Azure Speech TTS Quick Start

Azure Speech (also called Azure Cognitive Services Speech) is Microsoft's text-to-speech service. It offers high-quality neural voices and supports many languages.

Prerequisites

1. Azure Account with Cognitive Services access
2. Azure Speech API Key from Azure Portal
3. Region where your Speech resource is deployed (e.g., eastus, westus2)
4. Python requests library (should already be installed)

Getting Your Credentials

1. Go to Azure Portal
2. Create a "Speech" resource (Cognitive Services → Speech)
3. After creation, go to "Keys and Endpoint"
4. Copy:
   • Key 1 or Key 2 (your API key)
   • Location/Region (e.g., eastus)

Configuration

1. Add Azure Speech to your config

Edit .babulus/config.yml:

providers:
  azure_speech:
    api_key: "your_azure_api_key_here"
    region: "eastus"  # Your Azure region
    voice: "en-US-JennyNeural"  # Default voice

2. Set provider in your DSL

Edit your .babulus.yml file:

voiceover:
  provider: azure-speech  # or just "azure"
  sample_rate_hz: 24000  # Azure supports: 8000, 16000, 24000, 44100

3. Environment-Based Configuration (Recommended)

For multi-environment workflows:

voiceover:
  provider:
    development: openai
    aws: aws
    azure: azure        # Use Azure Speech for testing
    production: elevenlabs

  sample_rate_hz:
    development: 24000  # OpenAI
    aws: 16000          # Polly
    azure: 24000        # Azure Speech (good balance)
    production: 44100   # ElevenLabs

Then generate with:

BABULUS_ENV=azure babulus generate your-video.babulus.yml

Available Voices

Azure offers 400+ voices across 140+ languages. Here are popular English options:

US English Neural Voices

• en-US-JennyNeural (Female, friendly)
• en-US-GuyNeural (Male, warm)
• en-US-AriaNeural (Female, professional)
• en-US-DavisNeural (Male, professional)
• en-US-AmberNeural (Female, young adult)
• en-US-AnaNeural (Female, child)
• en-US-AndrewNeural (Male, young adult)
• en-US-BrianNeural (Male, child)

UK English Neural Voices

• en-GB-SoniaNeural (Female)
• en-GB-RyanNeural (Male)
• en-GB-LibbyNeural (Female, young)

Other English Variants

• en-AU (Australian): Natasha, William, Annette
• en-CA (Canadian): Clara, Liam
• en-IN (Indian): Neerja, Prabhat
• en-IE (Irish): Emily, Connor

See full voice list for all languages.

Voice Selection

Override the voice per DSL:

voiceover:
  provider: azure
  voice: en-US-GuyNeural  # Override default voice
  sample_rate_hz: 24000

Or use the default from config.

Sample Rate Options

Azure Speech supports flexible sample rates:

• 8000 Hz: Telephony quality (smallest files)
• 16000 Hz: Good quality (medium files)
• 24000 Hz: High quality (recommended, good balance)
• 44100 Hz: Maximum quality (largest files)

Pricing

• Standard voices: Free for first 0.5M characters/month, then $4/1M chars
• Neural voices: Free for first 0.5M characters/month, then **16/1Mchars * *( 0.016/1K chars)
• More expensive than OpenAI (0.015/1K)andAWSPolly(0.004-0.016/1K)
• Less expensive than ElevenLabs (~$0.30/1K)

Troubleshooting

Error: "Azure TTS failed (401)"

Solution: Check your API key and region are correct in config.

providers:
  azure_speech:
    api_key: "YOUR_ACTUAL_KEY"  # Get from Azure Portal
    region: "YOUR_REGION"       # e.g., eastus, westus2

Error: "Azure TTS sample_rate_hz must be one of 8000, 16000, 24000, 44100"

Solution: Use a supported sample rate:

voiceover:
  sample_rate_hz: 24000  # Change to 8000, 16000, 24000, or 44100

Voice Not Found

Solution: Check the voice name is correct. Voice names are case-sensitive and must include language prefix:

• ✅ Correct: en-US-JennyNeural
• ❌ Wrong: jenny, JennyNeural, en-us-jennyneural

SSML Support

Azure Speech uses SSML (Speech Synthesis Markup Language) internally. Currently, Babulus sends plain text which Azure wraps in basic SSML. Future versions may expose more SSML features for:

• Prosody control (rate, pitch, volume)
• Pauses and breaks
• Pronunciation hints
• Emphasis and expression

Comparison with Other Providers

Example: Complete DSL

voiceover:
  provider: azure
  voice: en-US-GuyNeural
  sample_rate_hz: 24000
  seed: 1337
  lead_in_seconds: 0.25

scenes:
  - id: intro
    title: "Introduction"
    cues:
      - id: welcome
        label: "Welcome"
        voice: "Welcome to our presentation about Azure Speech services."

Advanced: Multiple Voices

You can mix voices within a video by overriding voice per cue:

voiceover:
  provider: azure
  voice: en-US-JennyNeural  # Default female voice
  sample_rate_hz: 24000

scenes:
  - id: dialog
    cues:
      - id: host
        voice: "Hello, I'm the host."
        # Uses default: en-US-JennyNeural

      - id: guest
        voice:
          model: en-US-GuyNeural  # Override to male voice
          segments:
            - voice: "And I'm the guest speaker."

Regional Endpoints

Azure Speech uses regional endpoints. Common regions:

• eastus - East US (Virginia)
• westus2 - West US 2 (Washington)
• westeurope - West Europe (Netherlands)
• southeastasia - Southeast Asia (Singapore)

Choose the region closest to you for lower latency, or use the region where you created the Azure resource.