Configure live transcription and AI analysis

To use live transcription and AI analysis, you'll need to configure the following settings in System Configurator.

Prerequisites

Make sure that call recording is configured for the necessary queues. For more information, see the Call recording document.

To use live transcription and AI analysis, please create a service request or contact your Sinch account manager.

Configure

  1. Configure live transcription and AI analysis for either one of these:

    • voice channel at System Management > Channels > Voice Channel if you'd like to use live transcription and AI analysis for all phone queues

      Select Use live transcription and Use AI analysis.

    • phone queue at Queue Management > [phone queue] > Live transcription and AI settings if you'd like to use live transcription and AI analysis only in specific phone queue(s)

      Select On for Use live transcription and Use AI analysis.

    Enter the value in the Transcription parameters field as JSON. To format your JSON into a reader-friendly form, click the Prettify button .

    The parameters available depend on whether your system is using Sinch's Base or Plus transcription tier:

    Base tier

    If you want to turn off capitalization and punctuation, which is on by default, set the value of the format_turns parameter as false in the Transcription parameters field.

    Example:

    {
  "format_turns": false
}

    Plus tier

    In the Transcription parameters field, you can suggest language options to the AI for transcribing speech. We recommend defining either one of these options. Use ISO language codes to define the language(s).

    Preference:

    {
  "language_hints": ["fi", "sv"]
}

    These are the languages the transcription model expects in the audio. It will prioritize them when interpreting speech but can still detect and transcribe other languages when they are spoken.

    This is also the behavior if language_hints_strict is set to false.

    Best-effort restriction:

    {
  "language_hints": ["fi"],
  "language_hints_strict": true
}

    Transcription will strongly prioritize these languages and will try to produce a transcription only using them. This is a best-effort restriction rather than a strict limitation: in ambiguous or exceptional cases, another language may still be used in transcription. You can define several languages but defining only one gives the best results.

    Use this option when the expected language is known and transcription in other languages is undesirable.

    By default, the AI analysis includes fields:

    • Suggested actions: three best suggestions for next steps in the conversation

    • Context: a concise characterization of the context of the conversation

    • Keywords: a few keywords to describe the conversation

    • Language: the language of the conversation

    • Sentiment: the customer sentiment in the conversation

    • Summary: a compact summary of the conversation

    You can customize the AI analysis to suit your needs by specifying the prompt in the Custom AI instructions field. This makes it possible, for example, to:

    • Highlight any billing issues in the summary.

    • Replace personal information with a placeholder to hide sensitive data from the summary.

    • Add a dedicated Offer field to flag upsell possibilities to offer the customer upgrades or additional services.

  2. Select if agents are allowed to see the live transcription and the AI analysis in Communication Panel on the user settings template at User and Role Management > User Settings Template > [user settings template] > Communication Panel (CP) AI settings .

  3. If you want users to be able to access the transcription in the History view after the call, make sure they have Listen to recording rights on the level (queue/agent/campaign) that does the call recording.