#TGIFun🎈 Building GenAI apps with managed AI services

"""
Describe for Me (Bedrock edition ⛰️)

Image ==> Bedrock -> Translate -> Polly ==> Speech
"""

import argparse
import base64
import json

import boto3

# Parse arguments
parser = argparse.ArgumentParser(
prog='Describe for Me (Bedrock Edition) ⛰️',
description="""
Describe for Me is an 'image-to-speech' app
built on top of AI services that was created
to help the visually impaired understand images
through lively audio captions.
""",
)
parser.add_argument('image')
parser.add_argument('-m', '--model', default="anthropic.claude-3-sonnet-20240229-v1:0")
parser.add_argument('-t', '--translate')
parser.add_argument('-v', '--voice')
args = parser.parse_args()

# Initialize clients
bedrock = boto3.client("bedrock-runtime")
translate = boto3.client("translate")
polly = boto3.client("polly")

# Process image
with open(args.image, "rb") as image_file:
image = base64.b64encode(image_file.read()).decode("utf8")

# Generate a description of the image
response = bedrock.invoke_model(
modelId=args.model,
body=json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [{
"role": "user",
"content": [{
"type": "text",
"text": "Describe the image for me.",
},
{
"type": "image",
"source": {
"type": "base64",
"media_type": f"image/{args.image.split('.')[-1]}",
"data": image,
},
}],
}],
}),
)

result = json.loads(response.get("body").read())
description = " ".join([output["text"] for output in result.get("content", [])])

print(f"##### Original #####\n\n{description}\n")

# Translate the description to a target language
if args.translate:
description = translate.translate_text(
Text=description,
SourceLanguageCode='en',
TargetLanguageCode=args.translate
)['TranslatedText']

print(f"##### Translation #####\n\n{description}\n")

# Read the description back to the user
if args.voice:
# Translate -> Polly language map
lang_map = {
'en': 'en-US',
'pt': 'pt-BR'
}
lang_code = lang_map.get(args.translate, args.translate)

# Synthesize audio description
print(f"\nSynthesizing audio description\n> Language: {lang_code}\n> Voice: {args.voice}\n")
audio_data = polly.synthesize_speech(
Engine="neural", # hardcoded, because we want natural-sounding voices only
LanguageCode=lang_code,
OutputFormat="mp3",
Text=description,
TextType='text',
VoiceId=args.voice,
)

# Save audio description
with open(f"{''.join(args.image.split('.')[:-1])}.mp3", 'wb') as audio_file:
audio_file.write(audio_data['AudioStream'].read())

python describeforme.py --translate pt-PT --voice Ines seurat.jpeg

##### Original #####

The image depicts a park scene from the late 19th or early 20th century.
It's a lush green landscape with trees and people gathered enjoying the outdoors.
Many are dressed in fashions typical of that era, with men wearing top hats
and women in long dresses with parasols to shade themselves from the sun.

Some people are seated on the grass, while others are strolling along the paths.
There's a body of water in the background, likely a lake or river, with boats
and people gathered near the shoreline.

The style of painting is distinctive, with areas of flat color and simplified forms,
characteristic of the Post-Impressionist or early modern art movements.

The scene captures a leisurely day in an urban park setting, conveying a sense
of relaxation and appreciation for nature amidst the city environment.

##### Translation #####

A imagem retrata uma cena de parque do final do século XIX ou início do século XX.
É uma paisagem verdejante com árvores e pessoas reunidas a desfrutar do ar livre.
Muitos estão vestidos com modas típicas daquela época, com homens a usar chapéus
e mulheres em vestidos longos com guarda-sóis para se proteger do sol.

Algumas pessoas estão sentadas na relva, enquanto outras estão a passear pelos caminhos.
Há um corpo de água ao fundo, provavelmente um lago ou rio, com barcos
e pessoas reunidas perto da costa.

O estilo de pintura é distinto, com áreas de cor plana e formas simplificadas,
características dos movimentos de arte pós-impressionista ou do início da arte moderna.

A cena captura um dia de lazer num cenário de parque urbano, transmitindo uma sensação
de relaxamento e apreciação pela natureza em meio ao ambiente da cidade.

Synthesizing audio description
> Language: pt-PT
> Voice: Ines

%% Use the mermaid-cli
%% > mmdc -i meh.mmd -o meh.png -t forest -b transparent
%% or the online version
%% https://mermaid.live
%% to create the diagram
sequenceDiagram
You 😒 ->> +MEH! 🐑 : Prompt ✍️
MEH! 🐑 ->> +Amazon Translate : translate_text 🌐
Amazon Translate ->> MEH! 🐑 :
MEH! 🐑 ->> You 😒 : Translated Prompt
MEH! 🐑 ->> Amazon Bedrock: invoke_model 🧠
Amazon Bedrock ->> MEH! 🐑 :
MEH! 🐑 ->> You 😒 : Response
MEH! 🐑 ->> +Amazon Translate : translate_text 🌐
Amazon Translate ->> MEH! 🐑 :
MEH! 🐑 ->> You 😒 : Translated Response
MEH! 🐑 ->> Amazon Polly : synthesize_speech 🦜
Amazon Polly ->> MEH! 🐑 :
MEH! 🐑 ->> You 😒 : Play 🔊

streamlit run meh.py

docker build --rm -t meh .

docker run --rm --device /dev/snd -p 8501:8501 meh

wsl docker run --rm -e PULSE_SERVER=/mnt/wslg/PulseServer -v /mnt/wslg/:/mnt/wslg/ -p 8501:8501 meh