Skip to content

AZURE-SETUP.md

Latest commit

 

History

History
343 lines (259 loc) · 13.1 KB

AZURE-SETUP.md

File metadata and controls

343 lines (259 loc) · 13.1 KB

Microsoft Azure Cognitive Services TTS Setup Guide

Updated: October 2025 | Status: ✅ PRODUCTION READY

Complete setup and configuration guide for Azure Cognitive Services Text-to-Speech integration with TextToSpeech Generator v2.0.

Azure TTS Status

🔵 Overview

Microsoft Azure Cognitive Services Text-to-Speech delivers industry-leading neural voices with natural prosody and clear articulation. As of October 2025, Azure offers 490+ voices across 140+ languages with advanced neural capabilities, making it the most comprehensive TTS solution available.

✅ Full Implementation Status: This provider is completely implemented in TextToSpeech Generator v2.0 with real API calls, SSML support, and enterprise-grade error handling.

Key Benefits

  • Neural Voice Quality: Human-like speech with natural intonation
  • Global Reach: 140+ languages and regional variants
  • Flexible Pricing: Free tier available, pay-per-use scaling
  • Enterprise Features: Custom neural voices, SSML support
  • High Availability: 99.9% uptime SLA with global datacenters

🚀 Getting Started

Prerequisites

  • Azure subscription (free tier available)
  • Valid email address for account creation
  • Credit card for paid features (optional for free tier)

Cost Overview (October 2025 Pricing)

Tier Monthly Limit Cost per 1M chars Neural Voices Custom Neural Real-time/Batch
Free (F0) 500,000 characters Free ✅ Limited neural ❌ Real-time only
Standard (S0) Unlimited $15.00 Neural / $4.00 Standard ✅ All voices ✅ Available Both modes
Premium Unlimited $25.00 Ultra-neural ✅ Premium quality ✅ Advanced Both + priority

Note: Pricing updated as of October 2025. Microsoft has increased neural voice quality and pricing reflects enhanced capabilities.

📋 Step-by-Step Setup

Step 1: Create Azure Account

  1. Visit Azure Portal: https://portal.azure.com
  2. Sign Up: Click "Free account" if you don't have one
  3. Provide Information: Email, phone verification, credit card (for identity verification)
  4. Complete Setup: Follow the guided setup process

Step 2: Create Speech Resource

Option A: Speech-Only Resource (Recommended)

  1. Navigate to Create Resource: Portal home → "Create a resource"
  2. Search for Speech: Type "Speech" in the search box
  3. Select Speech: Choose "Speech" by Microsoft
  4. Click Create: Begin configuration

Option B: Multi-Service Cognitive Services

  1. Create Resource: Portal → "Create a resource"
  2. Search Cognitive Services: Find "Cognitive Services"
  3. Select Multi-Service: Provides access to all cognitive services
  4. Click Create: Begin configuration

Step 3: Configure Resource Settings

Fill out the resource creation form:

Basic Settings

  • Subscription: Select your Azure subscription
  • Resource Group:
    • Create new: tts-resources (recommended)
    • Or use existing group
  • Region: Choose based on your location:
    • East US: Best for North America East Coast
    • West Europe: Best for Europe
    • Southeast Asia: Best for Asia Pacific
    • Australia East: Best for Australia/New Zealand

Resource Details

  • Name: Unique name (e.g., my-company-tts-service)
  • Pricing Tier:
    • F0 (Free): 5,000 transactions/month, standard voices only
    • S0 (Standard): Pay-per-use, all features

Network and Tags (Optional)

  • Network: Leave default (All networks) for simplicity
  • Tags: Add for organisation (optional)

Click Review + Create → Create

Step 4: Get API Credentials

After deployment completes:

  1. Go to Resource: Click "Go to resource"
  2. Keys and Endpoint: Click in left navigation menu
  3. Copy Credentials:
    • Key 1: 32-character hexadecimal string (keep secure!)
    • Location/Region: Note the region code (e.g., "eastus")
    • Endpoint: The base URL for API calls

Security Best Practices

  • Key Security: Never share or commit keys to code repositories
  • Key Rotation: Regenerate keys monthly for production use
  • Least Privilege: Use resource-specific keys, not subscription-wide keys
  • Monitor Usage: Set up billing alerts to track consumption

⚙️ Application Configuration

Initial Setup in TextToSpeech Generator

  1. Launch Application: Run TextToSpeech-Generator-v1.1.ps1
  2. Select Azure Provider: Click "Azure" radio button
  3. Enter API Key: Paste your 32-character key
  4. Select Datacenter: Choose matching region from dropdown
  5. Test Connection: Click in the datacenter field to validate

Voice Selection

The application will automatically load available voices. Popular options:

English (US) - Professional (2025 Voices)

  • en-US-AvaNeural - Modern female voice, professional and warm
  • en-US-AndrewNeural - Modern male voice, confident and clear
  • en-US-AriaNeural - Expressive female voice, natural conversation
  • en-US-BrianNeural - Mature male voice, authoritative tone
  • en-US-ChristopherNeural - Young male voice, friendly and energetic
  • en-US-EmmaNeural - Young female voice, cheerful and engaging
  • en-US-JennyNeural - Versatile female voice, widely used
  • en-US-GuyNeural - Clear male voice, professional standard

English (UK) - British Accent (2025 Update)

  • en-GB-SoniaNeural - Professional British female, RP accent
  • en-GB-RyanNeural - Professional British male, RP accent
  • en-GB-LibbyNeural - Modern British female, friendly tone
  • en-GB-MaisieNeural - Young British female, contemporary accent
  • en-GB-ThomasNeural - Young British male, modern pronunciation

Multi-Language Support

  • fr-FR-DeniseNeural - French female
  • de-DE-KatjaNeural - German female
  • es-ES-ElviraNeural - Spanish female
  • it-IT-ElsaNeural - Italian female

Audio Format Selection

Choose based on your use case:

High Quality (Recommended)

  • riff-24khz-16bit-mono-pcm - Highest quality WAV
  • audio-24khz-48kbitrate-mono-mp3 - High quality MP3

Standard Quality

  • riff-16khz-16bit-mono-pcm - Standard WAV (PSTN compatible)
  • audio-16khz-32kbitrate-mono-mp3 - Standard MP3 (SIP compatible)

Bandwidth Optimised

  • audio-16khz-64kbitrate-mono-mp3 - Balanced quality/size
  • raw-16khz-16bit-mono-pcm - Uncompressed for processing

🔧 Advanced Configuration

Custom Neural Voices

For enterprise customers, Azure offers custom neural voice creation:

  1. Requirements: Minimum 300 sentences of training data
  2. Cost: $2,400 setup fee + hosting costs
  3. Timeline: 4-6 weeks development
  4. Use Cases: Brand-specific voices, celebrity voices, multilingual consistency

Contact Microsoft for custom voice development.

SSML Support

Azure supports Speech Synthesis Markup Language for advanced control:

<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xml:lang="en-US">
    <voice name="en-US-SaraNeural">
        <prosody rate="slow" pitch="low">
            This text will be spoken slowly with a lower pitch.
        </prosody>
        <break time="500ms"/>
        <emphasis level="strong">This text is emphasized.</emphasis>
    </voice>
</speak>

Regional Datacenters (October 2025)

Region Code Location Latency (US East) Neural Voices Best For
eastus Virginia, US ~15ms ✅ Full Support US East Coast
eastus2 Virginia, US ~20ms ✅ Full Support US East Coast (backup)
westus2 Washington, US ~60ms ✅ Full Support US West Coast
westus3 Phoenix, US ~65ms ✅ Full Support US Southwest
centralus Iowa, US ~35ms ✅ Full Support US Central
southcentralus Texas, US ~40ms ✅ Full Support US South
westeurope Netherlands ~100ms ✅ Full Support Europe West
northeurope Ireland ~110ms ✅ Full Support Europe North
uksouth London, UK ~105ms ✅ Full Support United Kingdom
francecentral Paris, France ~115ms ✅ Full Support France
germanywelcentral Frankfurt, Germany ~120ms ✅ Full Support Germany
southeastasia Singapore ~170ms ✅ Full Support Asia Pacific
eastasia Hong Kong ~180ms ✅ Full Support East Asia
japaneast Tokyo, Japan ~160ms ✅ Full Support Japan
australiaeast Sydney, Australia ~190ms ✅ Full Support Australia/NZ
canadacentral Toronto, Canada ~25ms ✅ Full Support Canada
brazilsouth São Paulo, Brazil ~150ms ✅ Full Support South America

Choose the closest region for optimal performance. All regions support the full range of neural voices as of October 2025.

Choose the closest region for best performance.

📊 Usage Monitoring

Azure Portal Monitoring

  1. Resource Overview: View usage statistics
  2. Metrics:
    • Total Calls
    • Data In/Out
    • Latency
    • Error Rate
  3. Alerts: Set up notifications for quota limits
  4. Cost Management: Track spending and set budgets

Application Logging

The TextToSpeech Generator logs all API interactions:

2025-10-10 14:30:15 [INFO] Azure token obtained successfully
2025-10-10 14:30:16 [INFO] Loaded 187 voices from eastus datacenter
2025-10-10 14:30:45 [INFO] Generated: welcome_message (en-US-SaraNeural)
2025-10-10 14:30:47 [ERROR] Rate limit exceeded, retrying in 1 second

🚨 Troubleshooting Azure-Specific Issues

Authentication Errors

"Invalid subscription key" (401 Error)

Causes:

  • Incorrect API key
  • Key for wrong service type
  • Expired or deactivated key

Solutions:

  1. Verify key in Azure Portal → Resource → Keys and Endpoint
  2. Ensure you're using Key1 or Key2 (not endpoint URL)
  3. Check service isn't suspended due to billing issues

"Access denied" (403 Error)

Causes:

  • Insufficient quota
  • Billing issues
  • Service disabled

Solutions:

  1. Check quota in Azure Portal
  2. Verify billing information is current
  3. Confirm service tier supports requested features

Regional Issues

"Region mismatch" errors

Cause: API key region doesn't match selected datacenter

Solution: Ensure datacenter selection matches resource location:

# Check resource location in PowerShell
Get-AzCognitiveServicesAccount -ResourceGroupName "your-rg" -Name "your-resource"

High latency or timeouts

Solutions:

  1. Switch to closer datacenter
  2. Check internet connection stability
  3. Contact Azure support if persistent

Voice Loading Issues

"No voices available"

Causes:

  • Network connectivity issues
  • Invalid authentication
  • Service outage

Diagnostic Steps:

  1. Test manual API call:
$headers = @{"Authorisation"="Bearer $token"}
$uri = "https://eastus.tts.speech.microsoft.com/cognitiveservices/voices/list"
Invoke-RestMethod -Uri $uri -Headers $headers
  1. Check Azure Service Health dashboard
  2. Try different datacenter region

💡 Best Practices

Production Deployment

  1. Use Standard Tier: Free tier has limitations unsuitable for production
  2. Implement Retry Logic: Handle transient failures gracefully
  3. Cache Tokens: Tokens are valid for 10 minutes, reuse when possible
  4. Monitor Quotas: Set up alerts before hitting limits
  5. Multiple Keys: Use key rotation for zero-downtime updates

Performance Optimisation

  1. Batch Requests: Group multiple TTS requests when possible
  2. Regional Deployment: Use multiple regions for global applications
  3. Caching: Cache generated audio for repeated content
  4. Connection Pooling: Reuse HTTP connections for multiple requests

Security Considerations

  1. Key Management: Use Azure Key Vault in production
  2. Network Security: Implement firewall rules and VPN access
  3. Audit Logging: Enable diagnostic logging for compliance
  4. Data Residency: Consider region for data sovereignty requirements

📞 Support and Resources

Microsoft Support

Community Resources

Application Support

Next Steps: After setting up Azure, refer to the main README.md for application usage instructions or CSV-FORMAT.md for bulk processing guidance.