AWS · © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. 2-H1-1-18 AWS

© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.

2-H1-1-18

AWS


❏

❏

❏ AWS

❖ Amazon EMR

❖ Amazon ECS

❖ Amazon SageMaker


❏Amazon

❏AWS ( )

❏

API


Amazon


Amazon

•

•

• Alexa

•

•

•

•

•

• Q&A

•

•

•

•

•

•

• /

20



AWS


AWS ML

Services

Platform

Engines

CPU IoT EdgeInfrastructure GPU

AWS Deep Learning AMI (Amazon Linux / Ubuntu / Windows)

Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon

KinesisAWS Batch Amazon ECS

Amazon

SageMaker

TensorFlowCaffe2 TorchPyTorchCNTK KerasApacheMXNetChainer Gluon

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


AWS ML

Services

Platform

Engines

Infrastructure CPU IoT EdgeGPU


Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


AWS ML

ServicesAmazon

Rekognition

Image

Platform

Engines



Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video

Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker



AWS ML

Services

Platform

Engines



Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


AWS ML

Services

Platform

Engines



Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


AWS ML

Services

Platform

Engines



Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


※ fine-tuning AWS

API

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


Amazon Rekognition Image

Amazon Rekognition Video

Text-to-Speech Amazon Polly

Speech-to-Text Amazon Transcribe

Amazon Comprehend

Amazon Translate

Amazon Lex

AWS ML Services Portfolio

※ New ! re:Invent 2017






Amazon Comprehend

Amazon Translate

Amazon Lex








Amazon Comprehend

Amazon Translate

Amazon Lex




API

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video



Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video



❏

❏ API

❏

❏

❏


Rekognition -

❏

❏ 0 100

Maple

Plant

Villa

Garden

Water

Swimming Pool

Tree

Potted Plant

Backyard

❖

❖


Rekognition -

❏

❖

❖

❖


Rekognition -

❖

❖

❏ 2


Rekognition -

❖

❖

❏ 2


❏

Rekognition -


❏

Rekognition -

( DB)


❏

Rekognition -

( DB)

{ ; }

JSON


Rekognition -

❏ 1 15 100

❏

❏ API


Rekognition -

❏

❏

❏ - Explicit Nudity

❖

❖

❖

❖

❖

❏ - Suggestive

❖

❖

❖


Rekognition -

❏

❏ IMDb


Rekognition -

❏

❏ 50

1 ISO

❏ (@ / $ % - _ + * # )



Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video




❖ JSON

❖ “Timesamp”

❖ JSON


3


❏ S3

❖

❖

❖

❖

❖

❖

❖

❏ H.264 .mp4 / .mov



:


:

Amazon Kinesis Video Streams


Rekognition -


Amazon Polly

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


❏

(Text-To-Speech)

❏

Amazon Polly


Amazon Polly – Text-to-Speech


Market grew by > 20%.



Market grew by > 20%.Market grew by more

thantwentypercent



Market grew by > 20%.Market grew by more

thantwentypercent




{ { { { {ˈtwɛn.tipɚ.ˈsɛnt

ˈmɑɹ.kət ˈgɹu baɪ ˈmoʊɹ ˈðæn

Market grew by morethan

twentypercent







twentypercent







twentypercent







twentypercent







twentypercent



❏

(Text-To-Speech)

❏

❏

❏ SSML* Lexicon**

❖ SSML

❖ Lexicon ( )

Amazon Polly

*SSML(Speech Syntheis Language) **Lexicon( )

<lexeme>

<grapheme>Kaja</grapheme>

<grapheme>kaja</grapheme>

<grapheme>KAJA</grapheme>

<phoneme>"kaI.@</phoneme>

</lexeme>

My daughter’s name is Kaja.<speak>

<amazon:effect name="whispered">

</amazon:effect>

</speak>


Polly –

Viseme

1

1

❑ ( )


https://youtu.be/T0RMDveyY_g?t=636

Polly – Demo:


Polly – 25 52

( ) Naja Mads

( ) Raveena Aditi

Seoyeon

Mizuki Takumi

/ / : :/ :

( ) Chantal

( ) Vitória Ricardo

( ) Joanna Joey

Salli Justin

Kendra Matthow

Kimberly

Ivy

( ) Penélope Miguel

Naja Mads

Lotte Ruben

Céline Mathieu

Marlene Hans

Vicki

Dóra Karl

Carla Giorgio

Liv

Ewa Jacek

Maja Jan

( ) Inês Cristiano

Carmen

Tatyana Maxim

( ) Conchita Enrique

Astrid

Filiz

( ) Amy Brian

Emma

( ) Geraint

Gwyneth


Polly –


Amazon Transcribe

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


Amazon Transcribe

API( ) -

❏ (Speech-To-Text)

(ASR)

Hello

/Hola


Transcribe -

1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0

“Hi, I would like to reschedule my flight to Seattle to later tonight”

3.480 sec

Confidence : 1

6.402 sec

Confidence : 0.95

10.541 sec

Confidence : 1


Transcribe -

please extrapolate the projections based on market

growth and segment share can you email it to me once

you are done

Please extrapolate the projections based on market

growth and segment share. Can you email it to me once

you’re done?


Transcribe -

Deep Learning based

Speech Recognition

Pre-Processing

Post-Processing

Audio

Transcription

Amazon

Transcribe


:

Transcribe

meetings

(Transcribe) (Amazon ES)(Amazon S3)


Amazon Comprehend

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


Amazon Comprehend

❏

(NLP)

❏

❏


Comprehend -


Comprehend -


Comprehend -


Comprehend -

❏ Amazon S3 ( )

❖ ( )

❖


:


:

Twitter Comprehend

Amazon Redshift


:

Audio Input

Amazon

S3

Amazon

Lambda

Amazon

Transcribe

Speech-To-Text

Amazon

Comprehend

Amazon

QuickSight

Amazon

Athena


Amazon Translate

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


Amazon Translate -

HTML

30

<500 ms

⇄

IAM SSL S3

AWS

CLI

SDK


Amazon Translate


Amazon Lex

Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


Amazon Lex

❏

( )

❏ Amazon Alexa

❖(ASR)

❖(NLU)

❏

Facebook Messenger

Slack Kik Twilio SMS


Lex - :

BookHotelIntents

intent


Lex - :

BookHotelIntents

intent

Utterances

intent


Lex - : Utterances

I’d like to book a hotel

I want to make my hotel reservations

I want to book a hotel in New York City

Can you help me book my hotel?


Lex - : Slots

Destination AMAZON.US_CITY New York City, Seattle, London, …

Slot Type Values

CheckIn AMAZON.DATE Valid dates

CheckOut AMAZON.DATE Valid dates


Lex - : Prompt

I’d like to book a hotel

What date do you check in?

New York City

Sure what city do you want to book?

Nov 30th Check In

11/30/2016

City

New York City


Lex - : Fullfillment

AWS Lambda

Integration

Intent slot Lambda

Return to Client

intent slot


Lex - :

BookHotelIntents

intent

Fulfillment

intent

Utterances

intent

Slots

Slot intent

Prompt

Slot


Lex -


Lex - : “BookHotel”

“Book a Hotel in

NYC”

BookHotel

City

CheckIn

CheckOut

(NLU)

Book Hotel

NYC

(ASR)

Book a Hotel

New York City

a

in

Intent/Slot

New York City

Utterance



BookHotel

City New York City

CheckIn

CheckOut“Can I go ahead

with the booking?

Prompt

“When would you check in ?”

No

“When would you

check in ?”Polly(TTS)



“November 30th“

BookHotel

City New York City

CheckIn

CheckOut

“When would you

check in ?”Polly(TTS)

Prompt

“When would you check in ?”



“November 30th“

BookHotel

City New York City

CheckIn

CheckOut

(NLU)(ASR)

Hotel Booking

November 30thNovember 30th

Intent/Slot

November 30th



BookHotel

City New York City

CheckIn


with the booking?

Prompt

“When would you check out ?”

No

November 30th

“When would you

check out ?”Polly(TTS)



BookHotel

City New York City

CheckIn


with the booking?

No

“December 2nd“

November 30th

“When would you

check out ?”Polly(TTS)

Prompt

“When would you check out ?”



“December 2nd“

BookHotel

City New York City

CheckIn

CheckOut

(NLU)(ASR)

Hotel Booking

December 2ndDecember 2nd

Intent/Slot

December 2nd

November 30th



BookHotel

City New York City

CheckIn

CheckOut December 2nd

“Your hotel is booked

for Nov 30th” Confirmation: “Your hotel is

booked for Nov 30th”

November 30th

Yes

Polly(TTS)

“Can I go ahead

with the booking?

Fullfillment

( )


Lex - 8kHz

❏

❏

❖ Amazon Connect


Lex - bot Alexa Skill Kit export

❏ Lex Alexa Skill Kit JSON

❖ JSON


AWS ML

Services

Platform

Engines



Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


AWS ML

Services

Platform

Engines



Amazon

Machine

Learning

Apache Spark

Amazon EMR

Amazon


Amazon

SageMaker


Amazon

Rekognition

Image

Amazon

Polly

Amazon

Lex

Amazon

Comprehend

Amazon

Translate

Amazon

Transcribe

Amazon

Rekognition

Video


IoT / AWS DeepLens


AWS DeepLens

•

ML

•

DL

• AWS Lambda

• Amazon SageMaker

• Apache MXNet / Tensorflow / Caffe


• Intel Atom Processor

• Gen9 graphics

• Ubuntu OS- 16.04 LTS

• 100 GFLOPS performance

• Dual band Wi-Fi

• 8-GB RAM

• 16-GB storage (eMMC)

• 32-GB SD card

• 4 MP camera with MJPEG

• H.264 encoding at 1080p resolution

• 2 USB ports

• Micro HDMI

• Audio out

• AWS Greengrass preconfigured

• clDNN-optimized for MXNet

• Key Differentiators/Technologies

• Intel cLDNN Library optimized for MXNet

• Intel Deep Learning Deployment Toolkit

AWS DeepLens Specifications


AWS DeepLens

HD video camera

Custom-designed

deep learning

inference engine

Micro-SD

Mini-HDMI

USB

USB

Reset

Audio out

Power

HD

Amazon SageMaker

AWS Lambda

10


A m a z o n S a g e M a k e r

A W S D e e p L e n s

AWS DeepLens 10



❏ Amazon

❏ AWS 4

❏ API


Thank you!

AWS · © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. 2-H1-1-18 AWS

Documents