Files
LocalAI/pkg/grpc/client.go
Richard Palethorpe 5d0c43ec6e feat(realtime): Semantic VAD EOU token (#10444)
* 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>
2026-06-30 09:01:22 +02:00

1334 lines
30 KiB
Go

package grpc
import (
"context"
"errors"
"fmt"
"io"
"sync"
"time"
pb "github.com/mudler/LocalAI/pkg/grpc/proto"
"google.golang.org/grpc"
"google.golang.org/grpc/credentials/insecure"
)
const maxGRPCMessageSize = 50 * 1024 * 1024 // 50MB
// bearerToken implements credentials.PerRPCCredentials to inject a bearer token
// into every gRPC call.
type bearerToken struct {
token string
}
func (b bearerToken) GetRequestMetadata(ctx context.Context, uri ...string) (map[string]string, error) {
return map[string]string{"authorization": "Bearer " + b.token}, nil
}
func (b bearerToken) RequireTransportSecurity() bool { return false }
type Client struct {
address string
busy bool
parallel bool
token string
sync.Mutex
opMutex sync.Mutex
wd WatchDog
}
type WatchDog interface {
Mark(address string)
UnMark(address string)
}
func (c *Client) IsBusy() bool {
c.Lock()
defer c.Unlock()
return c.busy
}
func (c *Client) setBusy(v bool) {
c.Lock()
c.busy = v
c.Unlock()
}
func (c *Client) wdMark() {
if c.wd != nil {
c.wd.Mark(c.address)
}
}
func (c *Client) wdUnMark() {
if c.wd != nil {
c.wd.UnMark(c.address)
}
}
// dial creates a gRPC client connection with common options.
// If c.token is set, bearer token credentials are included.
func (c *Client) dial() (*grpc.ClientConn, error) {
opts := []grpc.DialOption{
grpc.WithTransportCredentials(insecure.NewCredentials()),
grpc.WithDefaultCallOptions(
grpc.MaxCallRecvMsgSize(maxGRPCMessageSize),
grpc.MaxCallSendMsgSize(maxGRPCMessageSize),
),
}
if c.token != "" {
opts = append(opts, grpc.WithPerRPCCredentials(bearerToken{token: c.token}))
}
return grpc.NewClient(c.address, opts...)
}
func (c *Client) HealthCheck(ctx context.Context) (bool, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
conn, err := c.dial()
if err != nil {
return false, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
// The healthcheck call shouldn't take long time
ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
defer cancel()
res, err := client.Health(ctx, &pb.HealthMessage{})
if err != nil {
return false, err
}
if string(res.Message) == "OK" {
return true, nil
}
return false, fmt.Errorf("health check failed: %s", res.Message)
}
func (c *Client) Embeddings(ctx context.Context, in *pb.PredictOptions, opts ...grpc.CallOption) (*pb.EmbeddingResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.Embedding(ctx, in, opts...)
}
func (c *Client) Predict(ctx context.Context, in *pb.PredictOptions, opts ...grpc.CallOption) (*pb.Reply, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.Predict(ctx, in, opts...)
}
func (c *Client) LoadModel(ctx context.Context, in *pb.ModelOptions, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.LoadModel(ctx, in, opts...)
}
func (c *Client) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(reply *pb.Reply), opts ...grpc.CallOption) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
stream, err := client.PredictStream(ctx, in, opts...)
if err != nil {
return err
}
for {
// Check if context is cancelled before receiving
select {
case <-ctx.Done():
return ctx.Err()
default:
}
reply, err := stream.Recv()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
// Check if error is due to context cancellation
if ctx.Err() != nil {
return ctx.Err()
}
fmt.Println("Error", err)
return err
}
f(reply)
}
return nil
}
func (c *Client) GenerateImage(ctx context.Context, in *pb.GenerateImageRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.GenerateImage(ctx, in, opts...)
}
func (c *Client) GenerateVideo(ctx context.Context, in *pb.GenerateVideoRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.GenerateVideo(ctx, in, opts...)
}
func (c *Client) TTS(ctx context.Context, in *pb.TTSRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.TTS(ctx, in, opts...)
}
func (c *Client) TTSStream(ctx context.Context, in *pb.TTSRequest, f func(reply *pb.Reply), opts ...grpc.CallOption) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
stream, err := client.TTSStream(ctx, in, opts...)
if err != nil {
return err
}
for {
// Check if context is cancelled before receiving
select {
case <-ctx.Done():
return ctx.Err()
default:
}
reply, err := stream.Recv()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
// Check if error is due to context cancellation
if ctx.Err() != nil {
return ctx.Err()
}
return err
}
f(reply)
}
return nil
}
func (c *Client) SoundGeneration(ctx context.Context, in *pb.SoundGenerationRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.SoundGeneration(ctx, in, opts...)
}
func (c *Client) AudioTranscription(ctx context.Context, in *pb.TranscriptRequest, opts ...grpc.CallOption) (*pb.TranscriptResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.AudioTranscription(ctx, in, opts...)
}
func (c *Client) AudioTranscriptionStream(ctx context.Context, in *pb.TranscriptRequest, f func(chunk *pb.TranscriptStreamResponse), opts ...grpc.CallOption) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
stream, err := client.AudioTranscriptionStream(ctx, in, opts...)
if err != nil {
return err
}
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
chunk, err := stream.Recv()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
if ctx.Err() != nil {
return ctx.Err()
}
return err
}
f(chunk)
}
return nil
}
func (c *Client) TokenizeString(ctx context.Context, in *pb.PredictOptions, opts ...grpc.CallOption) (*pb.TokenizationResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
res, err := client.TokenizeString(ctx, in, opts...)
if err != nil {
return nil, err
}
return res, nil
}
func (c *Client) Status(ctx context.Context) (*pb.StatusResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.Status(ctx, &pb.HealthMessage{})
}
func (c *Client) StoresSet(ctx context.Context, in *pb.StoresSetOptions, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StoresSet(ctx, in, opts...)
}
func (c *Client) StoresDelete(ctx context.Context, in *pb.StoresDeleteOptions, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.wdMark()
defer c.wdUnMark()
c.setBusy(true)
defer c.setBusy(false)
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StoresDelete(ctx, in, opts...)
}
func (c *Client) StoresGet(ctx context.Context, in *pb.StoresGetOptions, opts ...grpc.CallOption) (*pb.StoresGetResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StoresGet(ctx, in, opts...)
}
func (c *Client) StoresFind(ctx context.Context, in *pb.StoresFindOptions, opts ...grpc.CallOption) (*pb.StoresFindResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StoresFind(ctx, in, opts...)
}
func (c *Client) Rerank(ctx context.Context, in *pb.RerankRequest, opts ...grpc.CallOption) (*pb.RerankResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.Rerank(ctx, in, opts...)
}
func (c *Client) TokenClassify(ctx context.Context, in *pb.TokenClassifyRequest, opts ...grpc.CallOption) (*pb.TokenClassifyResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.TokenClassify(ctx, in, opts...)
}
func (c *Client) Score(ctx context.Context, in *pb.ScoreRequest, opts ...grpc.CallOption) (*pb.ScoreResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.Score(ctx, in, opts...)
}
func (c *Client) GetTokenMetrics(ctx context.Context, in *pb.MetricsRequest, opts ...grpc.CallOption) (*pb.MetricsResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.GetMetrics(ctx, in, opts...)
}
func (c *Client) VAD(ctx context.Context, in *pb.VADRequest, opts ...grpc.CallOption) (*pb.VADResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.VAD(ctx, in, opts...)
}
func (c *Client) Diarize(ctx context.Context, in *pb.DiarizeRequest, opts ...grpc.CallOption) (*pb.DiarizeResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.Diarize(ctx, in, opts...)
}
func (c *Client) SoundDetection(ctx context.Context, in *pb.SoundDetectionRequest, opts ...grpc.CallOption) (*pb.SoundDetectionResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.SoundDetection(ctx, in, opts...)
}
func (c *Client) Detect(ctx context.Context, in *pb.DetectOptions, opts ...grpc.CallOption) (*pb.DetectResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.Detect(ctx, in, opts...)
}
func (c *Client) Depth(ctx context.Context, in *pb.DepthRequest, opts ...grpc.CallOption) (*pb.DepthResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.Depth(ctx, in, opts...)
}
func (c *Client) FaceVerify(ctx context.Context, in *pb.FaceVerifyRequest, opts ...grpc.CallOption) (*pb.FaceVerifyResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.FaceVerify(ctx, in, opts...)
}
func (c *Client) FaceAnalyze(ctx context.Context, in *pb.FaceAnalyzeRequest, opts ...grpc.CallOption) (*pb.FaceAnalyzeResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.FaceAnalyze(ctx, in, opts...)
}
func (c *Client) VoiceVerify(ctx context.Context, in *pb.VoiceVerifyRequest, opts ...grpc.CallOption) (*pb.VoiceVerifyResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.VoiceVerify(ctx, in, opts...)
}
func (c *Client) VoiceAnalyze(ctx context.Context, in *pb.VoiceAnalyzeRequest, opts ...grpc.CallOption) (*pb.VoiceAnalyzeResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.VoiceAnalyze(ctx, in, opts...)
}
func (c *Client) VoiceEmbed(ctx context.Context, in *pb.VoiceEmbedRequest, opts ...grpc.CallOption) (*pb.VoiceEmbedResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.VoiceEmbed(ctx, in, opts...)
}
func (c *Client) AudioEncode(ctx context.Context, in *pb.AudioEncodeRequest, opts ...grpc.CallOption) (*pb.AudioEncodeResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.AudioEncode(ctx, in, opts...)
}
func (c *Client) AudioDecode(ctx context.Context, in *pb.AudioDecodeRequest, opts ...grpc.CallOption) (*pb.AudioDecodeResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.AudioDecode(ctx, in, opts...)
}
func (c *Client) AudioTransform(ctx context.Context, in *pb.AudioTransformRequest, opts ...grpc.CallOption) (*pb.AudioTransformResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer func() { _ = conn.Close() }()
client := pb.NewBackendClient(conn)
return client.AudioTransform(ctx, in, opts...)
}
// ForwardClient is the duplex interface returned by (*Client).Forward.
// First Send carries path/method/headers/body, subsequent Sends carry
// body_chunk only. First Recv carries status/headers, subsequent Recvs
// carry body_chunk. Caller closes via CloseSend when request is done;
// stream ends when the upstream finishes and the server closes.
type ForwardClient interface {
Send(*pb.ForwardRequest) error
Recv() (*pb.ForwardReply, error)
CloseSend() error
Context() context.Context
}
type forwardClient struct {
pb.Backend_ForwardClient
conn *grpc.ClientConn
closer func()
once sync.Once
}
// CloseSend signals end-of-requests to the server but keeps the
// underlying connection open so the server can still send replies.
// Connection cleanup happens when Recv returns a final error (EOF
// or any other terminal status).
func (s *forwardClient) CloseSend() error {
return s.Backend_ForwardClient.CloseSend()
}
// Recv wraps the embedded stream's Recv to fire the connection-level
// closer once the stream ends. On EOF or any other error the
// connection + operation-state cleanup runs exactly once.
func (s *forwardClient) Recv() (*pb.ForwardReply, error) {
reply, err := s.Backend_ForwardClient.Recv()
if err != nil && s.closer != nil {
s.once.Do(s.closer)
}
return reply, err
}
func (c *Client) Forward(ctx context.Context, opts ...grpc.CallOption) (ForwardClient, error) {
if !c.parallel {
c.opMutex.Lock()
}
c.setBusy(true)
c.wdMark()
cleanup := func() {
c.wdUnMark()
c.setBusy(false)
if !c.parallel {
c.opMutex.Unlock()
}
}
conn, err := c.dial()
if err != nil {
cleanup()
return nil, err
}
client := pb.NewBackendClient(conn)
stream, err := client.Forward(ctx, opts...)
if err != nil {
_ = conn.Close()
cleanup()
return nil, err
}
return &forwardClient{
Backend_ForwardClient: stream,
conn: conn,
closer: func() {
_ = conn.Close()
cleanup()
},
}, nil
}
// AudioTransformStreamClient is the duplex interface returned by
// (*Client).AudioTransformStream. Wraps the generated bidi client without
// leaking the proto package across the public boundary.
type AudioTransformStreamClient interface {
Send(*pb.AudioTransformFrameRequest) error
Recv() (*pb.AudioTransformFrameResponse, error)
CloseSend() error
Context() context.Context
}
// audioTransformStreamClient is the concrete wrapper. It also owns the
// underlying gRPC connection, released once the receive side terminates —
// NOT at CloseSend, because the server still streams responses (the tail of
// the transform) after the client closes its send side. Same lifecycle as
// forwardClient.
type audioTransformStreamClient struct {
pb.Backend_AudioTransformStreamClient
closeOnce sync.Once
closer func()
}
func (s *audioTransformStreamClient) Recv() (*pb.AudioTransformFrameResponse, error) {
resp, err := s.Backend_AudioTransformStreamClient.Recv()
if err != nil && s.closer != nil {
s.closeOnce.Do(s.closer)
}
return resp, err
}
func (c *Client) AudioTransformStream(ctx context.Context, opts ...grpc.CallOption) (AudioTransformStreamClient, error) {
if !c.parallel {
c.opMutex.Lock()
}
c.setBusy(true)
c.wdMark()
cleanup := func() {
c.wdUnMark()
c.setBusy(false)
if !c.parallel {
c.opMutex.Unlock()
}
}
conn, err := c.dial()
if err != nil {
cleanup()
return nil, err
}
client := pb.NewBackendClient(conn)
stream, err := client.AudioTransformStream(ctx, opts...)
if err != nil {
_ = conn.Close()
cleanup()
return nil, err
}
return &audioTransformStreamClient{
Backend_AudioTransformStreamClient: stream,
closer: func() {
_ = conn.Close()
cleanup()
},
}, nil
}
// AudioTranscriptionLiveClient is the duplex interface returned by
// (*Client).AudioTranscriptionLive. Wraps the generated bidi client without
// leaking the proto package across the public boundary.
type AudioTranscriptionLiveClient interface {
Send(*pb.TranscriptLiveRequest) error
Recv() (*pb.TranscriptLiveResponse, error)
CloseSend() error
Context() context.Context
}
type audioTranscriptionLiveClient struct {
pb.Backend_AudioTranscriptionLiveClient
closeOnce sync.Once
closer func()
}
// Recv releases the connection once the stream reaches a terminal state
// (io.EOF after the server finishes, or any error). The conn MUST survive
// CloseSend: the live protocol is close-send -> backend flushes the decode
// tail -> terminal FinalResult arrives. Closing the conn inside CloseSend
// killed that pending Recv with "grpc: the client connection is closing",
// losing the final transcript (and its tail words) on every turn.
func (s *audioTranscriptionLiveClient) Recv() (*pb.TranscriptLiveResponse, error) {
resp, err := s.Backend_AudioTranscriptionLiveClient.Recv()
if err != nil {
s.release()
}
return resp, err
}
func (s *audioTranscriptionLiveClient) release() {
s.closeOnce.Do(func() {
if s.closer != nil {
s.closer()
}
})
}
// AudioTranscriptionLive opens the bidirectional live ASR stream. Note the
// same caveat as AudioToAudioStream: the watchdog busy-mark (and, on
// non-parallel backends, opMutex) is held for the stream's lifetime, which
// for a realtime session can be minutes — enable parallel requests on
// backends meant to serve live sessions alongside unary work.
func (c *Client) AudioTranscriptionLive(ctx context.Context, opts ...grpc.CallOption) (AudioTranscriptionLiveClient, error) {
if !c.parallel {
c.opMutex.Lock()
}
c.setBusy(true)
c.wdMark()
cleanup := func() {
c.wdUnMark()
c.setBusy(false)
if !c.parallel {
c.opMutex.Unlock()
}
}
conn, err := c.dial()
if err != nil {
cleanup()
return nil, err
}
client := pb.NewBackendClient(conn)
stream, err := client.AudioTranscriptionLive(ctx, opts...)
if err != nil {
_ = conn.Close()
cleanup()
return nil, err
}
return &audioTranscriptionLiveClient{
Backend_AudioTranscriptionLiveClient: stream,
closer: func() {
_ = conn.Close()
cleanup()
},
}, nil
}
// AudioToAudioStreamClient is the duplex interface returned by
// (*Client).AudioToAudioStream. Mirrors AudioTransformStreamClient's
// shape so realtime-API callers can plug in interchangeable backends.
type AudioToAudioStreamClient interface {
Send(*pb.AudioToAudioRequest) error
Recv() (*pb.AudioToAudioResponse, error)
CloseSend() error
Context() context.Context
}
// audioToAudioStreamClient owns its gRPC connection, released once the
// receive side terminates — NOT at CloseSend, because the server still
// streams the response tail after the client closes its send side. Same
// lifecycle as forwardClient.
type audioToAudioStreamClient struct {
pb.Backend_AudioToAudioStreamClient
closeOnce sync.Once
closer func()
}
func (s *audioToAudioStreamClient) Recv() (*pb.AudioToAudioResponse, error) {
resp, err := s.Backend_AudioToAudioStreamClient.Recv()
if err != nil && s.closer != nil {
s.closeOnce.Do(s.closer)
}
return resp, err
}
func (c *Client) AudioToAudioStream(ctx context.Context, opts ...grpc.CallOption) (AudioToAudioStreamClient, error) {
if !c.parallel {
c.opMutex.Lock()
}
c.setBusy(true)
c.wdMark()
cleanup := func() {
c.wdUnMark()
c.setBusy(false)
if !c.parallel {
c.opMutex.Unlock()
}
}
conn, err := c.dial()
if err != nil {
cleanup()
return nil, err
}
client := pb.NewBackendClient(conn)
stream, err := client.AudioToAudioStream(ctx, opts...)
if err != nil {
_ = conn.Close()
cleanup()
return nil, err
}
return &audioToAudioStreamClient{
Backend_AudioToAudioStreamClient: stream,
closer: func() {
_ = conn.Close()
cleanup()
},
}, nil
}
func (c *Client) StartFineTune(ctx context.Context, in *pb.FineTuneRequest, opts ...grpc.CallOption) (*pb.FineTuneJobResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StartFineTune(ctx, in, opts...)
}
func (c *Client) FineTuneProgress(ctx context.Context, in *pb.FineTuneProgressRequest, f func(update *pb.FineTuneProgressUpdate), opts ...grpc.CallOption) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
stream, err := client.FineTuneProgress(ctx, in, opts...)
if err != nil {
return err
}
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
update, err := stream.Recv()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
if ctx.Err() != nil {
return ctx.Err()
}
return err
}
f(update)
}
return nil
}
func (c *Client) StopFineTune(ctx context.Context, in *pb.FineTuneStopRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StopFineTune(ctx, in, opts...)
}
func (c *Client) ListCheckpoints(ctx context.Context, in *pb.ListCheckpointsRequest, opts ...grpc.CallOption) (*pb.ListCheckpointsResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.ListCheckpoints(ctx, in, opts...)
}
func (c *Client) ExportModel(ctx context.Context, in *pb.ExportModelRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.ExportModel(ctx, in, opts...)
}
func (c *Client) StartQuantization(ctx context.Context, in *pb.QuantizationRequest, opts ...grpc.CallOption) (*pb.QuantizationJobResult, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StartQuantization(ctx, in, opts...)
}
func (c *Client) QuantizationProgress(ctx context.Context, in *pb.QuantizationProgressRequest, f func(update *pb.QuantizationProgressUpdate), opts ...grpc.CallOption) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
stream, err := client.QuantizationProgress(ctx, in, opts...)
if err != nil {
return err
}
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
update, err := stream.Recv()
if errors.Is(err, io.EOF) {
break
}
if err != nil {
if ctx.Err() != nil {
return ctx.Err()
}
return err
}
f(update)
}
return nil
}
func (c *Client) StopQuantization(ctx context.Context, in *pb.QuantizationStopRequest, opts ...grpc.CallOption) (*pb.Result, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.StopQuantization(ctx, in, opts...)
}
func (c *Client) Free(ctx context.Context) error {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
_, err = client.Free(ctx, &pb.HealthMessage{})
return err
}
func (c *Client) ModelMetadata(ctx context.Context, in *pb.ModelOptions, opts ...grpc.CallOption) (*pb.ModelMetadataResponse, error) {
if !c.parallel {
c.opMutex.Lock()
defer c.opMutex.Unlock()
}
c.setBusy(true)
defer c.setBusy(false)
c.wdMark()
defer c.wdUnMark()
conn, err := c.dial()
if err != nil {
return nil, err
}
defer conn.Close()
client := pb.NewBackendClient(conn)
return client.ModelMetadata(ctx, in, opts...)
}