mirror of
https://github.com/mudler/LocalAI.git
synced 2026-07-01 11:56:57 -04:00
* feat(realtime): EOU-driven semantic_vad turn detection Add a `semantic_vad` turn-detection mode to the realtime API that feeds the transcription model live and decides "the user finished speaking" from the `<EOU>` end-of-utterance token rather than from silence alone. When EOU fires the turn commits immediately (~0.3s); otherwise it falls back to an eagerness-scaled silence threshold (low/med/high = 8/4/2s). Plumbing, bottom to top: - proto: `AudioTranscriptionLive` bidirectional RPC (config-first oneof, mono float PCM @16k, ready-ack / Unimplemented degrade signal) plus `TranscriptResult.eou` for the unary retranscribe gate. - pkg/grpc: client/server/base/embed scaffolding for the bidi stream, modeled on AudioTransformStream; release stream conns on terminal Recv. - parakeet-cpp: live transcription RPC with per-C-call engine locking (one live stream per turn, finalize+free at commit); bump parakeet.cpp to ABI v5 — incremental StreamingMel (no more quadratic per-feed mel recompute that delayed EOU on long turns) and the <EOU>/<EOB> split; strip the literal <EOU>/<EOB> from offline text and set Eou. - core/backend: LiveTranscriptionSession wrapper + pipeline `turn_detection:` config block (type/eagerness/retranscribe). - realtime: semantic_vad integration — live input captions streamed as transcription deltas while the user speaks, EOU-immediate commit with eagerness fallback, optional retranscribe gate (batch re-decode must also end in <EOU> to confirm), clause synthesis off the LLM token callback, and per-turn live-transcription / model_load telemetry. - UI: show the realtime pipeline components as a vertical list. Docs and tests included; opt-in via the pipeline YAML or per-session `session.update`. Non-streaming STT backends degrade to silence-only. Assisted-by: Claude Code:claude-opus-4-8 [Read] [Edit] [Write] [Bash] Assisted-by: Claude Code:claude-fable-5 [Read] [Edit] [Bash] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): explicit formally-verified state machines + parakeet streaming driver The realtime API had several implicit state machines whose state was inferred from scattered booleans, channels, and five separate mutexes, leaving illegal/inconsistent states reachable. Make them explicit and keep the implementation in step with a formal design; rework the parakeet streaming backend along the same lines. Realtime state machines (M1-M5). Each is a sealed sum-type State/Event/Effect with a total, pure Next(state,event)->(state,[]effect) behind a single-writer Coordinator: M1 conncoord connection lifecycle: VAD toggle + once-only teardown (replaces vadServerStarted + a `done` channel closed from two sites). M2 turncoord turn detection: collapses speechStarted and the live-stream "turn open" flag into one state, so discardTurn can no longer desync them and suppress the next onset. M3 respcoord response coordination: serializes the dual-writer start/cancel so at most one response is live; one response.done per response.create. M4 compactcoord conversation compaction: single-flight (replaces the `compacting atomic.Bool` CAS). M5 ttscoord TTS pipeline: open->closing->closed, idempotent wait(), rejects enqueue-after-close (was a silent drop). The Coordinator/Sink/Next plumbing — only the sealed types and Next differed per machine — is extracted once into core/http/endpoints/openai/coordinator as a generic Coordinator[S,E,F]; each machine keeps its public API via type aliases, so no sink, call-site, or test moved. Hierarchy. session_lifecycle.fizz models M1 as the parent region with its children (M2/M3/M4) as one statechart and asserts ChildrenDieWithParent (conn torn => all children terminal, none start after teardown). respcoord and compactcoord gain an absorbing Terminated state + Shutdown event; conncoord's teardown drives the children terminal. This closes a compaction teardown gap: a fire-and-forget compaction could outlive a torn session — compactionSink now takes a session-scoped cancellable context + WaitGroup and joins the in-flight summarize+evict on shutdown. Formal verification. formal-verification/ holds one authoritative FizzBee spec per machine plus the composition spec, each with an always-assertion and a documented one-line edit that makes the checker fail (verified non-vacuous). scripts/realtime-conformance.sh is fail-closed: all Go conformance suites under -race AND a model-check of every .fizz spec; a missing FizzBee is a hard error (only the loud REALTIME_CONFORMANCE_SKIP_FIZZBEE=1 bypasses it, never in CI). FizzBee is pinned by sha256 and installed via scripts/install-fizzbee.sh into .tools/ (gitignored). Wired as make test-realtime-conformance, a CI workflow, and a pre-commit path filter. Go conformance tests are Ginkgo/Gomega (per the repo's forbidigo lint): transition tables + fixed-seed property walks + concurrent/-race specs, no rapid dependency. Design map: docs/design/realtime-state-machines.md. Parakeet streaming backend. The same treatment applied to the parakeet-cpp streaming paths: - AudioTranscriptionStream returns codes.Unimplemented for non-streaming models instead of decoding offline and emitting it as one delta + final. A client that asked for streaming learns the model cannot stream rather than receiving a batch result shaped like a stream. New grpcerrors.StreamTranscriptionUnsupported carries that signal; the HTTP /v1/audio/transcriptions stream path surfaces it as an SSE error event. Mirrors AudioTranscriptionLive, which already did this. - utteranceBoundary (boundary.go): a single definition of the end-of-utterance latch, replacing three open-coded finalEou toggles. Modelled as a two-valued type so illegal states are unrepresentable. - Shared decode driver (driver.go): streamFeedResult (one per-feed event) + feedChunk (hides the ABI v4 JSON vs text-only split) + feedSlices + flushTail. The feed loop is written once. - AudioTranscriptionLive becomes a bidi adapter: it streams the per-feed {delta,eou,eob,words} the realtime turn detector consumes and a terminal FinalResult carrying only Text. Segments/duration/eou are offline-only and no longer produced (nor read) on the live path; liveTraceState drops the terminal eou and keeps the per-feed eou_events count. - AudioTranscriptionStream + streamJSON merge into one driver-based function; streamSegmenter is generalized to the unified event with a text-only fallback that preserves the legacy (no-words) library's per-utterance segmentation. Verified: build/vet/gofumpt clean, golangci-lint 0 issues, all coordinator and parakeet packages under -race, the fail-closed conformance gate green, and make test-realtime (12 e2e WS+WebRTC). Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>
80 lines
3.3 KiB
Plaintext
80 lines
3.3 KiB
Plaintext
---
|
|
# Authoritative formal design for realtime machine M2: turn detection.
|
|
#
|
|
# Companion to:
|
|
# - docs/design/realtime-state-machines.md (the map + invariants)
|
|
# - core/http/endpoints/openai/turncoord (the Go implementation)
|
|
#
|
|
# The Go MBT adapter maps each action below onto turncoord.Coordinator.Apply
|
|
# and the StateGetter onto turncoord.Coordinator.State, so this spec is the
|
|
# source of truth the implementation is checked against.
|
|
#
|
|
# The property this machine must guarantee is the COUPLING of two facts the
|
|
# legacy code tracked in two separate variables that could disagree:
|
|
# - speech -- handleVAD's speechStarted bool
|
|
# - turn -- the semantic_vad live-stream-open flag (lts.open())
|
|
# A discardTurn (no-speech clear / mode switch / teardown) closed the live
|
|
# stream (turn -> 0) but left speechStarted set (speech stays 1). They then
|
|
# disagreed, and the next onset was suppressed by `if !speechStarted` -- no
|
|
# speech_started, no barge-in, no commit. See Part 2, failure mode 4.
|
|
#
|
|
# Here speech and turn are driven only ever TOGETHER, modelling the single
|
|
# turncoord State (Idle <-> Speaking) where both facts are one value.
|
|
#
|
|
# NOTE: FizzBee is pre-1.0. Validate the exact syntax/CLI against the version
|
|
# pinned in formal-verification/README.md before trusting the gate.
|
|
deadlock_detection: false
|
|
---
|
|
|
|
# Bound the number of turns so the state space is finite.
|
|
MAX_TURNS = 4
|
|
|
|
role Detector:
|
|
action Init:
|
|
self.speech = 0 # speechStarted (0/1)
|
|
self.turn = 0 # live-stream / turn open (0/1)
|
|
self.turns = 0 # how many turns have been opened (bound)
|
|
|
|
# Onset: VAD reports speech while idle -> open a turn. ONE indivisible
|
|
# transition sets BOTH facts, so they cannot be left disagreeing. Re-onset
|
|
# while already speaking is a no-op (legacy `if !speechStarted`).
|
|
atomic action Onset:
|
|
require self.turns < MAX_TURNS
|
|
if self.speech == 0:
|
|
self.turns += 1
|
|
self.speech = 1
|
|
self.turn = 1
|
|
|
|
# Silence: VAD-confirmed end-of-speech past the dynamic threshold -> commit.
|
|
# Both facts clear together (EmitSpeechStopped + CommitTurn return to Idle).
|
|
atomic action Silence:
|
|
if self.speech == 1:
|
|
self.speech = 0
|
|
self.turn = 0
|
|
|
|
# Abort: no-speech clear / teardown -> discard. BOTH facts clear together.
|
|
# (A semantic->server mode switch only drops the orphaned live stream and
|
|
# lets the turn continue, so it is NOT an Abort -- see turncoord.go.)
|
|
# THE FIX: clearing only `self.turn` here (deleting `self.speech = 0`)
|
|
# reproduces the legacy discardTurn bug --
|
|
# the checker then reports Coupled violated, exactly the desync that
|
|
# suppressed the next onset.
|
|
atomic action Abort:
|
|
if self.turn == 1:
|
|
self.turn = 0
|
|
self.speech = 0
|
|
|
|
action Init:
|
|
d = Detector()
|
|
|
|
# SAFETY: speechStarted and turn-open never disagree -- they are one state, so
|
|
# the legacy desync that suppressed the next onset is unrepresentable
|
|
# (Part 4, invariant #4; failure mode 4).
|
|
always assertion Coupled:
|
|
return d.speech == d.turn
|
|
|
|
# SAFETY: at most one turn open at any instant -- `turn` is a 0/1 fact, never
|
|
# incremented twice without a clear between (onset is a no-op while speaking).
|
|
always assertion AtMostOneTurnOpen:
|
|
return d.turn >= 0 and d.turn <= 1
|