# Split into stems

`POST https://api.eesi.ai/v1/speech/transcriptions/{session_id}/stems`

Split the audio into stems, in the background:

- `kinds: ["dme"]`: dialogue, music and effects, which add back up to the
  original.
- `speakers: ["S01"]`: one person's voice, isolated from everything else, using
  the transcript's speaker turns.
- `prompt: "piano"`: any sound you describe.

Stems are 48 kHz FLAC. Poll `GET /v1/speech/transcriptions/{session_id}/stems`
until each is `complete`. Asking for a stem again replaces the earlier one.

Parameters:

- `session_id` (path, required)

Request body: application/json.

- `kinds` (array of string): `["dme"]` for dialogue, music and effects (or any of `dialogue`, `music`, `effects`).
- `speakers` (array of string): Speaker labels, such as `S01`, whose voice to isolate.
- `prompt` (string): Any sound to isolate, described in words: "piano", "the crowd".

Responses: 202, 404, 422.

Schemas: https://docs.eesi.ai/api-reference/public.json