[kvz]

Why

/speech/transcribe can label who is speaking, but transloadit speech transcribe had no flag for it, and its replicate default would make the API reject speaker labels anyway. The speaker labels announcement post needs this to be one command. Closes #643.

Changes

• --speaker-labels sends speaker_labels: true. Without --provider, the CLI leaves provider out so the API picks one that supports labels (aws, or gcp with vocabulary). The replicate default still applies without the flag.
• --max-speakers <1–10> sends max_speakers.
• --speaker-labels --provider replicate, --max-speakers without --speaker-labels, and out-of-range values fail before an Assembly is created.
• New help example: transloadit speech transcribe --input interview.mp4 --speaker-labels --output interview.txt.
• sync:intent-docs regenerated the README and docs/intent-commands.md. Besides the new rows, it picked up option descriptions that earlier Robot schema syncs had left stale.

Text, SRT and WebVTT output starts showing the speakers once transloadit/api2#10812 is deployed. Until then, use --format json.

Verification

yarn check in packages/node: intent docs sync, tsc, Biome, and 1129 unit tests pass, including three new intent tests (provider left to the API, replicate rejected, --max-speakers without labels rejected).

_{🤖 Claimed by kvz on framework1/node-sdk since 2026-11-05 · resume: claude --resume e0b6ec91-9333-451d-94f2-93dea1b12f72}