The Audio API provides a single speech to text endpoint:

  • transcriptions

File uploads are limited to 1 GB (if you need larger file support, you can write to Support), and the following input file types are supported: mp3, wav, mp3, m4a, flac, ogg, opus, mp4, mov, avi, and mkv. Furthermore, the API supports sending direct links to the files and neural speaker diarization.

​
Transcriptions

The transcriptions API takes as input the audio (or video) file you want to transcribe (or the link to that file) and the desired output file format for the transcription of the audio. The model supports the following output formats (json, text, srt, verbose_json, vtt). Read more abouit the output formats on the Response Formats page.

import requests

url = "https://api.nexara.ru/api/v1/audio/transcriptions"
api_key = "NEXARA_API_KEY"  # Replace with your full API key

headers = {
    "Authorization": f"Bearer {api_key}",
}

file_path = "audio/example.mp3"

with open(file_path, "rb") as audio_file:
    files = {
        "file": (file_path, audio_file, "audio/mp3"),  # Change MIME type if needed
    }

    # Or you can pass the "url" parameter and send the direct link to the audio file.
    data = {
        "response_format": "json",
        # "task": "diarize" # if you want to perform speaker diarization
        # By default, the task is set to "transcribe"
    }

    response = requests.post(url, headers=headers, files=files, data=data)

    transcription = response.json()
    print(transcription)

By default, the response type will be json:

{
  "text": "Four score and seven years ago our fathers brought forth on this continent, a new nation, conceived in Liberty, and dedicated to the proposition that all men are created equal.
...
}

​
Supported Languages

We currently support the following languages through the /transcriptions endpoint:

Afrikaans, Arabic, Armenian, Azerbaijani, Belarusian, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuanian, Macedonian, Malay, Marathi, Maori, Nepali, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swahili, Swedish, Tagalog, Tamil, Thai, Turkish, Ukrainian, Urdu, Vietnamese, and Welsh.

While the model was trained on 98 languages, the list above only contains languages that exceeded 50% word error rate (WER). View the entire language list and their codes in the ISO-639-1 format on the Supported Languages page.

​
Timestamp Granularities

To get detailed timing information alongside your transcription, you can use the timestamp_granularities[] parameter. This feature requires setting the response_format to verbose_json. For more details, visit the API Documentation.

Currently, two levels of granularity are available:

  1. segment Granularity:

    • How to request: Include timestamp_granularities[]=segment in your API request (along with response_format=verbose_json).
    • What it returns: The verbose_json response will contain a segments array. Each object in this array represents a larger chunk or segment of the transcribed audio (30 seconds in length). Each segment object includes its start time, end time, and the transcribed text for that specific chunk.

  2. word Granularity:

    • How to request: Include timestamp_granularities[]=word in your API request (along with response_format=verbose_json).
    • What it returns: This provides the most detailed timing information. The verbose_json response will contain:
      • A words array: Each object in this array details an individual recognized word, including the word itself, its precise start time, and its end time in seconds.
      • A segments array: Even when requesting word granularity, the response still includes the segments array described above. This gives you both fine-grained word timing and the broader segment structure in a single response.

In Summary:

  • Use segment if you only need timing for larger chunks of speech.
  • Use word if you need precise start/end times for individual words. Requesting word granularity conveniently provides both the words and segments arrays.
  • Remember, both options require response_format=verbose_json.
import requests

url = "https://api.nexara.ru/api/v1/audio/transcriptions"
api_key = "NEXARA_API_KEY"  # Replace with your full API key

headers = {
    "Authorization": f"Bearer {api_key}",
}

file_path = "audio/example.mp3"

with open(file_path, "rb") as audio_file:
    files = {
        "file": (file_path, audio_file, "audio/mp3"),  # Change MIME type if needed
    }
    data = {
        "response_format": "verbose_json",
        "timestamp_granularities[]": "word", # Returns both word timestamps and segment timestamps.
    }

    response = requests.post(url, headers=headers, files=files, data=data)

    transcription = response.json()
    print(transcription)

Here’s an example of an output:

{
  "duration": 109.69609977324264,
  "language": "en",
  "task": "transcribe",
  "text": "Four score and seven years ago, our fathers brought forth on this continent a new nation, conceived in liberty and dedicated to the proposition that all men are created equal. ... ",
  "words": [
    {
      "end": 0.62,
      "prob": 0.07,
      "start": 0,
      "word": "Four"
    },
    {
      "end": 1.12,
      "prob": 0.21,
      "start": 0.62,
      "word": "score"
    },
    {
      "end": 1.44,
      "prob": 0.84,
      "start": 1.12,
      "word": "and"
    },
    {
      "end": 1.82,
      "prob": 0.95,
      "start": 1.44,
      "word": "seven"
    },
    {
      "end": 2.18,
      "prob": 0.99,
      "start": 1.82,
      "word": "years"
    },
    {
      "end": 3.54,
      "prob": 0.68,
      "start": 2.18,
      "word": "ago,"
    },
    {
      "end": 3.84,
      "prob": 0.94,
      "start": 3.54,
      "word": "our"
    },
    {
      "end": 4.16,
      "prob": 0.46,
      "start": 3.84,
      "word": "fathers"
    },
    {
      "end": 4.52,
      "prob": 0.97,
      "start": 4.16,
      "word": "brought"
    },
    {
      "end": 4.94,
      "prob": 0.97,
      "start": 4.52,
      "word": "forth"
    },
    {
      "end": 5.16,
      "prob": 0.92,
      "start": 4.94,
      "word": "on"
    },
    {
      "end": 5.28,
      "prob": 0.77,
      "start": 5.16,
      "word": "this"
    },
    {
      "end": 5.64,
      "prob": 0.93,
      "start": 5.28,
      "word": "continent"
    },
    {
      "end": 6.12,
      "prob": 0.82,
      "start": 5.64,
      "word": "a"
    },
    {
      "end": 6.18,
      "prob": 0.98,
      "start": 6.12,
      "word": "new"
    },
    {
      "end": 7.94,
      "prob": 0.75,
      "start": 6.18,
      "word": "nation,"
    },
    {
      "end": 8.16,
      "prob": 0.98,
      "start": 7.94,
      "word": "conceived"
    },
    {
      "end": 8.46,
      "prob": 0.95,
      "start": 8.16,
      "word": "in"
    },
    {
      "end": 8.8,
      "prob": 0.98,
      "start": 8.46,
      "word": "liberty"
    },
    {
      "end": 9.64,
      "prob": 0.43,
      "start": 8.8,
      "word": "and"
    },
    {
      "end": 10.02,
      "prob": 1,
      "start": 9.64,
      "word": "dedicated"
    },
    {
      "end": 10.64,
      "prob": 0.96,
      "start": 10.02,
      "word": "to"
    },
    {
      "end": 10.86,
      "prob": 1,
      "start": 10.64,
      "word": "the"
    },
    {
      "end": 11.32,
      "prob": 0.98,
      "start": 10.86,
      "word": "proposition"
    },
    {
      "end": 11.78,
      "prob": 1,
      "start": 11.32,
      "word": "that"
    },
    {
      "end": 11.88,
      "prob": 0.91,
      "start": 11.78,
      "word": "all"
    },
    {
      "end": 12.1,
      "prob": 0.99,
      "start": 11.88,
      "word": "men"
    },
    {
      "end": 12.4,
      "prob": 0.98,
      "start": 12.1,
      "word": "are"
    },
    {
      "end": 12.68,
      "prob": 0.98,
      "start": 12.4,
      "word": "created"
    },
    {
      "end": 14.68,
      "prob": 0.95,
      "start": 12.68,
      "word": "equal."
    }
  ],
  "segments": [
    {
      "avg_logprob": -1,
      "compression_ratio": 2.4,
      "end": 29,
      "id": 0,
      "no_speech_prob": 0.5,
      "seek": 0,
      "start": 0,
      "temperature": 1,
      "text": "Four score and seven years ago, our fathers brought forth on this continent a new nation, conceived in liberty and dedicated to the proposition that all men are created equal. ...",
      "tokens": [
        7451, 6175, 293, 3407, 924, 2057, 11, 527, 23450, 3038, 5220, 322, 341,
        18932, 257, 777, 4790, 11, 34898, 294, 22849, 293, 8374, 281, 264,
        24830, 300, 439, 1706, 366, 2942, 2681, 13, 823, 321, 366, 8237
      ]
    }
  ]
}

​
Response Formats

The Nexara Transcription API allows you to specify the format in which you want to receive the transcription results. By using the response_format parameter in your request, you can tailor the output to best suit your needs, whether you require simple text, structured data for programmatic use, or ready-to-use subtitle files. Refer to the API Documentation for examples.

The API supports the following output formats:

  • json: Returns a standard JSON object containing the transcribed text.
  • text: Returns the transcription as a single plain text string.
  • verbose_json: Returns a detailed JSON object containing the text, language, duration, and potentially segment and word-level timestamps (if requested via timestamp_granularities[]).
  • srt: Returns the transcription formatted as an SRT subtitle file.
  • vtt: Returns the transcription formatted as a WebVTT subtitle file.

​
Speaker Diarization

Nexara API supports neural speaker diarization. This feature identifies different speakers in an audio file and attributes transcribed segments to each one. To get started, simply set the task parameter to "diarize" in your API request.

Furthermore, you can pass the num_speakers parameter to specify the number of speakers you expect in the audio file. Note that this parameter does not guarantee the correct number of speakers but can guide the model.

The model also supports three diarization settings:

  • general: Default setting for general use.
  • meeting: Optimized for meetings.
  • telephonic: Optimized for telephonic conversations.

Note that performing speaker diarization is more computationally intensive than just transcribing an audio file, so the delay for the response will be longer. Also, note that diarization costs twice as much as a regular transcription.

import requests

url = "https://api.nexara.ru/api/v1/audio/transcriptions"
api_key = "NEXARA_API_KEY"  # Replace with your full API key

headers = {
    "Authorization": f"Bearer {api_key}",
}

file_path = "audio/example.mp3"

with open(file_path, "rb") as audio_file:
    files = {
        "file": (file_path, audio_file, "audio/mp3"),  # Change MIME type if needed
    }

    # Or you can pass the "url" parameter and send the direct link to the audio file.
    data = {
        "task": "diarize",
        # "num_speakers": 2,
        # "diarization_setting": "telephonic"
    }

    response = requests.post(url, headers=headers, files=files, data=data)

    transcription = response.json()
    print(transcription)

When you request diarization, the API will always return a JSON response. This response includes a segments array. Each object within this array will have a speaker field (e.g., “speaker_0”, “speaker_1”) identifying who spoke during that specific audio segment, along with their start time, end time, and the text they spoke.

The output will always be in the following format, no matter the response_format setting:

{
    "task": "diarize",
    "language": "en",
    "duration": 9.12,
    "text": "The beach was a popular spot on a hot summer day. People were swimming in the ocean, building sandcastles, and playing beach volleyball.",
    "segments": [
        {
            "speaker": "speaker_0",
            "start": 0.00,
            "end": 5.00,
            "text": "The beach was a popular spot on a hot summer day."
        },
        {
            "speaker": "speaker_1",
            "start": 5.00,
            "end": 9.12,
            "text": "People were swimming in the ocean, building sandcastles, and playing beach volleyball."
        }
    ]
}

Note that Nexara doesn’t support channel diarization. This means that the API will ignore channel information and identify speakers based on the audio content alone.