0
- package/src/services/voice/e2e-harness.ts +902 -0
- package/src/services/voice/eager-context-builder.ts +262 -0
- package/src/services/voice/echo-delay.test.ts +118 -0
- package/src/services/voice/echo-delay.ts +135 -0
- package/src/services/voice/echo-metrics.test.ts +17 -0
- package/src/services/voice/echo-metrics.ts +20 -0
- package/src/services/voice/echo-reference-buffer.test.ts +86 -0
- package/src/services/voice/echo-reference-buffer.ts +165 -0
- package/src/services/voice/eliza1-eot-scorer.ts +242 -0
- package/src/services/voice/embedding-server.ts +200 -0
- package/src/services/voice/embedding.test.ts +131 -0
- package/src/services/voice/embedding.ts +242 -0
- package/src/services/voice/emotion-attribution.test.ts +129 -0
- package/src/services/voice/emotion-attribution.ts +361 -0
- package/src/services/voice/engine-bridge-cancellation.test.ts +422 -0
- package/src/services/voice/engine-bridge-transcript-join.test.ts +278 -0
- package/src/services/voice/engine-bridge.test.ts +384 -0
- package/src/services/voice/engine-bridge.ts +2343 -0
- package/src/services/voice/eot-classifier-ggml.ts +569 -0
- package/src/services/voice/eot-classifier.test.ts +98 -0
- package/src/services/voice/eot-classifier.ts +422 -0
- package/src/services/voice/errors.ts +34 -0
- package/src/services/voice/expressive-tags.asr.test.ts +77 -0
- package/src/services/voice/expressive-tags.test.ts +102 -0
- package/src/services/voice/expressive-tags.ts +405 -0
- package/src/services/voice/ffi-bindings.test.ts +735 -0
- package/src/services/voice/ffi-bindings.ts +3387 -0
- package/src/services/voice/first-line-cache.ts +725 -0
- package/src/services/voice/fused-eot-scorer.ts +139 -0
- package/src/services/voice/index.ts +502 -0
- package/src/services/voice/kokoro/__tests__/kokoro-backend.test.ts +262 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-bridge.real.test.ts +236 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-bridge.test.ts +60 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-discovery.test.ts +277 -0
- package/src/services/voice/kokoro/__tests__/kokoro-ffi-runtime.test.ts +235 -0
- package/src/services/voice/kokoro/__tests__/kokoro-runtime.test.ts +95 -0
- package/src/services/voice/kokoro/__tests__/phonemizer.test.ts +53 -0
- package/src/services/voice/kokoro/__tests__/runtime-selection.test.ts +67 -0
- package/src/services/voice/kokoro/__tests__/voices.test.ts +57 -0
- package/src/services/voice/kokoro/index.ts +79 -0
- package/src/services/voice/kokoro/kokoro-backend.ts +223 -0
- package/src/services/voice/kokoro/kokoro-engine-discovery.ts +177 -0
- package/src/services/voice/kokoro/kokoro-ffi-runtime.ts +233 -0
- package/src/services/voice/kokoro/kokoro-runtime.ts +170 -0
- package/src/services/voice/kokoro/phoneme-stream.ts +123 -0
- package/src/services/voice/kokoro/phonemizer.ts +344 -0
- package/src/services/voice/kokoro/pick-runtime.test.ts +91 -0
- package/src/services/voice/kokoro/pick-runtime.ts +130 -0
- package/src/services/voice/kokoro/runtime-selection.ts +64 -0
- package/src/services/voice/kokoro/types.ts +95 -0
- package/src/services/voice/kokoro/voice-presets.ts +129 -0
- package/src/services/voice/kokoro/voices.ts +64 -0
- package/src/services/voice/lifecycle.test.ts +315 -0
- package/src/services/voice/lifecycle.ts +301 -0
- package/src/services/voice/live-diarization-session.echo.test.ts +232 -0
- package/src/services/voice/live-diarization-session.ts +622 -0
- package/src/services/voice/metric-math.test.ts +61 -0
- package/src/services/voice/metric-math.ts +25 -0
- package/src/services/voice/mic-source.test.ts +210 -0
- package/src/services/voice/mic-source.ts +503 -0
- package/src/services/voice/nlms-echo-canceller.test.ts +244 -0
- package/src/services/voice/nlms-echo-canceller.ts +317 -0
- package/src/services/voice/optimistic-policy.power-source.test.ts +36 -0
- package/src/services/voice/optimistic-policy.test.ts +101 -0
- package/src/services/voice/optimistic-policy.ts +192 -0
- package/src/services/voice/optimistic-rollback.ts +343 -0
- package/src/services/voice/partial-stabilizer.test.ts +68 -0
- package/src/services/voice/partial-stabilizer.ts +140 -0
- package/src/services/voice/phoneme-tokenizer.ts +158 -0
- package/src/services/voice/phrase-cache.test.ts +242 -0
- package/src/services/voice/phrase-cache.ts +186 -0
- package/src/services/voice/phrase-chunker.test.ts +239 -0
- package/src/services/voice/phrase-chunker.ts +281 -0
- package/src/services/voice/pipeline-impls.l6.test.ts +110 -0
- package/src/services/voice/pipeline-impls.test.ts +292 -0
- package/src/services/voice/pipeline-impls.ts +315 -0
- package/src/services/voice/pipeline.ts +504 -0
- package/src/services/voice/prefill-client.ts +316 -0
- package/src/services/voice/prefix-preserving-queue.ts +162 -0
- package/src/services/voice/profile-store.ts +887 -0
- package/src/services/voice/real-audio-decode.test.ts +148 -0
- package/src/services/voice/research/VOICE_8785_ASSESSMENT.md +141 -0
- package/src/services/voice/research/VOICE_PIPELINE_RESEARCH_2026.md +117 -0
- package/src/services/voice/research/VOICE_VALIDATION_RUNBOOK.md +135 -0
- package/src/services/voice/ring-buffer.test.ts +129 -0
- package/src/services/voice/ring-buffer.ts +123 -0
- package/src/services/voice/rollback-queue.ts +74 -0
- package/src/services/voice/samantha-preset-placeholder.test.ts +97 -0
- package/src/services/voice/samantha-preset-placeholder.ts +148 -0
- package/src/services/voice/samantha-preset-regenerator.ts +393 -0
- package/src/services/voice/samantha-preset-regenerator.wav.test.ts +90 -0
- package/src/services/voice/scheduler.t2.test.ts +141 -0
- package/src/services/voice/scheduler.ts +927 -0
- package/src/services/voice/self-voice-imprint.test.ts +59 -0
- package/src/services/voice/self-voice-imprint.ts +102 -0
- package/src/services/voice/shared-resources.ts +343 -0
- package/src/services/voice/speaker/attribution-pipeline.test.ts +221 -0
- package/src/services/voice/speaker/attribution-pipeline.ts +449 -0
- package/src/services/voice/speaker/diarizer-fused.real.test.ts +100 -0
- package/src/services/voice/speaker/diarizer-fused.ts +154 -0
- package/src/services/voice/speaker/diarizer.ts +218 -0
- package/src/services/voice/speaker/encoder-fused.real.test.ts +113 -0
- package/src/services/voice/speaker/encoder-fused.ts +138 -0
- package/src/services/voice/speaker/encoder-ggml.test.ts +59 -0
- package/src/services/voice/speaker/encoder-ggml.ts +79 -0
- package/src/services/voice/speaker/encoder.ts +105 -0
- package/src/services/voice/speaker-imprint.test.ts +185 -0
- package/src/services/voice/speaker-imprint.ts +312 -0
- package/src/services/voice/speaker-preset-cache.test.ts +154 -0
- package/src/services/voice/speaker-preset-cache.ts +195 -0
- package/src/services/voice/streaming-asr/streaming-pipeline-adapter.ts +292 -0
- package/src/services/voice/system-audio-sink.test.ts +29 -0
- package/src/services/voice/system-audio-sink.ts +366 -0
- package/src/services/voice/transcriber.asr-backend.test.ts +76 -0
- package/src/services/voice/transcriber.test.ts +392 -0
- package/src/services/voice/transcriber.ts +704 -0
- package/src/services/voice/transcript-knowledge.test.ts +68 -0
- package/src/services/voice/transcript-knowledge.ts +75 -0
- package/src/services/voice/transcript-service.test.ts +195 -0
- package/src/services/voice/transcript-service.ts +205 -0
- package/src/services/voice/transcript-store.test.ts +189 -0
- package/src/services/voice/transcript-store.ts +164 -0
- package/src/services/voice/turn-controller.test.ts +575 -0
- package/src/services/voice/turn-controller.ts +596 -0
- package/src/services/voice/types.ts +699 -0
- package/src/services/voice/vad.test.ts +498 -0
- package/src/services/voice/vad.ts +832 -0
- package/src/services/voice/vad.v1-v4.test.ts +222 -0
- package/src/services/voice/voice-budget.test.ts +415 -0
- package/src/services/voice/voice-budget.ts +635 -0
- package/src/services/voice/voice-duet.test.ts +375 -0
- package/src/services/voice/voice-emotion-classifier.test.ts +210 -0
- package/src/services/voice/voice-emotion-classifier.ts +273 -0
- package/src/services/voice/voice-hardening.fuzz.test.ts +116 -0
- package/src/services/voice/voice-preload-predictor.test.ts +130 -0
- package/src/services/voice/voice-preload-predictor.ts +113 -0
- package/src/services/voice/voice-preset-format.fuzz.test.ts +89 -0
- package/src/services/voice/voice-preset-format.test.ts +75 -0
- package/src/services/voice/voice-preset-format.ts +713 -0
- package/src/services/voice/voice-preset-generator.test.ts +89 -0
- package/src/services/voice/voice-profile-artifact.test.ts +138 -0
- package/src/services/voice/voice-profile-artifact.ts +518 -0
- package/src/services/voice/voice-profile-routes.test.ts +429 -0
- package/src/services/voice/voice-profile-routes.ts +425 -0
- package/src/services/voice/voice-scenario.test.ts +159 -0
- package/src/services/voice/voice-scenario.ts +280 -0
- package/src/services/voice/voice-scenario.turn-helpers.test.ts +77 -0
- package/src/services/voice/voice-state-machine.ts +727 -0
- package/src/services/voice/voice-workbench-report.test.ts +168 -0
- package/src/services/voice/voice-workbench-report.ts +367 -0
- package/src/services/voice/voice-workbench.test.ts +158 -0
- package/src/services/voice/voice.test.ts +1070 -0
- package/src/services/voice/wake-word-ggml.ts +319 -0
- package/src/services/voice/wake-word.test.ts +298 -0
- package/src/services/voice/wake-word.ts +554 -0
- package/src/services/voice/wav-codec.fuzz.test.ts +59 -0
- package/src/services/voice/wav-codec.test.ts +32 -0
- package/src/services/voice/wav-codec.ts +101 -0
- package/src/services/voice/workbench-entrypoint.test.ts +55 -0
- package/src/services/voice/workbench-entrypoint.ts +88 -0
- package/src/services/voice/workbench-headless-runner.test.ts +162 -0
- package/src/services/voice/workbench-headless-runner.ts +396 -0
- package/src/services/voice/workbench-logic-services.test.ts +225 -0
- package/src/services/voice/workbench-logic-services.ts +184 -0
- package/src/services/voice/workbench-real-services.ts +629 -0
- package/src/services/voice/workbench-scenarios.ts +407 -0
- package/src/services/voice/wrap-with-first-line-cache.ts +267 -0
- package/src/services/voice-model-updater.ts +724 -0
- package/src/services/voice-prewarm.ts +51 -0
- package/src/voice-workbench.ts +71 -0
|
@@ -0,0 +1,199 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* CheckpointManager — high-level slot-agnostic KV-cache checkpoint primitive
|
|
3
|
+
* used by the voice state machine (`voice-state-machine.ts`) to implement the
|
|
4
|
+
* optimistic-rollback path described in `docs/eliza-1-optimistic-rollback.md`.
|
|
5
|
+
*
|
|
6
|
+
* Why a manager on top of `CheckpointClient`?
|
|
7
|
+
*
|
|
8
|
+
* - `CheckpointClient` (in `checkpoint-client.ts`) is a thin REST
|
|
9
|
+
* adapter. It is keyed by `(slotId, filename)` because that is what the
|
|
10
|
+
* fork's `POST /slots/<id>/save?filename=<n>` REST API expects.
|
|
11
|
+
* - Callers (the voice state machine, voice-bench drivers) don't want to
|
|
12
|
+
* mint filenames, track active handles, or know about the upstream URL
|
|
13
|
+
* scheme. They want `save("pre-draft") → handle`, `restore(handle)`,
|
|
14
|
+
* `discard(handle)` and an in-memory mock for unit tests.
|
|
15
|
+
* - Backend reference shape is isolated here: the voice state machine keeps
|
|
16
|
+
* the same public interface (`saveCheckpoint`, `restoreCheckpoint`,
|
|
17
|
+
* `discardCheckpoint`) while this module owns how the REST backend stores
|
|
18
|
+
* and addresses the snapshot.
|
|
19
|
+
*
|
|
20
|
+
* The handle returned from `saveCheckpoint` carries enough information to
|
|
21
|
+
* `restore` and `discard` the same snapshot later even after the underlying
|
|
22
|
+
* URL scheme changes. The current backend reference points at
|
|
23
|
+
* `(slotId, filename)` and can evolve without changing callers.
|
|
24
|
+
*
|
|
25
|
+
* **MockCheckpointManager** stores a caller-supplied snapshot (token sequence
|
|
26
|
+
* + arbitrary metadata) in memory keyed by the handle. Tests use it to drive
|
|
27
|
+
* the voice state machine deterministically without spinning up a real
|
|
28
|
+
* checkpoint runtime.
|
|
29
|
+
*/
|
|
30
|
+
import { type CheckpointFetch, type CheckpointHandle as RestCheckpointHandle } from "../checkpoint-client";
|
|
31
|
+
/**
|
|
32
|
+
* Opaque-to-callers handle returned by `saveCheckpoint`. The fields are
|
|
33
|
+
* exposed so tests can assert on them but callers should treat the handle
|
|
34
|
+
* as an opaque blob.
|
|
35
|
+
*/
|
|
36
|
+
export interface CheckpointHandle {
|
|
37
|
+
/**
|
|
38
|
+
* Caller-supplied conversation/turn-scoped slot id. Maps onto the REST
|
|
39
|
+
* `slotId` today; on v1 it maps onto the upstream checkpoint-server's
|
|
40
|
+
* session id.
|
|
41
|
+
*/
|
|
42
|
+
slotId: string;
|
|
43
|
+
/**
|
|
44
|
+
* Human-readable name passed to `saveCheckpoint`. Used as part of the
|
|
45
|
+
* REST filename today (`C1-<slotId>-<name>`).
|
|
46
|
+
*/
|
|
47
|
+
name: string;
|
|
48
|
+
/**
|
|
49
|
+
* Monotonically increasing per-manager id. Lets tests assert that two
|
|
50
|
+
* checkpoints from the same `(slotId, name)` are distinct.
|
|
51
|
+
*/
|
|
52
|
+
id: number;
|
|
53
|
+
/** ISO timestamp of the save call. */
|
|
54
|
+
createdAt: string;
|
|
55
|
+
/**
|
|
56
|
+
* Backend-specific reference. For the REST-backed manager this is the
|
|
57
|
+
* `CheckpointClient` handle (`{slotId: number, filename: string,
|
|
58
|
+
* createdAt: string}`). For the mock manager this is `null` (the mock
|
|
59
|
+
* is keyed by the handle's `id` field).
|
|
60
|
+
*/
|
|
61
|
+
readonly backendRef: RestCheckpointHandle | null;
|
|
62
|
+
}
|
|
63
|
+
/**
|
|
64
|
+
* Common interface implemented by both `CheckpointManager` (REST-backed)
|
|
65
|
+
* and `MockCheckpointManager` (in-memory). The voice state machine accepts
|
|
66
|
+
* this interface so tests can substitute the mock.
|
|
67
|
+
*/
|
|
68
|
+
export interface CheckpointManagerLike {
|
|
69
|
+
/**
|
|
70
|
+
* Snapshot the slot's KV state. Returns a handle the caller passes back
|
|
71
|
+
* to `restoreCheckpoint` / `discardCheckpoint`. Each call returns a new
|
|
72
|
+
* handle; identical `(slotId, name)` pairs do NOT alias.
|
|
73
|
+
*/
|
|
74
|
+
saveCheckpoint(slotId: string, name: string): Promise<CheckpointHandle>;
|
|
75
|
+
/**
|
|
76
|
+
* Restore a previously-saved snapshot. After `restore`, the handle is
|
|
77
|
+
* still valid — the same checkpoint can be restored again (e.g. for two
|
|
78
|
+
* consecutive barge-ins against the same C1).
|
|
79
|
+
*/
|
|
80
|
+
restoreCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
81
|
+
/**
|
|
82
|
+
* Free server-side storage for `handle`. After `discard`, the handle is
|
|
83
|
+
* invalid; subsequent restore/discard calls reject with
|
|
84
|
+
* `CheckpointHandleInvalidError`.
|
|
85
|
+
*/
|
|
86
|
+
discardCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
87
|
+
}
|
|
88
|
+
/**
|
|
89
|
+
* Raised when a caller passes a handle that this manager did not create,
|
|
90
|
+
* or that has already been `discardCheckpoint`-ed. Distinct from REST
|
|
91
|
+
* errors (which surface as `CheckpointHttpError` from the client).
|
|
92
|
+
*/
|
|
93
|
+
export declare class CheckpointHandleInvalidError extends Error {
|
|
94
|
+
constructor(message: string);
|
|
95
|
+
}
|
|
96
|
+
export interface CheckpointManagerOptions {
|
|
97
|
+
/**
|
|
98
|
+
* Base URL of the checkpoint runtime. Same shape as
|
|
99
|
+
* `CheckpointClient` — `http://host:port`.
|
|
100
|
+
*/
|
|
101
|
+
baseUrl: string;
|
|
102
|
+
/**
|
|
103
|
+
* Slot-id-string → numeric slot-id mapping. The REST layer takes a
|
|
104
|
+
* non-negative integer; voice callers prefer string ids that travel
|
|
105
|
+
* with the conversation/turn id. Defaults to a hash of the string.
|
|
106
|
+
*/
|
|
107
|
+
resolveSlotId?: (slotIdString: string) => number;
|
|
108
|
+
/** Optional custom fetch (mostly for unit-testing the REST surface). */
|
|
109
|
+
fetchImpl?: CheckpointFetch;
|
|
110
|
+
/**
|
|
111
|
+
* Default per-request timeout (ms). Forwarded to `CheckpointClient`. The
|
|
112
|
+
* REST checkpoint REST surface is latency-critical on the restore path —
|
|
113
|
+
* keep this short.
|
|
114
|
+
*/
|
|
115
|
+
requestTimeoutMs?: number;
|
|
116
|
+
/**
|
|
117
|
+
* Source of monotonically increasing ids. Injected for deterministic
|
|
118
|
+
* tests; defaults to a per-manager counter.
|
|
119
|
+
*/
|
|
120
|
+
now?: () => Date;
|
|
121
|
+
}
|
|
122
|
+
/**
|
|
123
|
+
* REST-backed `CheckpointManager`. Wraps `CheckpointClient` and exposes
|
|
124
|
+
* the slot-agnostic save/restore/discard contract. The class is stateful
|
|
125
|
+
* only in so far as it tracks which handles are still live so a double-
|
|
126
|
+
* `discard` is detected; the actual snapshot lives in the runtime's
|
|
127
|
+
* `--slot-save-path` directory.
|
|
128
|
+
*/
|
|
129
|
+
export declare class CheckpointManager implements CheckpointManagerLike {
|
|
130
|
+
private readonly client;
|
|
131
|
+
private readonly resolveSlotId;
|
|
132
|
+
private readonly now;
|
|
133
|
+
private nextId;
|
|
134
|
+
private readonly live;
|
|
135
|
+
constructor(opts: CheckpointManagerOptions);
|
|
136
|
+
saveCheckpoint(slotId: string, name: string): Promise<CheckpointHandle>;
|
|
137
|
+
restoreCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
138
|
+
discardCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
139
|
+
/**
|
|
140
|
+
* Probe the underlying runtime for checkpoint support. Forwarded
|
|
141
|
+
* from `CheckpointClient.probeSupported`. Callers gate the feature flag
|
|
142
|
+
* on this.
|
|
143
|
+
*/
|
|
144
|
+
probeSupported(signal?: AbortSignal): Promise<boolean>;
|
|
145
|
+
private assertLive;
|
|
146
|
+
}
|
|
147
|
+
/**
|
|
148
|
+
* Caller-supplied snapshot the mock stores against a handle. The voice
|
|
149
|
+
* state machine tests use this to record the token sequence at the
|
|
150
|
+
* `speech-pause` instant and assert that a barge-in restores the same
|
|
151
|
+
* sequence.
|
|
152
|
+
*/
|
|
153
|
+
export interface MockCheckpointSnapshot {
|
|
154
|
+
/** Token ids at the time of `saveCheckpoint`. */
|
|
155
|
+
tokens: readonly number[];
|
|
156
|
+
/** Free-form metadata (partial transcript, turn id, etc.). */
|
|
157
|
+
metadata?: Record<string, unknown>;
|
|
158
|
+
}
|
|
159
|
+
export type MockSnapshotSource = (slotId: string, name: string) => MockCheckpointSnapshot;
|
|
160
|
+
/**
|
|
161
|
+
* In-memory `CheckpointManager` for tests. Records every save / restore /
|
|
162
|
+
* discard call so tests can assert on them, and stores a snapshot of the
|
|
163
|
+
* token sequence keyed by handle.
|
|
164
|
+
*/
|
|
165
|
+
export declare class MockCheckpointManager implements CheckpointManagerLike {
|
|
166
|
+
private readonly snapshotSource?;
|
|
167
|
+
private nextId;
|
|
168
|
+
private readonly snapshots;
|
|
169
|
+
/**
|
|
170
|
+
* Operations recorded in arrival order. Useful for assertions like
|
|
171
|
+
* "discard happened after restore".
|
|
172
|
+
*/
|
|
173
|
+
readonly operations: Array<{
|
|
174
|
+
kind: "save";
|
|
175
|
+
slotId: string;
|
|
176
|
+
name: string;
|
|
177
|
+
handleId: number;
|
|
178
|
+
} | {
|
|
179
|
+
kind: "restore";
|
|
180
|
+
handleId: number;
|
|
181
|
+
} | {
|
|
182
|
+
kind: "discard";
|
|
183
|
+
handleId: number;
|
|
184
|
+
}>;
|
|
185
|
+
/**
|
|
186
|
+
* Token sequence the active "slot" most recently restored to. Lets
|
|
187
|
+
* tests assert that a restore actually replayed the saved tokens.
|
|
188
|
+
*/
|
|
189
|
+
currentTokens: readonly number[];
|
|
190
|
+
constructor(snapshotSource?: MockSnapshotSource | undefined);
|
|
191
|
+
saveCheckpoint(slotId: string, name: string): Promise<CheckpointHandle>;
|
|
192
|
+
restoreCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
193
|
+
discardCheckpoint(handle: CheckpointHandle): Promise<void>;
|
|
194
|
+
/** Live handles count — for leak assertions. */
|
|
195
|
+
liveHandleCount(): number;
|
|
196
|
+
/** Look up the snapshot saved against `handle.id`. */
|
|
197
|
+
snapshotFor(handle: CheckpointHandle): MockCheckpointSnapshot | undefined;
|
|
198
|
+
}
|
|
199
|
+
//# sourceMappingURL=checkpoint-manager.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"checkpoint-manager.d.ts","sourceRoot":"","sources":["../../../src/services/voice/checkpoint-manager.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,OAAO,EAEN,KAAK,eAAe,EACpB,KAAK,gBAAgB,IAAI,oBAAoB,EAC7C,MAAM,sBAAsB,CAAC;AAE9B;;;;GAIG;AACH,MAAM,WAAW,gBAAgB;IAChC;;;;OAIG;IACH,MAAM,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,IAAI,EAAE,MAAM,CAAC;IACb;;;OAGG;IACH,EAAE,EAAE,MAAM,CAAC;IACX,sCAAsC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB;;;;;OAKG;IACH,QAAQ,CAAC,UAAU,EAAE,oBAAoB,GAAG,IAAI,CAAC;CACjD;AAED;;;;GAIG;AACH,MAAM,WAAW,qBAAqB;IACrC;;;;OAIG;IACH,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAAC;IACxE;;;;OAIG;IACH,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3D;;;;OAIG;IACH,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC3D;AAED;;;;GAIG;AACH,qBAAa,4BAA6B,SAAQ,KAAK;gBAC1C,OAAO,EAAE,MAAM;CAI3B;AAMD,MAAM,WAAW,wBAAwB;IACxC;;;OAGG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;;;OAIG;IACH,aAAa,CAAC,EAAE,CAAC,YAAY,EAAE,MAAM,KAAK,MAAM,CAAC;IACjD,wEAAwE;IACxE,SAAS,CAAC,EAAE,eAAe,CAAC;IAC5B;;;;OAIG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;;;OAGG;IACH,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC;CACjB;AAID;;;;;;GAMG;AACH,qBAAa,iBAAkB,YAAW,qBAAqB;IAC9D,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAmB;IAC1C,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAmC;IACjE,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAa;IACjC,OAAO,CAAC,MAAM,CAAK;IACnB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAqB;gBAE9B,IAAI,EAAE,wBAAwB;IAYpC,cAAc,CACnB,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,GACV,OAAO,CAAC,gBAAgB,CAAC;IAmBtB,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC;IAc1D,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC;IAchE;;;;OAIG;IACG,cAAc,CAAC,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC;IAI5D,OAAO,CAAC,UAAU;CAOlB;AAMD;;;;;GAKG;AACH,MAAM,WAAW,sBAAsB;IACtC,iDAAiD;IACjD,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC;IAC1B,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACnC;AAED,MAAM,MAAM,kBAAkB,GAAG,CAChC,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,KACR,sBAAsB,CAAC;AAE5B;;;;GAIG;AACH,qBAAa,qBAAsB,YAAW,qBAAqB;IAkBtD,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC;IAjB5C,OAAO,CAAC,MAAM,CAAK;IACnB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAA6C;IACvE;;;OAGG;IACH,QAAQ,CAAC,UAAU,EAAE,KAAK,CACvB;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GAChE;QAAE,IAAI,EAAE,SAAS,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GACrC;QAAE,IAAI,EAAE,SAAS,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CACvC,CAAM;IACP;;;OAGG;IACH,aAAa,EAAE,SAAS,MAAM,EAAE,CAAM;gBAET,cAAc,CAAC,EAAE,kBAAkB,YAAA;IAE1D,cAAc,CACnB,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,GACV,OAAO,CAAC,gBAAgB,CAAC;IAkBtB,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC;IAW1D,iBAAiB,CAAC,MAAM,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC;IAUhE,gDAAgD;IAChD,eAAe,IAAI,MAAM;IAIzB,sDAAsD;IACtD,WAAW,CAAC,MAAM,EAAE,gBAAgB,GAAG,sBAAsB,GAAG,SAAS;CAGzE"}
|
|
@@ -0,0 +1,178 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Voice checkpoint policy — thin VAD-event → checkpoint-op translator that
|
|
3
|
+
* sits between the voice loop (turn-controller / pipeline / vad) and the
|
|
4
|
+
* `GatedCheckpointManager` (`../checkpoint-manager.ts`).
|
|
5
|
+
*
|
|
6
|
+
* Why a separate policy module?
|
|
7
|
+
*
|
|
8
|
+
* - The constraint envelope on this scaffold explicitly forbids editing
|
|
9
|
+
* `turn-controller.ts`, `pipeline.ts`, `pipeline-impls.ts`, `vad.ts`,
|
|
10
|
+
* `scheduler.ts`, `phrase-chunker.ts`, `barge-in.ts`, `transcriber.ts`,
|
|
11
|
+
* and anything under `voice/kokoro/` / `voice/streaming-asr/`. Those
|
|
12
|
+
* files are owned by other agents.
|
|
13
|
+
* - But the upstream merge for `--ctx-checkpoints` lands "any week now,"
|
|
14
|
+
* and the JS-side rollback policy is what the merge unlocks. So the
|
|
15
|
+
* policy lives here as a free-standing module that the turn controller
|
|
16
|
+
* can pick up in a follow-up PR by injecting it into its VAD handler
|
|
17
|
+
* and calling `onSpeechPause` / `onSpeechResume` / `onSpeechEndCommit`
|
|
18
|
+
* / `onHardStop` at the matching transitions.
|
|
19
|
+
* - The wiring required in `turn-controller.ts` is documented in the
|
|
20
|
+
* `WIRING-INSTRUCTIONS` comment at the bottom of this file and in
|
|
21
|
+
* `docs/eliza-1-ctx-checkpoints-integration.md`. We intentionally do
|
|
22
|
+
* NOT apply the wiring here — that is a follow-up PR scoped to the
|
|
23
|
+
* turn-controller owner.
|
|
24
|
+
*
|
|
25
|
+
* Policy summary (one C1 per turn, named `pre-speculative-T<turnId>`):
|
|
26
|
+
*
|
|
27
|
+
* - `onSpeechPause(turnId)` — VAD reports the user stopped speaking but
|
|
28
|
+
* hangover hasn't elapsed. Save C1 and let the caller kick the
|
|
29
|
+
* speculative drafter. If the save fails the policy logs and continues
|
|
30
|
+
* (callers MUST treat speculative work as best-effort).
|
|
31
|
+
*
|
|
32
|
+
* - `onSpeechResume(turnId)` — VAD fires `speech-active` within the
|
|
33
|
+
* rollback window. If we previously kicked a speculative draft (the
|
|
34
|
+
* caller flips `speculativeFired=true` to tell us), restore C1 so the
|
|
35
|
+
* KV state is rolled back to the pre-draft point. Otherwise no-op.
|
|
36
|
+
*
|
|
37
|
+
* - `onSpeechEndCommit(turnId)` — VAD's hangover elapsed; the pause was a
|
|
38
|
+
* real turn boundary. The speculative draft is promoted. Erase C1: we
|
|
39
|
+
* no longer need a rollback target for this turn.
|
|
40
|
+
*
|
|
41
|
+
* - `onHardStop(turnId)` — caller-initiated cancellation (e.g. user
|
|
42
|
+
* pressed mute, app backgrounded). If C1 exists, prefer restoring to
|
|
43
|
+
* it so the KV cache is in a known-clean state for the next turn; if
|
|
44
|
+
* C1 isn't around, fall back to `cancel` (the gated manager will
|
|
45
|
+
* either issue `DELETE /slots/<id>` or invoke the SSE-disconnect
|
|
46
|
+
* callback depending on the gate).
|
|
47
|
+
*
|
|
48
|
+
* All four hooks are idempotent and survive a missing C1 by no-op'ing.
|
|
49
|
+
* Errors from the underlying manager are caught and reported through the
|
|
50
|
+
* `events.onError` sink — the policy NEVER throws back into the voice
|
|
51
|
+
* loop, because a failing checkpoint endpoint must not be able to break
|
|
52
|
+
* audio.
|
|
53
|
+
*
|
|
54
|
+
* The policy holds no state of its own beyond the per-turn name; the
|
|
55
|
+
* `GatedCheckpointManager` owns the registry, the REST client, and the
|
|
56
|
+
* capability cache.
|
|
57
|
+
*
|
|
58
|
+
* --- WIRING-INSTRUCTIONS (turn-controller.ts) -----------------------------
|
|
59
|
+
*
|
|
60
|
+
* The turn-controller owner adds (after the upstream merge lands):
|
|
61
|
+
*
|
|
62
|
+
* 1. Construct a `GatedCheckpointManager` once at session start and
|
|
63
|
+
* pass it into a `CheckpointPolicy` instance (one per slot).
|
|
64
|
+
* 2. In the VAD `speech-pause` handler, immediately after the pause
|
|
65
|
+
* hangover timer is armed:
|
|
66
|
+
*
|
|
67
|
+
* await policy.onSpeechPause(this.turnId, this.slotId);
|
|
68
|
+
* // ...kick speculative drafter against the partial transcript
|
|
69
|
+
*
|
|
70
|
+
* 3. In the VAD `speech-active` handler (only when arriving within the
|
|
71
|
+
* rollback window — the controller already tracks this):
|
|
72
|
+
*
|
|
73
|
+
* await policy.onSpeechResume(this.turnId, this.slotId, {
|
|
74
|
+
* speculativeFired: this.speculativeFired,
|
|
75
|
+
* });
|
|
76
|
+
* // ...abort the speculative drafter
|
|
77
|
+
*
|
|
78
|
+
* 4. In the `speech-end` → SPEAKING transition (after the verifier
|
|
79
|
+
* promotes the draft):
|
|
80
|
+
*
|
|
81
|
+
* await policy.onSpeechEndCommit(this.turnId, this.slotId);
|
|
82
|
+
*
|
|
83
|
+
* 5. In the `dispose()` path and any other hard-stop site (mute, app
|
|
84
|
+
* background, error shutdown, barge-in mid-SPEAKING):
|
|
85
|
+
*
|
|
86
|
+
* await policy.onHardStop(this.turnId, this.slotId, () => {
|
|
87
|
+
* this.speculativeAbort?.abort(); // SSE-disconnect callback
|
|
88
|
+
* });
|
|
89
|
+
*
|
|
90
|
+
* 6. Feature flag: pass `useCtxCheckpoints` through to the
|
|
91
|
+
* `GatedCheckpointManager` constructor; when off the policy still
|
|
92
|
+
* runs but every call is a logged no-op.
|
|
93
|
+
*
|
|
94
|
+
* The turn-controller must NOT call `mgr.save/restore/erase/cancel`
|
|
95
|
+
* directly — those names are reserved for the policy so the gated/no-op
|
|
96
|
+
* branching stays in one place. The `policy.events.onError` sink lets the
|
|
97
|
+
* controller forward checkpoint failures into its existing voice-loop
|
|
98
|
+
* telemetry without coupling to the REST error type.
|
|
99
|
+
*/
|
|
100
|
+
import type { CheckpointHandle, GatedCheckpointManager, SseDisconnectFn } from "../checkpoint-manager";
|
|
101
|
+
/**
|
|
102
|
+
* Errors are surfaced through this sink rather than rethrown. The voice
|
|
103
|
+
* loop wires it into its existing telemetry; tests assert on it directly.
|
|
104
|
+
*/
|
|
105
|
+
export interface CheckpointPolicyEvents {
|
|
106
|
+
onError?(op: "save" | "restore" | "erase" | "cancel", error: unknown, turnId: string): void;
|
|
107
|
+
/**
|
|
108
|
+
* Called after a successful save so callers can record the handle in
|
|
109
|
+
* their per-turn state if they want to bypass the name-based lookup on
|
|
110
|
+
* the matching restore.
|
|
111
|
+
*/
|
|
112
|
+
onSaved?(turnId: string, handle: CheckpointHandle): void;
|
|
113
|
+
/** Called after a successful restore. */
|
|
114
|
+
onRestored?(turnId: string, handle: CheckpointHandle): void;
|
|
115
|
+
/** Called when the policy decides to no-op (registry miss, gate off). */
|
|
116
|
+
onNoop?(op: "save" | "restore" | "erase" | "cancel", turnId: string, reason: "gate-off" | "registry-miss" | "no-speculative"): void;
|
|
117
|
+
}
|
|
118
|
+
export interface CheckpointPolicyOptions {
|
|
119
|
+
/** Gated manager. Owned by the caller; one per session. */
|
|
120
|
+
manager: GatedCheckpointManager;
|
|
121
|
+
/** Events sink (errors + observability). Optional. */
|
|
122
|
+
events?: CheckpointPolicyEvents;
|
|
123
|
+
}
|
|
124
|
+
/** Optional second arg to `onSpeechResume` so the policy knows whether
|
|
125
|
+
* a speculative draft actually fired. When `false`, the resume is a no-op
|
|
126
|
+
* (no draft means nothing to roll back).
|
|
127
|
+
*/
|
|
128
|
+
export interface SpeechResumeContext {
|
|
129
|
+
speculativeFired: boolean;
|
|
130
|
+
}
|
|
131
|
+
/**
|
|
132
|
+
* Voice checkpoint policy. Stateless w.r.t. checkpoints (the manager owns
|
|
133
|
+
* the registry) — only holds the manager + event sink. One instance per
|
|
134
|
+
* voice session is enough; the `turnId` argument scopes each operation.
|
|
135
|
+
*/
|
|
136
|
+
export declare class CheckpointPolicy {
|
|
137
|
+
private readonly manager;
|
|
138
|
+
private readonly events;
|
|
139
|
+
constructor(opts: CheckpointPolicyOptions);
|
|
140
|
+
/**
|
|
141
|
+
* VAD `speech-pause`. Save C1. Caller kicks the speculative drafter on
|
|
142
|
+
* its own — the policy doesn't care; it just guarantees the rollback
|
|
143
|
+
* target exists.
|
|
144
|
+
*/
|
|
145
|
+
onSpeechPause(turnId: string, slotId: number): Promise<void>;
|
|
146
|
+
/**
|
|
147
|
+
* VAD `speech-active` within the rollback window. Restore C1 ONLY if
|
|
148
|
+
* the caller actually kicked a speculative draft — otherwise the KV
|
|
149
|
+
* state hasn't been mutated and we'd be doing a needless REST round
|
|
150
|
+
* trip.
|
|
151
|
+
*/
|
|
152
|
+
onSpeechResume(turnId: string, slotId: number, ctx: SpeechResumeContext): Promise<void>;
|
|
153
|
+
/**
|
|
154
|
+
* VAD's hangover elapsed → real turn boundary. Speculative draft is
|
|
155
|
+
* being promoted, so C1 is no longer needed. Erase frees the registry
|
|
156
|
+
* slot (the server-side LRU handles its own eviction independently).
|
|
157
|
+
*/
|
|
158
|
+
onSpeechEndCommit(turnId: string, slotId: number): Promise<void>;
|
|
159
|
+
/**
|
|
160
|
+
* Hard-stop: caller-initiated cancellation. Prefer rolling back to C1
|
|
161
|
+
* (clean KV state for the next turn) when available, else cancel any
|
|
162
|
+
* in-flight decode on the slot. `sseDisconnect` is the existing voice-
|
|
163
|
+
* loop abort hook — required because the gated manager falls back to
|
|
164
|
+
* it when the REST endpoints aren't available.
|
|
165
|
+
*/
|
|
166
|
+
onHardStop(turnId: string, slotId: number, sseDisconnect: SseDisconnectFn): Promise<void>;
|
|
167
|
+
}
|
|
168
|
+
/**
|
|
169
|
+
* Per-turn checkpoint name. Keeps the namespace stable so a hard-stop
|
|
170
|
+
* after a normal commit doesn't collide with the next turn's C1.
|
|
171
|
+
*
|
|
172
|
+
* The format is the only thing callers outside the policy ever see —
|
|
173
|
+
* `GatedCheckpointManager.getNamedHandle('pre-speculative-T123')` returns
|
|
174
|
+
* the same handle the policy used. Keep it stable; if the format changes
|
|
175
|
+
* in a later change, audit every consumer of `getNamedHandle`.
|
|
176
|
+
*/
|
|
177
|
+
export declare function checkpointNameFor(turnId: string): string;
|
|
178
|
+
//# sourceMappingURL=checkpoint-policy.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"checkpoint-policy.d.ts","sourceRoot":"","sources":["../../../src/services/voice/checkpoint-policy.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkGG;AAGH,OAAO,KAAK,EACX,gBAAgB,EAChB,sBAAsB,EACtB,eAAe,EACf,MAAM,uBAAuB,CAAC;AAE/B;;;GAGG;AACH,MAAM,WAAW,sBAAsB;IACtC,OAAO,CAAC,CACP,EAAE,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,QAAQ,EAC3C,KAAK,EAAE,OAAO,EACd,MAAM,EAAE,MAAM,GACZ,IAAI,CAAC;IACR;;;;OAIG;IACH,OAAO,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACzD,yCAAyC;IACzC,UAAU,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAC;IAC5D,yEAAyE;IACzE,MAAM,CAAC,CACN,EAAE,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,GAAG,QAAQ,EAC3C,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,UAAU,GAAG,eAAe,GAAG,gBAAgB,GACrD,IAAI,CAAC;CACR;AAED,MAAM,WAAW,uBAAuB;IACvC,2DAA2D;IAC3D,OAAO,EAAE,sBAAsB,CAAC;IAChC,sDAAsD;IACtD,MAAM,CAAC,EAAE,sBAAsB,CAAC;CAChC;AAED;;;GAGG;AACH,MAAM,WAAW,mBAAmB;IACnC,gBAAgB,EAAE,OAAO,CAAC;CAC1B;AAED;;;;GAIG;AACH,qBAAa,gBAAgB;IAC5B,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAyB;IACjD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAyB;gBAEpC,IAAI,EAAE,uBAAuB;IAKzC;;;;OAIG;IACG,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IA0BlE;;;;;OAKG;IACG,cAAc,CACnB,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,EACd,GAAG,EAAE,mBAAmB,GACtB,OAAO,CAAC,IAAI,CAAC;IAqChB;;;;OAIG;IACG,iBAAiB,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAoBtE;;;;;;OAMG;IACG,UAAU,CACf,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,EACd,aAAa,EAAE,eAAe,GAC5B,OAAO,CAAC,IAAI,CAAC;CAyChB;AAED;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAOxD"}
|
|
@@ -0,0 +1,111 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Voice Workbench corpus augmentation — acoustic degradation DSP (#8785).
|
|
3
|
+
*
|
|
4
|
+
* Real rooms are not clean. The corpus generator produces dry speech; this
|
|
5
|
+
* module degrades it the way a microphone actually hears it: additive room
|
|
6
|
+
* noise at a target SNR, reverberation (near vs far), far-field attenuation,
|
|
7
|
+
* a low-quality/telephone line, and competing background talkers. Every
|
|
8
|
+
* function is PURE and DETERMINISTIC (seeded PRNG, no `Math.random`, no I/O), so
|
|
9
|
+
* the same scenario + seed always yields byte-identical audio — a labeled,
|
|
10
|
+
* reproducible corpus the real ASR/diarization/EOT models can be benchmarked
|
|
11
|
+
* against, and the DSP itself is unit-testable in CI with no models.
|
|
12
|
+
*
|
|
13
|
+
* Layering: this module knows nothing about scenarios. It operates on mono
|
|
14
|
+
* `Float32Array` PCM at a given sample rate. `corpus-generator.ts` translates a
|
|
15
|
+
* scenario's declarative {@link AugmentationSpec} into these calls.
|
|
16
|
+
*/
|
|
17
|
+
/** Root-mean-square amplitude over [start, end). */
|
|
18
|
+
export declare function measureRms(pcm: Float32Array, start?: number, end?: number): number;
|
|
19
|
+
/** Peak absolute amplitude over [start, end). */
|
|
20
|
+
export declare function measurePeak(pcm: Float32Array, start?: number, end?: number): number;
|
|
21
|
+
/** Linear amplitude ratio for a dB gain (+6 dB ≈ ×2, −6 dB ≈ ×0.5). */
|
|
22
|
+
export declare function dbToGain(db: number): number;
|
|
23
|
+
/** Estimated SNR (dB) of a signal region against a noise-only region. */
|
|
24
|
+
export declare function estimateSnrDb(signalRms: number, noiseRms: number): number;
|
|
25
|
+
export type NoiseKind = "white" | "pink" | "music";
|
|
26
|
+
/**
|
|
27
|
+
* Add background noise at a target SNR (dB) relative to the signal's voiced RMS.
|
|
28
|
+
* Lower `snrDb` = noisier. `pink` is a one-pole-filtered approximation of 1/f
|
|
29
|
+
* room rumble; `white` is flat; `music` is a seeded harmonic chord (a few
|
|
30
|
+
* detuned partials under a slow tremolo) — tonal and sustained, the kind of
|
|
31
|
+
* steady background that fools an energy-only VAD where flat hiss would not.
|
|
32
|
+
* The noise floor is added across the WHOLE stream (including silent gaps) so
|
|
33
|
+
* silence is no longer pristine — exactly the condition that makes a real
|
|
34
|
+
* VAD/EOT classifier work for its living.
|
|
35
|
+
*/
|
|
36
|
+
export declare function addNoise(pcm: Float32Array, opts: {
|
|
37
|
+
snrDb: number;
|
|
38
|
+
kind?: NoiseKind;
|
|
39
|
+
seed?: number;
|
|
40
|
+
}): Float32Array;
|
|
41
|
+
/**
|
|
42
|
+
* Simulate a low-quality / telephone line: band-limit to ~300–3400 Hz then
|
|
43
|
+
* companded 8-bit quantization. Cheap mics and phone codecs strip the highs and
|
|
44
|
+
* add quantization grunge, which is the dominant real-world ASR stressor for
|
|
45
|
+
* "voices near and far" / low-quality input.
|
|
46
|
+
*/
|
|
47
|
+
export declare function applyLowQualityLine(pcm: Float32Array, sampleRate: number): Float32Array;
|
|
48
|
+
/**
|
|
49
|
+
* Freeverb-style Schroeder reverb (4 parallel combs → 2 series allpasses),
|
|
50
|
+
* mixed with the dry signal. `room` (0..1) sets reflection density/decay;
|
|
51
|
+
* `wet` (0..1) the reflected level. Reverb spreads energy in time — speech
|
|
52
|
+
* keeps ringing after the talker stops — which is what makes a far/reverberant
|
|
53
|
+
* voice hard to endpoint and to diarize. The output is `dry.length + tail`
|
|
54
|
+
* samples so the decay is preserved (callers may keep or trim the tail).
|
|
55
|
+
*/
|
|
56
|
+
export declare function applyReverb(pcm: Float32Array, sampleRate: number, opts?: {
|
|
57
|
+
room?: number;
|
|
58
|
+
wet?: number;
|
|
59
|
+
tailSec?: number;
|
|
60
|
+
}): Float32Array;
|
|
61
|
+
/** Multiply the whole stream by a dB gain (far-field attenuation = negative). */
|
|
62
|
+
export declare function applyGainDb(pcm: Float32Array, db: number): Float32Array;
|
|
63
|
+
/**
|
|
64
|
+
* Mix an overlay stream (a competing talker, babble, or the agent's own TTS for
|
|
65
|
+
* an echo test) into a base stream at a given level, starting at `offsetSamples`
|
|
66
|
+
* and optionally looping the overlay to cover the base. The base length is
|
|
67
|
+
* preserved. Returns a new array; neither input is mutated.
|
|
68
|
+
*/
|
|
69
|
+
export declare function mixInto(base: Float32Array, overlay: Float32Array, opts?: {
|
|
70
|
+
gainDb?: number;
|
|
71
|
+
offsetSamples?: number;
|
|
72
|
+
loop?: boolean;
|
|
73
|
+
}): Float32Array;
|
|
74
|
+
/** Declarative degradation for one stream (the scenario's `environment`). */
|
|
75
|
+
export interface AugmentationSpec {
|
|
76
|
+
/** Additive room-noise SNR (dB) relative to voiced speech. Lower = noisier. */
|
|
77
|
+
noiseSnrDb?: number;
|
|
78
|
+
/** Noise character (default white). */
|
|
79
|
+
noiseKind?: NoiseKind;
|
|
80
|
+
/** Reverb room size 0..1 (near→far, small→large room). */
|
|
81
|
+
reverb?: number;
|
|
82
|
+
/** Reverb wet level 0..1 (defaults from `reverb` when omitted). */
|
|
83
|
+
reverbWet?: number;
|
|
84
|
+
/** Far-field attenuation in dB (how many dB QUIETER; positive number). */
|
|
85
|
+
farFieldDb?: number;
|
|
86
|
+
/** Band-limit + 8-bit companding (telephone / cheap-mic line). */
|
|
87
|
+
lowQuality?: boolean;
|
|
88
|
+
/** Competing background talkers, mixed this many dB BELOW the speech. */
|
|
89
|
+
backgroundTalkersDb?: number;
|
|
90
|
+
/** Deterministic seed for noise/babble. */
|
|
91
|
+
seed?: number;
|
|
92
|
+
}
|
|
93
|
+
/** True when the spec asks for any degradation at all. */
|
|
94
|
+
export declare function specIsClean(spec: AugmentationSpec | undefined): boolean;
|
|
95
|
+
export interface AugmentPcmOptions {
|
|
96
|
+
/**
|
|
97
|
+
* A babble source (a competing-talker stream) for `backgroundTalkersDb`.
|
|
98
|
+
* The generator supplies one synthesized from other voices; omitted = no
|
|
99
|
+
* background talkers even if the spec asks (the runner logs the gap).
|
|
100
|
+
*/
|
|
101
|
+
babble?: Float32Array;
|
|
102
|
+
/** Trim reverb tail back to the input length (keeps corpus timing exact). */
|
|
103
|
+
trimReverbTail?: boolean;
|
|
104
|
+
}
|
|
105
|
+
/**
|
|
106
|
+
* Apply a full degradation chain to one stream, in acoustically sensible order:
|
|
107
|
+
* background talkers → reverb (room reflections) → far-field gain → low-quality
|
|
108
|
+
* line → additive noise floor. Pure; returns a new array.
|
|
109
|
+
*/
|
|
110
|
+
export declare function augmentPcm(pcm: Float32Array, sampleRate: number, spec: AugmentationSpec, options?: AugmentPcmOptions): Float32Array;
|
|
111
|
+
//# sourceMappingURL=corpus-augment.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"corpus-augment.d.ts","sourceRoot":"","sources":["../../../src/services/voice/corpus-augment.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAuBH,oDAAoD;AACpD,wBAAgB,UAAU,CACzB,GAAG,EAAE,YAAY,EACjB,KAAK,SAAI,EACT,GAAG,SAAa,GACd,MAAM,CAOR;AAED,iDAAiD;AACjD,wBAAgB,WAAW,CAC1B,GAAG,EAAE,YAAY,EACjB,KAAK,SAAI,EACT,GAAG,SAAa,GACd,MAAM,CASR;AAED,uEAAuE;AACvE,wBAAgB,QAAQ,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,CAE3C;AAED,yEAAyE;AACzE,wBAAgB,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAIzE;AAED,MAAM,MAAM,SAAS,GAAG,OAAO,GAAG,MAAM,GAAG,OAAO,CAAC;AAEnD;;;;;;;;;GASG;AACH,wBAAgB,QAAQ,CACvB,GAAG,EAAE,YAAY,EACjB,IAAI,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,SAAS,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAA;CAAE,GACtD,YAAY,CAoDd;AAkFD;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAClC,GAAG,EAAE,YAAY,EACjB,UAAU,EAAE,MAAM,GAChB,YAAY,CAKd;AAED;;;;;;;GAOG;AACH,wBAAgB,WAAW,CAC1B,GAAG,EAAE,YAAY,EACjB,UAAU,EAAE,MAAM,EAClB,IAAI,GAAE;IAAE,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAA;CAAO,GAC1D,YAAY,CA+Cd;AAED,iFAAiF;AACjF,wBAAgB,WAAW,CAAC,GAAG,EAAE,YAAY,EAAE,EAAE,EAAE,MAAM,GAAG,YAAY,CAKvE;AAED;;;;;GAKG;AACH,wBAAgB,OAAO,CACtB,IAAI,EAAE,YAAY,EAClB,OAAO,EAAE,YAAY,EACrB,IAAI,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,aAAa,CAAC,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,OAAO,CAAA;CAAO,GACpE,YAAY,CAad;AAED,6EAA6E;AAC7E,MAAM,WAAW,gBAAgB;IAChC,+EAA+E;IAC/E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,uCAAuC;IACvC,SAAS,CAAC,EAAE,SAAS,CAAC;IACtB,0DAA0D;IAC1D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,mEAAmE;IACnE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0EAA0E;IAC1E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,kEAAkE;IAClE,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,yEAAyE;IACzE,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,2CAA2C;IAC3C,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,0DAA0D;AAC1D,wBAAgB,WAAW,CAAC,IAAI,EAAE,gBAAgB,GAAG,SAAS,GAAG,OAAO,CASvE;AAED,MAAM,WAAW,iBAAiB;IACjC;;;;OAIG;IACH,MAAM,CAAC,EAAE,YAAY,CAAC;IACtB,6EAA6E;IAC7E,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB;AAED;;;;GAIG;AACH,wBAAgB,UAAU,CACzB,GAAG,EAAE,YAAY,EACjB,UAAU,EAAE,MAAM,EAClB,IAAI,EAAE,gBAAgB,EACtB,OAAO,GAAE,iBAAsB,GAC7B,YAAY,CA0Dd"}
|
|
@@ -0,0 +1,134 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Voice Workbench corpus generator (#8785).
|
|
3
|
+
*
|
|
4
|
+
* Turns a declarative {@link VoiceScenario} into one labeled audio stream + a
|
|
5
|
+
* ground-truth JSON the headless runner scores against. Two synthesis paths,
|
|
6
|
+
* one shape:
|
|
7
|
+
*
|
|
8
|
+
* - Synthetic (default, deterministic, NO native model): formant-resonator
|
|
9
|
+
* speech-like PCM (`__test-helpers__/synthetic-speech.ts`) the Silero VAD
|
|
10
|
+
* reads as speech. Reproducible in CI with no artifacts — it exercises the
|
|
11
|
+
* pipeline plumbing + the scorers/labels, not diarization/WER *accuracy*.
|
|
12
|
+
* - Real TTS (gated): an injected {@link CorpusTtsSynthesizer} (Kokoro /
|
|
13
|
+
* OmniVoice via the TTS route) produces natural speech. Real diarization
|
|
14
|
+
* DER and transcription WER benchmarking need this path.
|
|
15
|
+
*
|
|
16
|
+
* `generateVoiceCorpus` is pure (no I/O) so it is unit-testable without disk;
|
|
17
|
+
* `writeVoiceCorpus` / `readVoiceCorpus` handle the versioned on-disk corpus.
|
|
18
|
+
* A turn's labels (speaker, transcript, respond decision, entity) come straight
|
|
19
|
+
* from the scenario, so the ground truth is reproducible regardless of path.
|
|
20
|
+
*/
|
|
21
|
+
import { type VoiceEnvironment, type VoiceScenario } from "./voice-scenario";
|
|
22
|
+
/** Per-turn ground-truth label with sample-accurate timing. */
|
|
23
|
+
export interface CorpusTurnLabel {
|
|
24
|
+
index: number;
|
|
25
|
+
/** Diarization ground-truth label (the participant who spoke). */
|
|
26
|
+
speaker: string;
|
|
27
|
+
/** Resolved elizaOS entity id for the speaker, when the scenario binds one. */
|
|
28
|
+
entityId?: string;
|
|
29
|
+
/** First sample of voiced speech in this turn (after any lead silence). */
|
|
30
|
+
speechStartSample: number;
|
|
31
|
+
/** Sample just past the voiced speech (before trailing pauses). */
|
|
32
|
+
speechEndSample: number;
|
|
33
|
+
/** First sample of this turn's whole segment in the stream. */
|
|
34
|
+
segmentStartSample: number;
|
|
35
|
+
/** Sample just past this turn's whole segment (incl. trailing pauses). */
|
|
36
|
+
segmentEndSample: number;
|
|
37
|
+
/** Reference transcript for WER scoring. */
|
|
38
|
+
referenceTranscript: string;
|
|
39
|
+
/** Ground truth: should the agent respond to this turn? */
|
|
40
|
+
expectRespond: boolean;
|
|
41
|
+
/** Ground truth: is this segment a real end-of-turn boundary? */
|
|
42
|
+
expectEndOfTurn?: boolean;
|
|
43
|
+
/** Expected inferred/recognized entity, when the scenario asserts one. */
|
|
44
|
+
expectedEntity?: string;
|
|
45
|
+
/** TTS voice id used for this turn (real-TTS path), when set. */
|
|
46
|
+
ttsVoiceId?: string;
|
|
47
|
+
/** True when this turn was formant-synthesized rather than real TTS. */
|
|
48
|
+
synthetic: boolean;
|
|
49
|
+
/** Acoustic degradation applied to this turn's audio (when any). */
|
|
50
|
+
environment?: VoiceEnvironment;
|
|
51
|
+
/** True when this "turn" is the agent's own TTS echoed back (not a user turn). */
|
|
52
|
+
isAgentEcho?: boolean;
|
|
53
|
+
/** Ground truth: the speaker is the device owner / primary enrolled voice. */
|
|
54
|
+
isOwner?: boolean;
|
|
55
|
+
/** The agent's spoken reply to this turn (drives the echo gate downstream). */
|
|
56
|
+
agentReplyText?: string;
|
|
57
|
+
}
|
|
58
|
+
/**
|
|
59
|
+
* On-disk corpus ground-truth schema version. Bump when the labeled-corpus
|
|
60
|
+
* shape changes incompatibly; `readVoiceCorpusGroundTruth` treats a corpus
|
|
61
|
+
* written by a different version as absent (→ `skipped`, never a stale `pass`).
|
|
62
|
+
*/
|
|
63
|
+
export declare const CORPUS_SCHEMA_VERSION = 1;
|
|
64
|
+
export interface CorpusGroundTruth {
|
|
65
|
+
/** Labeled-corpus schema version (see {@link CORPUS_SCHEMA_VERSION}). */
|
|
66
|
+
schemaVersion: number;
|
|
67
|
+
scenarioId: string;
|
|
68
|
+
classes: VoiceScenario["classes"];
|
|
69
|
+
sampleRate: number;
|
|
70
|
+
totalSamples: number;
|
|
71
|
+
durationSec: number;
|
|
72
|
+
participants: Array<{
|
|
73
|
+
label: string;
|
|
74
|
+
entityId?: string;
|
|
75
|
+
isOwner?: boolean;
|
|
76
|
+
ttsVoiceId?: string;
|
|
77
|
+
}>;
|
|
78
|
+
agents?: string[];
|
|
79
|
+
/** Entity ids the agent answers without a wake word (owner + enrolled). */
|
|
80
|
+
knownSpeakerEntityIds?: string[];
|
|
81
|
+
turns: CorpusTurnLabel[];
|
|
82
|
+
/** True when EVERY turn was synthetic (no real TTS used anywhere). */
|
|
83
|
+
synthetic: boolean;
|
|
84
|
+
}
|
|
85
|
+
export interface GeneratedVoiceCorpus {
|
|
86
|
+
pcm: Float32Array;
|
|
87
|
+
sampleRate: number;
|
|
88
|
+
groundTruth: CorpusGroundTruth;
|
|
89
|
+
}
|
|
90
|
+
/**
|
|
91
|
+
* Real-TTS synthesizer. Gated: when omitted, the generator uses deterministic
|
|
92
|
+
* synthetic speech. An implementation wraps the TTS route / Kokoro engine and
|
|
93
|
+
* returns mono PCM at the requested sample rate.
|
|
94
|
+
*/
|
|
95
|
+
export interface CorpusTtsSynthesizer {
|
|
96
|
+
synthesize(args: {
|
|
97
|
+
text: string;
|
|
98
|
+
voiceId?: string;
|
|
99
|
+
speakerLabel: string;
|
|
100
|
+
turnIndex: number;
|
|
101
|
+
isAgentEcho: boolean;
|
|
102
|
+
sampleRate: number;
|
|
103
|
+
}): Promise<Float32Array>;
|
|
104
|
+
}
|
|
105
|
+
export interface GenerateVoiceCorpusOptions {
|
|
106
|
+
sampleRate?: number;
|
|
107
|
+
/** Inject a real-TTS synthesizer to produce natural speech (else synthetic). */
|
|
108
|
+
synthesizer?: CorpusTtsSynthesizer;
|
|
109
|
+
/** Silence (s) spliced after a turn that declares no explicit pauses. */
|
|
110
|
+
interTurnSilenceSec?: number;
|
|
111
|
+
/** Synthetic-speech sizing: characters of text per second of audio. */
|
|
112
|
+
charsPerSecond?: number;
|
|
113
|
+
}
|
|
114
|
+
/**
|
|
115
|
+
* Generate one labeled audio stream for a scenario. Throws on an invalid
|
|
116
|
+
* scenario (fail loud — a malformed corpus must not silently produce garbage).
|
|
117
|
+
* A turn that carries only `audioRef` (no `text`) is not synthesizable here and
|
|
118
|
+
* is rejected; pre-rendered audio is supplied through the corpus on disk.
|
|
119
|
+
*/
|
|
120
|
+
export declare function generateVoiceCorpus(scenario: VoiceScenario, options?: GenerateVoiceCorpusOptions): Promise<GeneratedVoiceCorpus>;
|
|
121
|
+
export interface VoiceCorpusPaths {
|
|
122
|
+
dir: string;
|
|
123
|
+
audioPath: string;
|
|
124
|
+
groundTruthPath: string;
|
|
125
|
+
}
|
|
126
|
+
/** Persist a generated corpus as `audio.wav` + `ground-truth.json` under `dir`. */
|
|
127
|
+
export declare function writeVoiceCorpus(corpus: GeneratedVoiceCorpus, dir: string): VoiceCorpusPaths;
|
|
128
|
+
/**
|
|
129
|
+
* Read a previously-written corpus's ground truth. Returns null when the corpus
|
|
130
|
+
* directory or its ground-truth file is absent (the honesty contract — the
|
|
131
|
+
* runner reports `skipped`, never `pass`, when corpus artifacts are missing).
|
|
132
|
+
*/
|
|
133
|
+
export declare function readVoiceCorpusGroundTruth(dir: string): CorpusGroundTruth | null;
|
|
134
|
+
//# sourceMappingURL=corpus-generator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"corpus-generator.d.ts","sourceRoot":"","sources":["../../../src/services/voice/corpus-generator.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAeH,OAAO,EAIN,KAAK,gBAAgB,EACrB,KAAK,aAAa,EAElB,MAAM,kBAAkB,CAAC;AAa1B,+DAA+D;AAC/D,MAAM,WAAW,eAAe;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,kEAAkE;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,+EAA+E;IAC/E,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,2EAA2E;IAC3E,iBAAiB,EAAE,MAAM,CAAC;IAC1B,mEAAmE;IACnE,eAAe,EAAE,MAAM,CAAC;IACxB,+DAA+D;IAC/D,kBAAkB,EAAE,MAAM,CAAC;IAC3B,0EAA0E;IAC1E,gBAAgB,EAAE,MAAM,CAAC;IACzB,4CAA4C;IAC5C,mBAAmB,EAAE,MAAM,CAAC;IAC5B,2DAA2D;IAC3D,aAAa,EAAE,OAAO,CAAC;IACvB,iEAAiE;IACjE,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,0EAA0E;IAC1E,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,iEAAiE;IACjE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,wEAAwE;IACxE,SAAS,EAAE,OAAO,CAAC;IACnB,oEAAoE;IACpE,WAAW,CAAC,EAAE,gBAAgB,CAAC;IAC/B,kFAAkF;IAClF,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,8EAA8E;IAC9E,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,+EAA+E;IAC/E,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAED;;;;GAIG;AACH,eAAO,MAAM,qBAAqB,IAAI,CAAC;AAEvC,MAAM,WAAW,iBAAiB;IACjC,yEAAyE;IACzE,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,aAAa,CAAC,SAAS,CAAC,CAAC;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,KAAK,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,OAAO,CAAC,EAAE,OAAO,CAAC;QAClB,UAAU,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC,CAAC;IACH,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,2EAA2E;IAC3E,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;IACjC,KAAK,EAAE,eAAe,EAAE,CAAC;IACzB,sEAAsE;IACtE,SAAS,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,oBAAoB;IACpC,GAAG,EAAE,YAAY,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,iBAAiB,CAAC;CAC/B;AAED;;;;GAIG;AACH,MAAM,WAAW,oBAAoB;IACpC,UAAU,CAAC,IAAI,EAAE;QAChB,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,YAAY,EAAE,MAAM,CAAC;QACrB,SAAS,EAAE,MAAM,CAAC;QAClB,WAAW,EAAE,OAAO,CAAC;QACrB,UAAU,EAAE,MAAM,CAAC;KACnB,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;CAC1B;AAED,MAAM,WAAW,0BAA0B;IAC1C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,gFAAgF;IAChF,WAAW,CAAC,EAAE,oBAAoB,CAAC;IACnC,yEAAyE;IACzE,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,uEAAuE;IACvE,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAoCD;;;;;GAKG;AACH,wBAAsB,mBAAmB,CACxC,QAAQ,EAAE,aAAa,EACvB,OAAO,GAAE,0BAA+B,GACtC,OAAO,CAAC,oBAAoB,CAAC,CA6K/B;AAED,MAAM,WAAW,gBAAgB;IAChC,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,eAAe,EAAE,MAAM,CAAC;CACxB;AAED,mFAAmF;AACnF,wBAAgB,gBAAgB,CAC/B,MAAM,EAAE,oBAAoB,EAC5B,GAAG,EAAE,MAAM,GACT,gBAAgB,CAUlB;AAED;;;;GAIG;AACH,wBAAgB,0BAA0B,CACzC,GAAG,EAAE,MAAM,GACT,iBAAiB,GAAG,IAAI,CAa1B"}
|