Batch production & throughput
SOW §3.5 — parallel, high-volume processing with performance metrics. NLS produces ~3,000 titles a year (≈12 per business day).
Workers
—
parallel titles (Fargate autoscaling in prod)
Queued / running
0 / 0
Completed
—
Audio produced
—
Avg wall-clock / title
—
Titles / hour
—
observed, this session
| Title | Source | Words | Audio | Synth | Master | Wall-clock | × realtime | QC |
|---|
Synthesis is cached per narration unit and voice, so re-runs after structure edits only synthesize what changed. Numbers here are from this machine’s CPU with the open-source Piper voice; Amazon Polly runs the synthesis stage in the Library’s account and is bounded by API throughput rather than local compute.