Deploying OpenAI Text-to-Speech (TTS) Model on Azure: A Step-by-Step Guide

Azure Cognitive Services provides a straightforward way to deploy OpenAI models, including powerful text-to-speech capabilities. In this guide, I’ll demonstrate how to deploy a text-to-speech model using Azure CLI commands.

Prerequisites

• An Azure subscription
• Azure CLI installed and logged in (az login)

Step 1: Define Environment Variables

Set your environment variables to simplify and standardize deployments.

export AZURE_RESOURCE_GROUP="example-rg"
export AZURE_REGION="eastus"
export OPENAI_NAME="example-openai"
export AZURE_SUBSCRIPTION_ID="your-subscription-id"

# Keep these variables as is.
export AUDIO_MODEL="gpt-4o-mini-tts"
export AUDIO_MODEL_VERSION="2025-03-20"

Explanation:

• AZURE_RESOURCE_GROUP: Azure Resource Group name. Choose a unique name for your resource group to contain all related resources.
• AZURE_REGION: Azure region where resources will be created. Common regions include eastus, westus, westeurope, etc.
• OPENAI_NAME: Unique name for your OpenAI resource. This will be used to access your deployed model.
• AZURE_SUBSCRIPTION_ID: Azure subscription identifier. You can find this in your Azure portal under “Subscriptions”.
• AUDIO_MODEL: The specific model you want to deploy. For text-to-speech, gpt-4o-mini-tts is a suitable choice.
• AUDIO_MODEL_VERSION: The version of the model you wish to use. This should match the version available in Azure OpenAI service.

Step 2: Create Resource Group

Create a resource group to contain your OpenAI resources, if it doesn’t already exist, by running the following command:

az group create \
  --name "${AZURE_RESOURCE_GROUP}" \
  --location "${AZURE_REGION}" \
  --subscription "${AZURE_SUBSCRIPTION_ID}"

Step 3: Create Cognitive Services Account

Create an Azure Cognitive Services account for hosting your OpenAI model:

az cognitiveservices account create \
  --name "${OPENAI_NAME}" \
  --resource-group "${AZURE_RESOURCE_GROUP}" \
  --location "${AZURE_REGION}" \
  --kind OpenAI \
  --sku S0 \
  --custom-domain "${OPENAI_NAME}" \
  --subscription "${AZURE_SUBSCRIPTION_ID}"

Explanation:

• --name: Sets the unique name for the service.
• --kind: Specifies the kind of service (OpenAI).
• --sku: Selects the pricing tier (S0 is standard).
• --custom-domain: Custom domain name for accessing the service.

Step 4: Deploy the OpenAI Model

Deploy your chosen model (gpt-4o-mini-tts) to your account:

az cognitiveservices account deployment create \
  --name "${OPENAI_NAME}" \
  --resource-group "${AZURE_RESOURCE_GROUP}" \
  --deployment-name "${OPENAI_NAME}-${AUDIO_MODEL}" \
  --model-name "${AUDIO_MODEL}" \
  --model-version "${AUDIO_MODEL_VERSION}" \
  --model-format OpenAI \
  --sku-capacity 1 \
  --sku-name "GlobalStandard"

Explanation:

• --deployment-name: Unique name for this specific model deployment.
• --model-name: The OpenAI model to deploy.
• --model-version: Specific version of the model.
• --model-format: Format of the model (OpenAI).
• --sku-capacity: Number of instances allocated for the deployment.
• --sku-name: SKU type that dictates the availability and performance.

Step 5: Retrieve API Key

Fetch the API key necessary for accessing your deployed model:

export AZURE_OPENAI_API_KEY=$(az cognitiveservices account keys list \
  --name "${OPENAI_NAME}" \
  --resource-group "${AZURE_RESOURCE_GROUP}" \
  --query key1 -o tsv)

Explanation:

• --query key1: Retrieves the primary key of your Cognitive Services account.
• -o tsv: Outputs the result in a tab-separated format for easy scripting.

Step 6: Set Endpoint URL

Create an environment variable for your model endpoint:

export OPENAI_ENDPOINT="https://${OPENAI_NAME}.openai.azure.com/openai/deployments/${OPENAI_NAME}-${AUDIO_MODEL}/audio/speech?api-version=2025-04-01-preview"

Verify Deployment with cURL

To verify your deployment, run the following cURL command:

curl "$OPENAI_ENDPOINT" \
 -H "api-key: $AZURE_OPENAI_API_KEY" \
 -H "Content-Type: application/json" \
 -d '{
  "model": "gpt-4o-mini-tts",
  "input": "I am excited to try text to speech.",
  "voice": "alloy",
  "instructions": "Speak in a cheerful and positive tone."
 }' --output speech.mp3

Explanation:

• Sends a POST request to the OpenAI endpoint.
• -H: Includes headers specifying API key and content type.
• -d: Defines the JSON payload with model details, input text, voice choice, and speaking instructions.
• --output: Saves the generated speech audio to speech.mp3.

Summary

You’ve successfully deployed an OpenAI text-to-speech model on Azure. You can now leverage Azure’s powerful infrastructure and your custom deployment to build advanced audio-based applications. Always ensure you securely handle your API keys in production environments.