Features & formats

Every request can enable the full feature set — there are no paid add-ons.

Transcription options

Option	Default	Description
`word_timestamps`	`true`	Start/end time per word (seconds)
`speaker_labels`	`true`	Diarization / speaker turns
`diarize`	—	Deepgram-compatible alias for `speaker_labels`
`nltk`	`true`	Restore punctuation & capitalization
`custom_vocabulary`	—	List of names / jargon to recover
`tier`	`standard`	`standard` — the only tier currently available
`output_type`	`json`	Result file format

Output types

Type	Use for
`json`	Structured `words` (+ optional diarization). SDKs parse this into `text`, `words`, `utterances`.
`txt`	Plain transcript (speaker blocks when labels on)
`srt` / `vtt`	Subtitles / captions
`docx` / `pdf`	Shareable documents

JSON shape (native)

Default JSON looks like:

{
  "words": [
    {"word": "Hello,", "start": 0.12, "end": 0.40, "speaker": "SPEAKER_0", "language": "en"},
    {"word": "world.", "start": 0.41, "end": 0.70, "speaker": "SPEAKER_0", "language": "en"}
  ],
  "diarization": [
    {"start": 0.12, "end": 0.70, "speaker": "SPEAKER_0"}
  ],
  "languages": [
    {"start": 0.12, "end": 0.70, "language": "en"}
  ]
}

Live progress

Every SDK reports real-time progress for both the file upload and the transcription — as console bars (progress toggle) or as onProgress / onUploadProgress callbacks that carry completed, total, step, and a computed percent. It is derived from the job SSE stream's completed/total counts, and neither AssemblyAI nor Deepgram exposes it for pre-recorded audio.

SDK adapters

Python/JS SDKs expose AssemblyAI-style result.text / result.utterances, plus result.to_deepgram() for Deepgram-shaped migrations.