Amazon Polly: Text-to-Speech (TTS), Engines, SSML & Custom Lexicons

Key Takeaways​

Amazon Polly is a fully managed, serverless Text-to-Speech (TTS) service that converts written text into lifelike spoken audio using advanced deep learning and generative AI voice synthesis engines.

Amazon Polly provides four distinct voice engines (Standard, Neural, Long-Form, and Generative), fine-grained speech control via Speech Synthesis Markup Language (SSML), custom pronunciation rules via Custom Lexicons, and temporal synchronization metadata via Speech Marks for visual lip-syncing and word-highlighting animations.

Main Discussion​

The Four Voice Engine Tiers​

Speech Synthesis Markup Language (SSML)​

SSML is an XML-based standard that allows fine-grained customization of generated audio pronunciation, volume, pacing, and tone:

<speak>
     Hello! <break time="1s"/> Welcome to the training course.
     <emphasis level="strong">Pay close attention</emphasis> to this section.
     <amazon:effect name="whispered">This is a confidential secret.</amazon:effect>
     I will spell this out: <say-as interpret-as="spell-out">AWS</say-as>.
</speak>

Custom Lexicons vs. Speech Marks​

Speech Marks JSON Metadata Output Structure​

{"time": 450, "type": "word", "start": 12, "end": 19, "value": "Rendy"}
{"time": 450, "type": "viseme", "value": "s"}
{"time": 520, "type": "viseme", "value": "t"}

Exam Guide​

Exam Tips​

• Core Service Direction:
  • Amazon Transcribe: Spoken Audio $\rightarrow$ Written Text (Speech-to-Text).
  • Amazon Polly: Written Text $\rightarrow$ Spoken Audio (Text-to-Speech).
• Pronunciation Customization: If an exam question asks how to ensure Polly expands acronyms (e.g., "W3C" to "World Wide Web Consortium") or correctly pronounces unusual industry jargon, select Custom Lexicons (PLS format).
• Fine-Grained Speech Tuning: If the requirement is adding pauses, whispering, adjusting pitch, or emphasizing specific words, the answer is SSML (Speech Synthesis Markup Language).
• Lip-Syncing & Animations: Any requirement for lip-syncing 3D avatars, animated mouth movements, or highlighting spoken words in real time maps directly to Amazon Polly Speech Marks (visemes).
• Engine Selection Rules:
  • Multi-chapter audiobooks / long articles $\rightarrow$ Long-Form Engine.
  • Highly conversational, emotionally engaging voice $\rightarrow$ Generative Engine.
  • Standard cost-effective alerts $\rightarrow$ Standard / Neural Engine.

Sample AIF-C01 Questions​

Question 1​

An educational technology company is creating a 3D animated virtual teacher for a mobile learning app. The app converts written lesson scripts into spoken audio using Amazon Polly. The graphics team needs precise timestamp data indicating when specific facial and mouth shapes (visemes) occur in the synthesized audio to animate the avatar's lips in real time. Which Amazon Polly feature should the engineering team use?

• A. Amazon Polly Custom Lexicons
• B. Amazon Polly Speech Marks
• C. Amazon Transcribe Speaker Diarization
• D. Amazon Comprehend Syntax Tokenization

Question 2​

A developer is configuring Amazon Polly for a telecommunications customer portal. The portal must convert written account alerts into audio messages. When the text contains the abbreviation "VoIP", the portal must pronounce it as "Voice over Internet Protocol" instead of reading the letters individually. Which configuration accomplishes this?

• A. Train an Amazon Comprehend Custom Classifier
• B. Define a Custom Lexicon using Pronunciation Lexicon Specification (PLS) XML and apply it to synthesis requests
• C. Enable Automatic Multi-Language Identification in Amazon Transcribe
• D. Apply an Amazon Translate parallel data file