ReadWhatYouSee

Read What You See

Read What You See chains three AWS services together: load an image, click on any text in it, and the app will extract it (Textract), translate it (Translate), and read it aloud (Polly).

What's demonstrated

Three services, one interaction:

• Amazon Textract — TTextractClient.DetectDocumentText to extract text and bounding boxes from the image
• Amazon Translate — TTranslateClient.TranslateText with automatic source language detection
• Amazon Polly — TPollyClient.DescribeVoices and SynthesizeSpeech to speak the translated text

Architecture

Each AWS service lives in its own DataModule, keeping the main form focused on UI and coordination:

Unit	Responsibility
DataModules.TextExtraction	Calls Textract, stores detected blocks, performs hit-testing by coordinate
DataModules.Translation	Translates text with automatic source language detection
DataModules.TextToSpeech	Manages Polly voices and synthesizes speech for the target language
Forms.Main	Coordinates the above: image display, click handling, language selection

The flow

User clicks text in image
        |
        v
TextExtractionDM.SelectTextAt(point)     -- hit-test against Textract bounding boxes
        |
        v
TranslationDM.TranslateText(selected)    -- translate to target language
        |
        v
TextToSpeechDM.ConvertTextToSpeech(text) -- synthesize and play audio

Running the sample

1. Open "ReadWhatYouSee.dproj" in Delphi or RAD Studio.
2. Select "Run > Run" from the menu or press F9.
3. Click "Open file..." and select an image containing text.
4. Once the image has loaded, click on any text in the image to select it.
5. Choose the output language from the drop-down on the top-left.
6. The selected text will be translated and spoken aloud automatically.

Required IAM permissions

• textract:DetectDocumentText
• translate:TranslateText
• polly:DescribeVoices
• polly:SynthesizeSpeech