Hearing and Speaking

In the world of AWS AI, audio processing is split into two distinct paths.

1. Speech-to-Text: Listening to a recording and writing it down (Amazon Transcribe).
2. Text-to-Speech: Taking a text string and turning it into a natural-sounding voice (Amazon Polly).

For the AWS Certified AI Practitioner exam, you must understand the specialized features of both and, most importantly, never confuse the two names.

---

1. Amazon Transcribe (The Listener)

Amazon Transcribe uses deep learning to provide high-quality speech-to-text. It doesn't just write words; it understands the structure of a conversation.

Key Features:

• Speaker Identification (Diarization): It can tell when "Speaker 1" stops and "Speaker 2" starts. (Crucial for meeting notes).
• Custom Vocabulary: You can tell it what your company's product names or technical jargon sound like so it doesn't get them wrong.
• Redaction: It can automatically remove sensitive data (like a credit card number spoken over a phone call).
• Subtitling: It can output a time-coded file (.srt) for video subtitles.

Industry Variation: Amazon Transcribe Medical

Just like Comprehend Medical, this version is trained on terms like "Metformin" or "Clarithromycin," allowing doctors to dictate notes hands-free.

---

2. Amazon Polly (The Speaker)

Amazon Polly turns text into lifelike speech. It is the "Voice" for your applications.

Key Features:

• Neural TTS (NTTS): High-fidelity voices that sound like a human person, with natural breathing and rhythm.
• SSML (Speech Synthesis Markup Language): You can "code" the voice. You can tell Polly to whisper, to <emphasis>shout</emphasis>, or to take a specific pause.
• Speech Marks: It can tell your app exactly when it is pronouncing a specific word—perfect for syncing an avatar's lips or highlighting text as it's read.
• Lexicons: You can teach Polly how to pronounce a specific acronym (e.g., "AWS") so it says "Amazon Web Services" or "A-W-S" depending on your choice.

---

3. The Functional Distinction

graph LR
    subgraph Audio_Processing
    A[Raw Audio File / Stream] -->|TRANSCRIBE| B[Text Document]
    B -->|POLLY| C[Digital Voice .mp3]
    end
    
    subgraph Use_Cases
    D[Call Center Analytics]
    E[Video Subtitles]
    F[Handbooks for the Blind]
    G[Podcast Generation]
    end
    
    B --> D
    B --> E
    C --> F
    C --> G

The "A-ha!" Moment for the Exam:

• If the question asks to "Support users with visual impairments" -> Use Amazon Polly (to read to them).
• If the question asks to "Analyze customer complaints on phone calls" -> Use Amazon Transcribe first (to get the text) and then Amazon Comprehend (to find the sentiment).

---

4. Summary: The Audio Loop

• Polly is a Synthetic Voice.
• Transcribe is an Automatic Scribe.
• Together, they allow for high-level automated phone systems (IVRs) and accessibility tools.

---

Exercise: The Strategic Chain

A hospital wants to build a system where a doctor can:

1. Record a voice memo of a patient's symptoms.
2. Have that memo turned into a text note in the system.
3. Have the text note specifically highlight any "Dangerous" medical conditions.

Which three services are needed in this order?

• A. Amazon Polly -> Amazon Transcribe -> Amazon Bedrock.
• B. Amazon Transcribe Medical -> Amazon Comprehend Medical.
• C. Amazon Rekognition -> Amazon Textract -> Amazon Translate.
• D. Amazon Transcribe -> Amazon Polly -> Amazon QuickSight.

The Answer is B! Transcribe Medical converts the voice to text. Comprehend Medical "Highlights" the specific medical conditions (Entities).

---

Knowledge Check

What's Next?

We’ve covered Audio. But what about the millions of paper documents and messy PDFs in the world? In our next lesson, we master Amazon Textract.