Sync Transcribe

Send an audio file and receive the transcript in the same HTTP request/response cycle — no webhooks or polling required. Optionally, the endpoint can also label who spoke when (diarisation) and detect music such as hold tones.

Code
POST https://api.lelapa.ai/v1/transcribe/sync

Request

Headers

Form data

Supported audio formats: WAV, MP3, FLAC, AAC, OGG, MP4, AIFF, OPUS.

Query parameters

Supported language codes

Note: our zul (isiZulu) support is code-switched isiZulu.

See Language Support for the current status of every language, and the model cards for details on model performance.

Example request

Code
import requests

API_KEY = "<INSERT-TOKEN>"

with open("recording.wav", "rb") as f:
    response = requests.post(
        "https://api.lelapa.ai/v1/transcribe/sync",
        headers={"X-CLIENT-TOKEN": API_KEY},
        files={"file": ("recording.wav", f, "audio/wav")},
        params={
            "lang_code": "<INSERT-LANGUAGE-CODE>",  # optional
            "diarise": 1,                           # optional
        },
    )

print(response.json()["transcription_text"])

Prefer not to hand-roll HTTP calls? The Python SDK wraps transcription in a single method call.

Response

A successful request returns 200 OK with the transcript and audio metadata. The example below shows a diarised, two-speaker call with detect_music enabled:

Code
{
  "id": "5f15e81b-53c2-4c5c-a779-1f6776100543",
  "upload_file_size": null,
  "audio_length_seconds": 20.0,
  "sample_rate": null,
  "channels": null,
  "frame_rate": null,
  "mime_type": null,
  "language_code": "eng",
  "warnings": null,
  "diarisation_result": {
    "timeline": [
      {
        "start_time": 0.0,
        "end_time": 2.14,
        "type": "silence"
      },
      {
        "start_time": 2.14,
        "end_time": 10.54,
        "type": "speech",
        "speaker_id": "speaker0",
        "text": "Thank you so much for calling. You're through to Thandi, how can help you?"
      },
      {
        "start_time": 10.54,
        "end_time": 15.98,
        "type": "speech",
        "speaker_id": "speaker1",
        "text": "Hi, Thandi, I would like to settle my bill for yesterday."
      },
      {
        "start_time": 15.98,
        "end_time": 20.0,
        "type": "music"
      }
    ],
    "words": [
      {
        "word": "Thank",
        "start_time": 2.14,
        "end_time": 2.45,
        "confidence": 0.94,
        "weight": null,
        "word_intensity": null,
        "best_path": true,
        "speaker_id": "speaker0"
      },
      {
        "word": "you",
        "start_time": 2.45,
        "end_time": 2.60,
        "confidence": 0.91,
        "weight": null,
        "word_intensity": null,
        "best_path": true,
        "speaker_id": "speaker0"
      }
      // More words would follow here...
    ]
  },
  "transcription_text": "Thank you so much for calling. You're through to Thandi, how can help you? Hi, Thandi, I would like to settle my bill for yesterday.",
  "transcription_status": "COMPLETED",
  "error_message": null,
  "status_datetime": "2025-04-16T10:45:00Z",
  "upload_datetime": "2025-04-16T10:43:12Z"
}

Response fields

Word-level results

You still get who spoke when, and the text of each turn, from timeline. What is not available for Sesotho:

• Word timings — the start and end time of each word. Captions, search and click-to-play can only work at the level of a whole speaker turn.
• Per-word speaker IDs — speaker labels apply to a whole turn, and the text at the edge of a turn may be attached to the neighbouring speaker.
• Per-word confidence — there is no score to filter or flag individual words with.

If your integration handles several languages, treat words as optional:

Code
result = response.json()["diarisation_result"]
words = result.get("words") or []   # empty for Sesotho

for segment in result["timeline"]:
    if segment["type"] == "speech":
        print(segment["speaker_id"], segment["text"])

Errors

Next steps

• Use the Python SDK for webhook-based transcription without managing HTTP calls
• Explore the model cards for accuracy details per language
• Translate your transcripts with the Translate API
• Browse the full API Reference