All the configuration properties described below are defined in the POST /v2/live endpoint.

​
Language configuration

​
Single language

If you know the language of the conversation in advance, specify it in the language_config.languages parameter to ensure the best transcription results. 
{
  ...rest of the config
  "language_config": {
    "languages": ["en"]
  }
}
If the spoken language is unknown, you can:
  • Omit the language_config.languages parameter; the model will automatically detect the language from the audio across all supported languages.
  • Specify multiple languages in the language_config.languages parameter; the model will detect the language from the audio within the provided options.
{
  ...rest of the config
  // Will limit to detect from English, French or Spanish
  "language_config": {
    "languages": ["en", "fr", "es"]
  }
}

​
Multiple languages
(Code-switching)

If you expect multiple languages to be spoken during the conversation, enable the language_config.code_switching parameter. This will allow the model to switch languages dynamically and reflect it in the transcription results. As with single-language configuration, you can either let the model detect the language from all supported languages or specify a set of options to narrow down the selection. 
{
  ...rest of the config
  "language_config": {
    // Will limit to detect from English, French or Spanish
    "languages": ["en", "fr", "es"],
    "code_switching": true
  }
}

It is recommended to limit the number of languages to avoid incorrect detection, either in single or multiple languages configuration. Some languages, such as those from Eastern European countries, have similar sounds, which may cause the model to confuse them and produce a transcription in the wrong language.

​
Word-level timestamps

Instead of just getting timestamps for when utterances begin and end, Gladia’s real-time API provides word-level timestamps. This lets you know the exact timestamp for each word, giving you a more precise transcription, facilitating detailed analysis and more accurate synchronization with audio and video files. To enable it, pass the following configuration: 
{
  "realtime_processing": {
    "words_accurate_timestamps": true
  }
}
Under each utterance, you’ll find a words property, like this: 
{
  // ... other utterance properties
  "words": [
    {
      "word": "Split",
      "start": 0.21001999999999998,
      "end": 0.69015,
      "confidence": 1
    },
    {
      "word": " infinity",
      "start": 0.91021,
      "end": 1.55038,
      "confidence": 0.95
    },
  ]
}

​
Custom vocabulary

To enhance the precision of words you know will recur often in your transcription, use the custom_vocabulary feature.
 
{
  "realtime_processing": {
    "custom_vocabulary": true,
    "custom_vocabulary_config": {
      "vocabulary": [
        "Westeros", 
        {"value": "Stark"}, 
        { 
          "value": "Night's Watch",
          "pronunciations": ["Nightz Watch"],
          "intensity": 0.4,
          "language": "en"
        }
      ],
      "default_intensity": 0.6
    }
  }
}
Using a string in place of an object is equivalent to use {"value": "string"}
  • default_intensity: [optional] The global intensity of the feature (minimum 0, maximum 1, default 0.5).
  • vocabulary.value: [required] The text used to replace in the transcription.
  • vocabulary.pronunciations: [optional] The pronunciations used in the transcription language, or vocabulary.language if present.
  • vocabulary.intensity: [optional] The intensity of the feature for this particular word (minimum 0, maximum 1, default 0.5).
  • vocabulary.language: [optional] Specify the language in which it will be pronounced when sound comparison occurs. Default to transcription language.

​
Custom spelling

You can customize how certain words, names or phrases are spelled in the final transcript.
To use custom spelling, provide a dictionary through the custom_spelling_config parameter. This dictionary should contain the correct spelling as the key and a list of one or more possible variations as the value. Custom spelling is useful in scenarios where consistent spelling of specific words is crucial (e.g., technical terms in industry-specific recordings).
The keys in the dictionary are case sensitive, while the values aren’t. Values can contain multiple words.
request data
{
  "realtime_processing": {
    "custom_spelling": true,
    "custom_spelling_config": {
      "spelling_dictionary": {
          "Gorish": ["ghorish", "gaurish", "gaureish"],
          "Data Science": ["data-science", "data science"],
          ".": ["period", "full stop"],
          "SQL": ["sequel"]
      }
    }
  }
}
In this example, the model will ensure that “Gorish” is spelled correctly throughout the transcript, even if it is pronounced in various ways such as “ghorish,” “gaurish,” or “gaureish.”

​
Translation

This feature allows you to get both the original transcription and its translation in your specified target language(s) in real-time. 
{
  "realtime_processing": {
    "translation": true,
    "translation_config": {
      "target_languages": ["fr", "es"],  // Translate to French and Spanish
      "model": "base",                    // "base" or "enhanced"
      "context_adaptation": true,         // Enable context-aware translation
      "context": "Technical support call about software issues",
      "informal": false                   // Use formal language
    }
  }
}

​
Translation Configuration Options:

  • target_languages: Array of language codes for translation outputs (e.g., ["fr", "es"])
  • model: Translation model - "base" (fast) or "enhanced" (higher quality, context-aware)
  • context_adaptation: Boolean to enable/disable context-aware translation features (default: true)
  • context: String providing context to improve translation accuracy (default: "")
  • informal: Boolean to force informal language forms when available (default: false)
With this feature enabled, you’ll receive a translation message for each transcript message and target language.

​
Multiple channels

If you have multiple channels in your audio stream, specify the count in the configuration: 
{
  "channels": 2
}
Gladia’s real-time API will automatically split the channels and transcribe them separately. For each utterance, you’ll get a channel key corresponding to the channel the utterance came from.
Transcribing an audio stream with multiple channels will be billed exponentially. For example, an audio stream with 2 channels will be billed as double the audio duration, even if the channels are identical.
For a detailed guide on how to merge multiple audio tracks into a single multi-channel stream and send it over a WebSocket, see the Sending multiple audio tracks over a single WebSocket section.

​
Attaching custom metadata

You can attach metadata to your real-time transcription session using the custom_metadata property. This’ll make it easy to recognize your transcription when you receive data from the GET /v2/live/:id endpoint. And more importantly, you’ll be able to use it as a filter in the GET /v2/live list endpoint. For example, you can add the following to your configuration: 
"custom_metadata": {
    "internalUserId": 2348739875894375,
    "paymentMethod": {
        "last4Digits": 4576
     },
     "internalUserName": "Spencer"
}
And use a GET request to filter results, like this: 
https://api.gladia.io/v2/live?custom_metadata={"internalUserId": "2348739875894375"}
or like this: 
https://api.gladia.io/v2/live?custom_metadata={"paymentMethod": {"last4Digits": 4576}, "internalUserName": "Spencer"}
custom_metadata cannot be longer than 2000 characters when stringified.