AWS Polly TTS Quick Start

AWS Polly is Amazon's text-to-speech service. It's cost-effective (~$4 per 1 million characters) and works well for development and production workloads.

Prerequisites

1. AWS Account with Polly access

2. AWS Credentials configured via one of:

   • Environment variables: AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
   • AWS credentials file: ~/.aws/credentials
   • IAM role (if running on EC2/ECS)

3. Python boto3 library installed:

   pip install boto3

Configuration

1. Add AWS Polly to your config

Edit .babulus/config.yml:

providers:
  aws_polly:
    region: "us-east-1"
    voice_id: "Joanna"
    engine: "standard"  # or "neural" for better quality

2. Set provider in your DSL

Edit your .babulus.yml file:

voiceover:
  provider: aws-polly  # or just "aws"
  sample_rate_hz: 16000  # Required: Polly PCM only supports 8000 or 16000 Hz

Important: AWS Polly's PCM format only supports 8000 Hz or 16000 Hz sample rates. Use 16000 Hz for best quality.

3. Environment-Based Configuration (Recommended)

For multi-environment workflows:

voiceover:
  provider:
    development: openai
    aws: aws           # Use Polly for AWS testing
    production: elevenlabs

  sample_rate_hz:
    development: 24000  # OpenAI
    aws: 16000          # Polly PCM maximum
    production: 44100   # ElevenLabs

Then generate with:

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

Available Voices

Standard Engine (Cheaper)

• English (US): Joanna, Matthew, Ivy, Kendra, Kimberly, Salli, Joey, Justin
• English (UK): Amy, Emma, Brian
• Many other languages: See AWS Polly Voices

Neural Engine (Better Quality)

• English (US): Joanna, Matthew, Ivy, Kendra, Kimberly, Salli, Joey, Justin, Kevin, Ruth, Stephen
• Requires engine: "neural" in config
• Slightly higher cost

Voice Selection

You can override the voice per DSL:

voiceover:
  provider: aws
  voice: Matthew  # Override default voice
  sample_rate_hz: 16000

Or use the default from config.

Pricing

• Standard voices: 4.00per1millioncharacters( 0.004/1K chars)
• Neural voices: 16.00per1millioncharacters( 0.016/1K chars)
• Much cheaper than ElevenLabs (~$0.30/1K chars)
• More expensive than OpenAI ($0.015/1K chars)

Troubleshooting

Error: "AWS Polly PCM only supports sample_rate_hz [8000, 16000]"

Solution: Set sample_rate_hz: 16000 in your voiceover config.

Error: "Unable to locate credentials"

Solutions:

1. Set environment variables:

   export AWS_ACCESS_KEY_ID=your_key_id
   export AWS_SECRET_ACCESS_KEY=your_secret_key

2. Or create ~/.aws/credentials:

   [default]
   aws_access_key_id = your_key_id
   aws_secret_access_key = your_secret_key

3. Or use IAM roles if running on AWS infrastructure

Low Audio Quality

Solution: Switch to neural engine for better quality:

providers:
  aws_polly:
    engine: "neural"

Note: Neural voices cost 4x more than standard but sound significantly better.

Comparison with Other Providers

Example: Complete DSL

voiceover:
  provider: aws
  voice: Matthew
  sample_rate_hz: 16000
  seed: 1337
  lead_in_seconds: 0.25

scenes:
  - id: intro
    title: "Introduction"
    cues:
      - id: welcome
        label: "Welcome"
        voice: "Welcome to our presentation. Let's get started."

AWS Polly Features Not (Yet) Supported

• SSML tags for prosody control
• Speech marks for word-level timing
• Lexicons for pronunciation customization

These may be added in future releases. For now, Babulus uses plain text synthesis.