Checkpoint speaker transcriptions incrementally

This commit is contained in:
2026-09-21 16:20:53 +00:00
parent 26aa0cce29
commit 2ae63dbc2b
8 changed files with 793 additions and 70 deletions

View File

@@ -8,7 +8,6 @@ import (
"path/filepath"
"sort"
"strings"
"sync"
"golang.org/x/sync/errgroup"
@@ -28,11 +27,14 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
if env.ArtifactStore == nil {
return nil, fmt.Errorf("transcribe: artifact store is required")
}
if env.ManifestStore == nil {
return nil, fmt.Errorf("transcribe: manifest store is required")
}
if env.Config.Pipeline == nil || env.Config.Session == nil {
return nil, fmt.Errorf("transcribe: resolved config must include pipeline and session")
}
if env.WhisperX == nil {
return nil, fmt.Errorf("transcribe: whisperx adapter is required")
if m == nil {
return nil, fmt.Errorf("transcribe: session manifest is required")
}
var sessionID string
@@ -47,21 +49,29 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
}
paths := sessionPathsForEnv(env, sessionID)
runLayout, err := resolveRunStageLayout(env, m, paths, sessionID, "transcribe")
reconciliation, err := reconcileTranscribeSpeakers(ctx, env, m)
if err != nil {
return nil, fmt.Errorf("transcribe: resolve run-stage layout: %w", err)
return nil, fmt.Errorf("transcribe: reconcile speakers: %w", err)
}
jobs, err := buildTranscribeJobs(m, paths)
if err != nil {
return nil, fmt.Errorf("transcribe: resolve audio inputs: %w", err)
state := newTranscribeExecutionState(env, m, paths, reconciliation.current)
if err := state.checkpointStartingCollection(ctx); err != nil {
return nil, fmt.Errorf("transcribe: checkpoint reconciled speaker collection before execution: %w", err)
}
if len(jobs) == 0 {
return nil, fmt.Errorf("transcribe: no prepared audio files found")
jobs := append([]transcribeReconciliationJob(nil), reconciliation.work...)
if len(jobs) > 0 && env.WhisperX == nil {
return nil, fmt.Errorf("transcribe: whisperx adapter is required")
}
for index := range jobs {
jobs[index].runOutputPath, err = runLocalPathForCanonical(runLayout, paths, jobs[index].canonicalOutputPath)
if len(jobs) > 0 {
runLayout, err := resolveRunStageLayout(env, m, paths, sessionID, "transcribe")
if err != nil {
return nil, fmt.Errorf("transcribe: resolve run-local output path for %q: %w", jobs[index].speakerID, err)
return nil, fmt.Errorf("transcribe: resolve run-stage layout: %w", err)
}
for index := range jobs {
jobs[index].runOutputPath, err = runLocalPathForCanonical(runLayout, paths, jobs[index].canonicalOutputPath)
if err != nil {
return nil, fmt.Errorf("transcribe: resolve run-local output path for %q: %w", jobs[index].speakerID, err)
}
}
}
@@ -69,15 +79,10 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
if env.Config.Pipeline.WhisperX.Concurrency != nil && *env.Config.Pipeline.WhisperX.Concurrency > 0 {
concurrency = *env.Config.Pipeline.WhisperX.Concurrency
}
if concurrency > len(jobs) {
if len(jobs) > 0 && concurrency > len(jobs) {
concurrency = len(jobs)
}
var (
mu sync.Mutex
completed int
perFile = map[string]map[string]any{}
)
group, groupCtx := errgroup.WithContext(ctx)
group.SetLimit(concurrency)
for _, job := range jobs {
@@ -86,6 +91,9 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
if err := groupCtx.Err(); err != nil {
return err
}
if err := state.checkpointFailure(); err != nil {
return err
}
res, runErr := env.WhisperX.Transcribe(groupCtx, whisperx.TranscribeRequest{
SpeakerID: job.speakerID,
AudioPath: job.audioPath,
@@ -95,13 +103,6 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
return fmt.Errorf("speaker %q (%s): %w", job.speakerID, filepath.Base(job.audioPath), runErr)
}
if _, err := authoritativeOutputPath(job.runOutputPath, res.OutputRawTranscriptPath); err != nil {
return fmt.Errorf("speaker %q: %w", job.speakerID, err)
}
if err := validateTranscriptJSONFile(job.runOutputPath); err != nil {
return fmt.Errorf("speaker %q output %q invalid: %w", job.speakerID, job.runOutputPath, err)
}
meta := map[string]any{
"attempts": res.Attempts,
"http_status": res.HTTPStatus,
@@ -109,17 +110,17 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
"output_path": job.runOutputPath,
}
mu.Lock()
perFile[job.speakerID] = meta
completed++
mu.Unlock()
if err := state.acceptSpeaker(ctx, job, res.OutputRawTranscriptPath, meta); err != nil {
return fmt.Errorf("speaker %q: %w", job.speakerID, err)
}
return nil
})
}
groupErr := group.Wait()
mu.Lock()
completedCount := completed
mu.Unlock()
if err := state.checkpointFailure(); err != nil {
return nil, err
}
completedCount := state.completedCount()
if err := ctx.Err(); err != nil {
return nil, fmt.Errorf("transcribe: canceled after completing %d of %d audio jobs: %w", completedCount, len(jobs), err)
}
@@ -129,34 +130,9 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
if completedCount != len(jobs) {
return nil, fmt.Errorf("transcribe: completed %d of %d planned audio jobs", completedCount, len(jobs))
}
speakers := make([]string, 0, len(perFile))
for speaker := range perFile {
speakers = append(speakers, speaker)
}
sort.Strings(speakers)
outputs := make([]artifacts.Ref, 0, len(speakers))
runOutputPaths := make([]string, 0, len(speakers))
outputPaths := make([]string, 0, len(speakers))
orderedPerFile := make(map[string]any, len(speakers))
for _, speaker := range speakers {
var selected transcribeJob
for _, job := range jobs {
if job.speakerID == speaker {
selected = job
break
}
}
runOutputPaths = append(runOutputPaths, selected.runOutputPath)
ref := artifacts.Ref{Kind: "transcript_raw", Category: "transcripts", SessionID: sessionID, AbsolutePath: selected.runOutputPath}
materialized, err := materializeRunLocalOutput(env.ArtifactStore, selected.runOutputPath, selected.canonicalOutputPath, ref)
if err != nil {
return nil, fmt.Errorf("transcribe: materialize %q output: %w", speaker, err)
}
outputs = append(outputs, materialized)
outputPaths = append(outputPaths, selected.canonicalOutputPath)
orderedPerFile[speaker] = perFile[speaker]
outputs, outputPaths, runOutputPaths, orderedPerFile, err := state.successDetails()
if err != nil {
return nil, fmt.Errorf("transcribe: assemble completed outputs: %w", err)
}
retries := 0
@@ -168,7 +144,7 @@ func (transcribeStage) Run(ctx context.Context, env *Env, m *manifest.Manifest)
Outputs: outputs,
Metadata: map[string]any{
"stage": "transcribe",
"audio_files_count": len(jobs),
"audio_files_count": len(reconciliation.jobs),
"language": env.Config.Pipeline.WhisperX.Language,
"concurrency": concurrency,
"retries": retries,