Get transcription

GET /transcriptions/{id} — one transcription: the flat text, the structured segments, speakers, topics and AI reports.

GET /transcriptions/{id}

Returns one transcription in full: the transcript both as a flat text and as structured segments, plus any AI reports (prompt_results). No query parameters — everything comes back every time. Find the id via List transcriptions.

Path parameters

Parameter	Type	Description
`id`	UUID	Transcription identifier.

Request

curl https://app.specala.ai/api/v1/developer/transcriptions/3fa85f64-5717-4562-b3fc-2c963f66afa6 \
  -H "Authorization: Bearer sk_live_your_key_here"

import httpx

tid = "3fa85f64-5717-4562-b3fc-2c963f66afa6"
resp = httpx.get(
    f"https://app.specala.ai/api/v1/developer/transcriptions/{tid}",
    headers={"Authorization": "Bearer sk_live_your_key_here"},
)
resp.raise_for_status()
t = resp.json()

const tid = "3fa85f64-5717-4562-b3fc-2c963f66afa6";
const res = await fetch(
  `https://app.specala.ai/api/v1/developer/transcriptions/${tid}`,
  { headers: { Authorization: "Bearer sk_live_your_key_here" } },
);
const t = await res.json();

Open the Acme discovery call and show me what was decided.

The assistant calls specala_get_transcription with the UUID.

Response

200 OK. The transcript comes back two ways in the same payload: text is the flat string to show a person, segments is the same speech as structured data to load into a CRM, a database or your own model. You do not have to ask for either.

{
  "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "title": "Acme — discovery call",
  "status": "completed",
  "summary": "Walked through current workflow and pain points...",
  "language": "en",
  "duration_seconds": 1840.5,
  "media_type": "audio",
  "project": { "id": "8b1f3e22-1c4a-4f7e-9a2b-6d5e8c1a2b3c", "name": "Sales" },
  "author": { "id": "9c2d4e6f-1a2b-4c3d-8e9f-0a1b2c3d4e5f", "email": "alex@company.com", "name": "Alex Morgan" },
  "speakers": [{ "id": "1", "name": "Alex" }, { "id": "2", "name": "Jordan" }],
  "topics": [{ "title": "Pricing" }, { "title": "Onboarding" }],

  "text": "[00:00] Alex: Thanks for hopping on...\n[00:42] Jordan: Of course...",
  "segments": [
    {
      "id": 1,
      "start": 0.0,
      "end": 4.82,
      "speaker_id": "1",
      "speaker_name": "Alex",
      "text": "Thanks for hopping on..."
    },
    {
      "id": 2,
      "start": 42.1,
      "end": 45.6,
      "speaker_id": "2",
      "speaker_name": "Jordan",
      "text": "Of course..."
    }
  ],

  "prompt_results": [
    {
      "id": "8d2f6a1e-0c4b-4f0e-9d3a-6b1c2e7f5a90",
      "slug": "followup-email",
      "name": "Follow-up plan",
      "status": "completed",
      "error_code": null,
      "text": "- Send pricing deck by Friday\n- Schedule technical follow-up",
      "created_at": "2026-05-20T09:31:00Z",
      "completed_at": "2026-05-20T09:31:40Z"
    }
  ],
  "created_at": "2026-05-20T09:14:00Z",
  "updated_at": "2026-05-20T09:31:00Z"
}

The transcript

Field	Type	Description
`text`	string \| null	The whole transcript as one string, with timestamps and speaker labels. `null` until `status` is `completed`.
`segments[]`	array	The same speech, one entry per turn. Empty until `status` is `completed`.
`segments[].id`	integer	Position of the turn, starting at 1. Stable for a given transcript.
`segments[].start` / `.end`	float \| null	Seconds from the start of the recording.
`segments[].speaker_id`	string \| null	Diarization identifier, matching `speakers[].id`. Survives renames — group by this. `null` when the recording was not diarized.
`segments[].speaker_name`	string \| null	The name shown today, with manual renames applied over the AI-assigned one. `null` when nobody has named this speaker — label them yourself from `speaker_id`, in your own language.
`segments[].text`	string	What was said, without the timestamp or speaker prefix.

Build on segments. text is the same speech assembled for a person to read: its exact shape is not something we promise to keep, and the speaker labels inside it change the moment someone renames a speaker in the app.

Everything else

Same as a list item, plus:

Field	Type	Description
`status`	string	`queued`, `processing`, `completed`, `failed` or `insufficient_balance` — see status lifecycle.
`error_code`	string \| null	Machine-readable failure reason when `status` is `failed`.
`required_minutes`	float \| null	How many minutes this file needs. Set only when `status` is `insufficient_balance`.
`available_minutes`	float \| null	How many minutes the workspace has left. Set only when `status` is `insufficient_balance` — the difference from `required_minutes` is exactly what needs topping up.
`speakers[]`	array	Everyone detected in the recording, as `{ id, name }`.
`external_id`	string \| null	The identifier you sent as `X-External-Id` at upload. `null` if the transcription did not come through the API. See Your own identifiers.
`metadata`	object \| null	The object you sent as `X-Metadata` at upload, returned as given.
`author`	object	The workspace member who owns the transcription, as `{ id, email, name }`. For API uploads this is the `X-Author` you sent, or the key's creator. `id` is a UUID and can be used as `X-Author: user:<id>`. See Assigning an author.
`prompt_results[]`	array	AI reports for this transcription — generated in the app, with `ai_metadata=true`, or requested through the API. Reports requested with `?reports=` appear here immediately with their own `status`; the transcription itself is `completed` as soon as the text is ready, so keep polling until every report you need is terminal. See AI reports.
`prompt_results[].id`	UUID	Report identifier.
`prompt_results[].slug`	string \| null	Slug of the report template that produced it. `null` for legacy reports without a template.
`prompt_results[].name`	string \| null	Report name, e.g. Follow-up plan, Summary.
`prompt_results[].status`	string	`queued` → `processing` → `completed` / `failed`.
`prompt_results[].error_code`	string \| null	Set only on `failed`.
`prompt_results[].text`	string \| null	Report content, as prose. Present only when `status` is `completed`.
`prompt_results[].created_at`	datetime	When the report was requested.
`prompt_results[].completed_at`	datetime \| null	When the report was generated.

Need the whole thing as a file instead? Export transcription with format=json returns exactly this body.

Errors

Status	When
`401`	Missing/invalid key, or key passed in the URL.
`403`	Plan without API access.
`403`	`transcription_locked` — the transcription is a locked preview. Should not happen on a plan with API access; if it does, the recording is not readable through the API.
`404`	No transcription with that `id` in this workspace.
`429`	Rate limit exceeded.

A 404 is also returned when the transcription exists but belongs to another workspace — existence is never revealed across workspaces. See Errors.