Skip to main content

Hands-On Lab: Amazon Polly Voice Synthesis, Engine Selection & SSML Markup

Key Takeaways​

Amazon Polly converts arbitrary text input into high-fidelity synthetic audio directly within the AWS Management Console using deep learning and generative foundation models.

The interactive console provides immediate hands-on controls for testing Engine Tiers (Generative, Long-form, Neural, Standard), selecting diverse regional voices/personas (such as Ruth or Gregory), wrapping phrases in SSML (Speech Synthesis Markup Language) tags like <speak> and <break time="..."/> to introduce custom pauses, and adjusting pronunciation lexicons and audio export formats.

Hands-On Workflow: Voice Synthesis, Engine Selection & SSML Pacing​

  1. Open the Amazon Polly Text-to-Speech Console:

    • Navigate to the Amazon Polly Console in your target AWS Region.
    • In the left navigation menu, click Text-to-Speech (or click Try Polly from the landing dashboard).
    • Locate the interactive input text area and the engine configuration panel on the right.

  2. Select and Compare Synthesis Engines & Voices:

    • Evaluate different voice personalities and engine capabilities:
    • Select an Engine:
      • Generative: Highly expressive, adaptive tone driven by foundation model architectures.
      • Long-form: Optimized for natural breathing rhythm and pacing across multi-sentence text.
      • Neural (NTTS): Natural, conversational prosody.
      • Standard: Cost-effective parametric synthesis.
    • Choose a Language and Voice persona (e.g., English, Australian →\rightarrow Olivia or Russell).
    • Enter a test string: Hi, I'm Olivia. I can read any text for you. Test it out!
    • Click Listen to audition the voice in real time.

  3. Enable Speech Synthesis Markup Language (SSML):

    • Add fine-grained timing and cadence control to the synthesized voice:
    • Under the input editor, switch from Plain text mode to SSML mode.
    • Wrap the entire text payload inside opening and closing <speak> tags.
    • Insert an explicit break element (<break time="..."/> or self-closing <break/>) between words where a natural pause is required:
    <speak>
      Hi, I'm Rendy and I'm learning <break time="1s"/> AWS, it's awesome!
    </speak>
    • Click Listen and verify that the synthesis engine inserts an audible silence between "learning" and "AWS".

  4. Explore Additional Settings & Output Formats:

    • Inspect production export and customization parameters:
    • Open Additional settings to adjust the output format (MP3, OGG Vorbis, or PCM).
    • Under Custom pronunciation, attach a configured Pronunciation Lexicon (PLS XML) to enforce custom brand or name pronunciations.
    • Click Download to save the generated audio file locally, or select Synthesize to S3 for asynchronous long-form generation.

Polly Synthesis Engines: Console Feature Matrix​

Engine OptionLatency & Expressiveness ProfileSSML SupportTarget Workload Demonstrated
GenerativeMaximum emotional range, conversational adaptation, and inflection.Full SSML supportVirtual assistants, dynamic character voiceovers, customer support.
Long-formNatural pacing, paragraph pauses, and consistent tone across extended text.Full SSML supportAudiobooks, long-form articles, training lectures, news summaries.
Neural (NTTS)High prosody and human-like naturalness.Extensive SSML supportIVR call centers, mobile applications, notifications.
StandardBasic concatenative synthesis, lowest cost per character.Standard SSML tagsLegacy alerts, high-volume automated status messaging.

Exam Guide​

Exam Tips​

  • SSML Syntax Rules: Every SSML payload sent to Amazon Polly must start with <speak> and end with </speak>. Individual modifier tags like <break/>, <emphasis>, or <whisper> reside within this root block.
  • Pause Insertion: The tag to introduce explicit pauses or silences in Amazon Polly speech is <break time="..."/> (e.g., <break time="2s"/> or <break strength="medium"/>).
  • Long Content Generation: For synthesizing large text documents that exceed synchronous API payload limits, use the StartSpeechSynthesisTask API to render speech asynchronously directly to an Amazon S3 bucket.
  • Audio Format Types: Amazon Polly can export audio in MP3, OGG Vorbis, and raw uncompressed PCM format for telephony/WAV encoding.

Practice Test​

Question 1​

A developer is testing Amazon Polly in the AWS Management Console to create an automated prompt for a contact center. The prompt must pause for 1.5 seconds between reciting two separate instructions. Which configuration should the developer apply?

  • A. Upload a Custom Language Model (CLM) with pause annotations
  • B. Switch the console editor to SSML mode, wrap the text in <speak> tags, and insert <break time="1.5s"/> between the instructions
  • C. Enable Automatic Multi-Language Identification
  • D. Apply an Amazon Translate parallel data glossary
Correct Answer
  • B. Switch the console editor to SSML mode, wrap the text in <speak> tags, and insert <break time="1.5s"/> between the instructions
    • Explanation: In Amazon Polly, Speech Synthesis Markup Language (SSML) is used to adjust speech pacing and timing. The <break time="1.5s"/> tag inside a <speak> block instructs the synthesizer to pause for 1.5 seconds at that exact position.

Question 2​

An online media publisher wants to automatically convert multi-page editorial articles and blog posts (averaging 3,000 words each) into high-quality spoken audio with natural breathing rhythms and conversational pacing. Which Amazon Polly engine is specifically tailored for this use case?

  • A. Standard Engine
  • B. Long-form Engine
  • C. Amazon Transcribe ASR Engine
  • D. Amazon Comprehend Syntax Engine
Correct Answer
  • B. Long-form Engine
    • Explanation: The Amazon Polly Long-form Engine is specifically designed for multi-paragraph articles, editorial content, and audiobooks, providing natural breathing rhythms and engaging pacing across long passages of text.