Vision & speech

Rekognition is AWS's vision service: you hand it an image and it tells you what's in the picture. This block points it at a photo sitting in an S3 bucket (AWS's file storage) and asks two questions — "what objects are here?" and "is this image safe?". Nothing is trained by you; the model already knows how to recognise common things.

1. boto3.client("rekognition", region_name="us-east-1") creates the connection object (AWS calls it a client) that lets Python send commands to Rekognition in the us-east-1 data-centre region. boto3 is the official AWS library for Python.
2. img = {"S3Object": {"Bucket": "my-media", "Name": "photo.jpg"}} does not upload anything — it just says "the picture is the file photo.jpg inside the bucket named my-media". You point at where the file already lives.
3. rek.detect_labels(Image=img, MaxLabels=5) asks "what do you see?" and caps the answer at the 5 most confident guesses. Each guess ("label") comes with a confidence score from 0–100.
4. The print([(l["Name"], round(l["Confidence"], 1)) ...]) line loops over the labels and builds a tidy list of (name, confidence) pairs, rounding the score to one decimal.
5. rek.detect_moderation_labels(Image=img) is a separate check for unsafe content (violence, nudity, etc.). It returns a list of flags — useful as a guardrail before you let a user's upload through.

What the output means: The first line lists what Rekognition saw, most-confident first — e.g. ('Truck', 99.4) means it is 99.4% sure the photo contains a truck. flagged: [] is an empty list, meaning nothing was flagged as unsafe — the image passed the moderation check.

Try this: Change MaxLabels=5 to MaxLabels=2 and you'll get only the top two guesses. A non-empty flagged list is your cue to reject or hold the upload for review.

This block chains two services. Transcribe turns recorded speech into written text (speech-to-text). Translate then rewrites text from one language into another. Transcribe is asynchronous: instead of answering instantly, you start a job, it works in the background, and you check back later — the way you might drop clothes at a laundromat and return when they're done.

1. ts.start_transcription_job(...) kicks off the job. TranscriptionJobName="call-001" is a label you invent so you can find this job again later. Media={"MediaFileUri": "s3://..."} tells it which audio file (in S3) to transcribe.
2. MediaFormat="mp3" and LanguageCode="en-US" tell it the file is an MP3 and the speech is US English — hints that make transcription more accurate.
3. The comment line # poll get_transcription_job until COMPLETED marks the step you'd normally add: polling means repeatedly asking "done yet?" in a loop until the status reads COMPLETED, then reading the finished transcript (returned as a JSON file in S3).
4. tr = boto3.client("translate", ...) opens a second client, this time for Translate.
5. tr.translate_text(Text="The database is down.", SourceLanguageCode="en", TargetLanguageCode="es") asks it to turn that English sentence (en) into Spanish (es). The result lives under the "TranslatedText" key of the reply.

What the output means: La base de datos está caída. — the Spanish translation of "The database is down." Only the Translate step prints here; the Transcribe job runs in the background and is fetched separately.

Try this: Change TargetLanguageCode="es" to "fr" (French) or "de" (German) and re-run just the Translate part. The source stays English; only the target language changes.

Polly is the mirror image of Transcribe: it turns written text into spoken audio (text-to-speech). This block hands Polly a sentence and saves the spoken version as an MP3 file you can play.

1. polly = boto3.client("polly", region_name="us-east-1") opens the connection to the Polly service.
2. polly.synthesize_speech(Text="Deployment complete...", OutputFormat="mp3", VoiceId="Joanna") is the core call: give it the words, ask for MP3 audio, and pick a voice — Joanna is one of Polly's built-in US-English voices. It returns the audio as a stream of bytes, not a file yet.
3. with open("status.mp3", "wb") as f: opens a new file for writing in binary mode ("wb") — binary because audio is raw bytes, not text. The with block automatically closes the file when done.
4. f.write(audio["AudioStream"].read()) reads all the audio bytes Polly sent back and writes them into status.mp3 on your disk.

What the output means: It prints wrote status.mp3 and leaves a real, playable status.mp3 file in your folder that says "Deployment complete. All systems healthy." in Joanna's voice.

Try this: Swap VoiceId="Joanna" for "Matthew" (a male US voice) and re-run — you'll get the same words in a different voice. Change the Text= to hear your own message.