Skip to main content

AI Transcription

AI Transcription is an optional feature that automatically processes call recordings to produce:

  • Speech-to-Text transcription — converts WAV recordings into text
  • Transcription cleanup — refines the raw transcript using a large language model
  • Call summary — generates a concise summary of the conversation
  • Tagging — automatically detects and assigns relevant tags to the call recording
  • Sentiment analysis — evaluates the emotional tone of the conversation
info

AI Transcription requires Azure cloud services. The following section describes which resources must be created in the Azure Portal before the feature can be enabled.

Azure Services​

The following items must be created in Azure:

#ItemUsed for
1App RegistrationAuthentication of UCS against Azure APIs
2Azure OpenAISummary, transcription cleanup, tagging (GPT-4.1)
3Azure AI Services multi-service accountSpeech-to-Text, sentiment analysis

App Registration​

UCS authenticates against Azure using a service principal (App Registration). This single credential is shared by all Azure AI services.

  1. In Azure Portal, navigate to Microsoft Entra ID → App registrations → New registration.
  2. Set a name (e.g. ucs-ai) and register.
  3. Note the Application (client) ID and Directory (tenant) ID.
  4. Go to Certificates & secrets → New client secret, create a secret and note its Value.

The following values will be needed for UCS configuration:

ParameterWhere to find it
Tenant IDApp registration → Overview → Directory (tenant) ID
Client IDApp registration → Overview → Application (client) ID
Client SecretApp registration → Certificates & secrets → Value (shown only once at creation)
warning

Copy the client secret value immediately after creation — it cannot be retrieved later.

Azure OpenAI​

Azure OpenAI provides the large language model used for transcription cleanup, call summarization, and tag detection.

  1. In Azure Portal, navigate to Azure OpenAI → Create.
  2. Select a resource group, region, and pricing tier.
  3. Once created, go to Azure AI Foundry (formerly Azure OpenAI Studio) → Deployments → Deploy model.
  4. Deploy at least the gpt-4.1 model. Note the Deployment name (e.g. gpt-4.1).
  5. Go to the resource Overview page and note the Endpoint URL (e.g. https://<your-resource>.openai.azure.com/).
  6. Assign the Cognitive Services OpenAI User role to the App Registration service principal: navigate to Access control (IAM) → Add role assignment, select the role, and assign it to the App Registration created above.

Azure AI Services (multi-service account)​

A single Azure AI Services multi-service account provides access to both Speech (Speech-to-Text) and Language (sentiment analysis) APIs.

  1. In Azure Portal, search for Azure AI services and select Azure AI services multi-service account → Create.
  2. Select a resource group, region, and pricing tier.
  3. Once created, note the Endpoint URL and Region from the resource Overview page.
  4. Assign the Cognitive Services User role to the App Registration service principal: navigate to Access control (IAM) → Add role assignment, select the role, and assign it to the App Registration created above.
tip

If you prefer separate resources instead of a multi-service account, you can create Speech and Language resources individually. The setup steps are the same — just create each resource separately and note their respective endpoints.

Budget​

AI services are billed based on usage (tokens, audio minutes, text records). To avoid unexpected costs, it is strongly recommended to create a budget for the resource group containing your AI resources.

  1. In Azure Portal, navigate to the resource group where your AI resources are located.
  2. Go to Cost Management → Budgets → Add.
  3. Set the expected monthly budget amount.
  4. Configure alert thresholds (e.g. 80%, 100%) and notification recipients.
warning

Without a budget, there is no upper limit on the costs incurred by AI processing. Always set a budget that matches the expected monthly spend for your call volume.

Activating the Feature​

Once all Azure resources are provisioned, send the following information to INSOFT for activation:

ParameterSource
Tenant IDApp Registration → Overview → Directory (tenant) ID
Client IDApp Registration → Overview → Application (client) ID
Client SecretApp Registration → Certificates & secrets → Value
Azure OpenAI EndpointAzure OpenAI → Overview → Endpoint
OpenAI Deployment NameAzure OpenAI → Azure AI Foundry → Deployments
OpenAI API VersionAzure OpenAI → Azure AI Foundry → Deployments → Deployment details
AI Services EndpointAzure AI Services → Overview → Endpoint

Example:

Tenant ID:              a1b2c3d4-e5f6-7890-abcd-ef1234567890
Client ID:              12345678-abcd-ef01-2345-6789abcdef01
Client Secret:          aBcD1~ExAmPlEsEcReTvAlUe0123456789abcdef
Azure OpenAI Endpoint:  https://my-company-openai.openai.azure.com/
OpenAI Deployment Name: gpt-4.1
OpenAI API Version:     2025-01-01-preview
AI Services Endpoint:   https://my-company-ai.cognitiveservices.azure.com/