Kick off an asynchronous stuttering classification job on a previously-uploaded recording. Returns a session_id immed…

REQUIRED. fileId from a prior /api/assignFileId upload.

Optional, default "unknown". Patient identifier — the literal value "PT-UNKNOWN" is explicitly rejected (use "unknown" or a real ID).

Optional, default 0.25. Overlap fraction between analysis chunks.

Optional, default 4. Chunk size in seconds.

Optional, default "en-US". Locale used by the speech-to-text step within the classification pipeline.

optional (default true). When true, the recording is transcribed once (real-time STT) and each result block includes its transcription plus a words array of per-word timestamps. Set false to skip transcription for minimum latency.

optional (default false). When true, per-block phonetic transcription is computed (slower). When false, each block's phonemes is "N/A".

Boolean — true when the job has been queued.

String of the form "cls-<uuid>". Pass to /api/therapy-status/:sessionId and /api/therapy-result/:sessionId to poll and retrieve.

Human-readable confirmation, e.g. "Classification started. Use session_id to poll for progress."