Nemotron 3.5 Transcribe

Transcribe speech to text with Nemotron 3.5 Transcribe, KIE hosted. Send a public audio URL, or a video with an audio track, and get the transcript; the spoken language is detected automatically. Optionally add word-level timestamps with speaker labels, or translate the speech instead. The model uses the parameter model=nvidia/nemotron-3-5-transcribe.

Capabilities

Transcription in 40 languages, detected automatically

Word-level timestamps with speaker labels (timestamps: true)

Speech translation to or from English (target_language)

Input: MP3, WAV, M4A/AAC, FLAC, OGG/Opus, WEBM, or an MP4/MOV video with an audio track; up to 60 minutes and 200 MB per file

Whole-file transcription; streaming is not supported

Choosing parameters

Plain transcription: send only audio_url.

Timestamps and speakers: add timestamps: true. The result file then includes every word's start and end time and a speaker label. Transcription only.

Translation: set target_language to the output language. Translation is supported to or from English: speech in another language into English (en), or English speech into any other listed language. Translation between two non-English languages may work but is not guaranteed; for best quality, translate into English. target_language and timestamps: true cannot be used in the same request.

audio_url must be directly downloadable without login. Avoid signed links that expire within a few minutes, and make sure large files can be downloaded within 3 minutes.

Supported transcription languages: English, Chinese (Mandarin), Japanese, Korean, French, German, Spanish, Portuguese, Italian, Russian, Dutch, Polish, Czech, Slovak, Hindi, Bulgarian, Croatian, Danish, Estonian, Finnish, Hungarian, Latvian, Lithuanian, Romanian, Slovenian, Swedish, Thai, Turkish, Ukrainian, Arabic, Bengali, Greek, Hebrew, Indonesian, Maltese, Marathi, Tagalog, Tamil, Telugu, Vietnamese.

Result format

When the task succeeds, Get Task Details returns resultJson as a JSON string with a resultObject:

{
  "code": 200,
  "msg": "success",
  "data": {
    "taskId": "task_nemotron_1762345821000",
    "model": "nvidia/nemotron-3-5-transcribe",
    "state": "success",
    "param": "{\"model\": \"nvidia/nemotron-3-5-transcribe\", \"callBackUrl\": \"https://your-domain.com/api/callback\", \"input\": \"{\\\"audio_url\\\": \\\"https://your-domain.com/audio/team-meeting.mp3\\\", \\\"timestamps\\\": true}\"}",
    "resultJson": "{\"resultObject\": {\"text\": \"Thanks for joining, everyone. Let's start with the launch checklist for Friday.\", \"duration\": 12.64, \"url\": \"https://tempfile.aiquickdraw.com/example/transcript-result.json\", \"expires_at\": \"2026-11-19T12:30:23Z\"}}",
    "failCode": "",
    "failMsg": "",
    "costTime": 3,
    "completeTime": 1762345824000,
    "createTime": 1762345821000,
    "updateTime": 1762345824000
  }
}

resultObject fields:

text: the full transcript, or the translation when target_language is set. With timestamps on, this is plain text without speaker tags

duration: audio length in seconds, measured from the decoded audio; credits are charged per second of this value

url: the full result as a JSON file (structure below)

expires_at: when url expires (about 14 days after completion). Download the file before then if you need to keep it

Result file at url (example with timestamps: true):

{
  "task": "transcribe",
  "duration": 12.64,
  "text": "Thanks for joining, everyone. Sure, let's start.",
  "text_with_speakers": "<spk:0> Thanks for joining, everyone. <spk:1> Sure, let's start.",
  "words": [
    {
      "word": "Thanks",
      "start": 0.32,
      "end": 0.61,
      "speaker": "0"
    },
    {
      "word": "for",
      "start": 0.61,
      "end": 0.74,
      "speaker": "0"
    },
    {
      "word": "joining,",
      "start": 0.74,
      "end": 1.18,
      "speaker": "0"
    },
    {
      "word": "everyone.",
      "start": 1.18,
      "end": 1.83,
      "speaker": "0"
    },
    {
      "word": "Sure,",
      "start": 2.46,
      "end": 2.8,
      "speaker": "1"
    },
    {
      "word": "let's",
      "start": 2.84,
      "end": 3.07,
      "speaker": "1"
    },
    {
      "word": "start.",
      "start": 3.07,
      "end": 3.52,
      "speaker": "1"
    }
  ],
  "diarization": [
    {
      "speaker": 0,
      "start": 0.32,
      "end": 1.83
    },
    {
      "speaker": 1,
      "start": 2.46,
      "end": 3.52
    }
  ]
}

task: transcribe or translate

text, duration: same as in resultObject

text_with_speakers: transcript with speaker tags <spk:N>; only with timestamps: true

words: start and end time in seconds from the beginning of the audio, and speaker label, for each word (each character for Chinese); only with timestamps: true

diarization: speaker turns detected from the audio; only with timestamps: true

silent: true when the audio has almost no sound; text is then an empty string

truncated: true when the output hit the length limit for the audio duration. This usually means the output repeats itself, so discard or review such results

Good to know

Speaker labels start at 0. Voices that sound very different (for example male and female) separate well; similar voices may be merged into one speaker.

Word timestamps: end times are typically within a few hundredths of a second; start times can be off by around half a second, more for English.

Running the same audio twice can produce small wording differences (for example "but is" and "but it's"); do not rely on exact text matches to detect duplicates.

In Chinese output, foreign names are usually transliterated, and some model numbers may be written out in characters.

Common failure messages

failMsg	Cause
Audio fetch failed. Check access settings or use our File Upload API instead.	The URL returns 4xx or the domain cannot be resolved
The request failed because the audio format is not supported by the API.	No decodable audio: not an audio file, a video without an audio track, an empty or too short file
The audio file exceeds the maximum allowed size of 200 MB.	File larger than 200 MB
The audio exceeds the maximum allowed duration of 60 minutes.	Audio longer than 60 minutes; split it and submit the parts

Query Task Status

After submitting a task, use the unified query endpoint to check progress and retrieve results:

Get Task Details

Learn how to query task status and retrieve generation results

For production use, we recommend using the callBackUrl parameter to receive automatic notifications when the task completes, rather than polling the status endpoint.

Market Overview

Explore all available models

File Upload API

Learn how to upload and manage files

Common API

Check credits and account usage

Bearer Token

Provide your bearer token in the

Authorization

header when making requests to protected resources.

Example:

Authorization: Bearer ********************

curl --location 'https://api.kie.ai/api/v1/jobs/createTask' \ --header 'Authorization: Bearer <token>' \ --header 'Content-Type: application/json' \ --data '{ "model": "nvidia/nemotron-3-5-transcribe", "callBackUrl": "https://your-domain.com/api/callback", "input": { "audio_url": "https://static.aiquickdraw.com/tools/example/1793873594707_8fvLCiGf.mp3", "timestamps": true } }'

Nemotron 3.5 Transcribe

Capabilities

Choosing parameters

Result format

Good to know

Common failure messages

Query Task Status

Request

Responses

Nemotron 3.5 Transcribe

Capabilities#

Choosing parameters#

Result format#

Good to know#

Common failure messages#

Query Task Status#

Related Resources#

Request

Responses

Capabilities

Choosing parameters

Result format

Good to know

Common failure messages

Query Task Status

Related Resources