Recorded Translation

PreviousNext

File-level translation for recorded audio.

File-level translation for offline transcription.

Translation is available now on the batch-jobs plane. Declare the target languages once when you submit the job; the set is frozen for the whole job.

Request

Send target_languages (array, max 8) or translate_to (single-language shorthand):

curl -X POST https://audio.lansonai.com/v1/audio/transcriptions/jobs \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "audio_url": "https://cdn.example.com/audio.wav",
    "language": "zh",
    "target_languages": ["en", "ja"]
  }'
ParameterDescription
languageSource language of the audio (optional; auto-detected when omitted)
target_languagesTarget languages (BCP-47, max 8). Each language is billed separately.
translate_toSingle-language shorthand for target_languages: [translate_to]. Ignored when target_languages is present.

Omit both fields to get transcription only, without translation.

Output

Each target language produces one track with per-segment timestamps:

{
  "result": {
    "segments": [
      { "id": 0, "start_time": 0.0, "end_time": 3.2, "duration": 3.2, "text": "今天天气很好" }
    ],
    "translations": [
      {
        "language": "en",
        "text": "The weather is nice today.",
        "segments": [
          {
            "id": 0,
            "start_time": 0.0,
            "end_time": 3.2,
            "duration": 3.2,
            "text": "The weather is nice today.",
            "source_text": "今天天气很好"
          }
        ],
        "failed_batches": 0
      }
    ],
    "metadata": {
      "translation": {
        "target_languages": ["en"],
        "model": "deepseek/deepseek-v4.1-flash-1017",
        "failed_batches": 0,
        "usage": {
          "calls": 1,
          "prompt_tokens": 120,
          "completion_tokens": 24,
          "total_tokens": 144
        }
      }
    }
  }
}

translations[].segments corresponds 1:1 to the source sentence units (same id and timestamps). A failed batch is retried once; if it still fails, its lines are omitted from the track and failed_batches is incremented. One language failing never blocks another.

Billing

Translation is billed per target language × processed audio duration at the recorded translation add-on rate. See API pricing. Token usage is reported in result.metadata.translation.usage for observability; it is not a separate line item.