[FAQs] AI Services

Question 1

In what modes can Amazon Translate receive text?

Answer 1

Either real-time or batch

Question 2

Can Amazon Translate be used if you don’t know what language the text is in?

Answer 2

Yes - it will use Comprehend behind-the-scenes to determine the source language

Question 3

How is Amazon Translate used to perform batch translation?

Answer 3

Asynchronously with an API call that points to an S3 bucket folder with up to 1 million documents (up to 5 GB)

Question 4

In what modes can Rekognition receive data?

Answer 4

• Images can be provided either as bytes in the API call or as an S3 path
• Videos can be stored or streaming

Question 5

What tasks can Rekognition perform

Answer 5

• Detect objects and scenes
• Detect and analyse faces
• Recognise celebrities
• Identify inappropriate content
• Match faces
• Custom labelling (images only)
• Text detection

Question 6

What video formats does Rekognition support?

Answer 6

MOV and MPEG-4 encoded with the H.264 codec

Question 7

What is the maximum runtime for stored videos used with Rekognition?

Answer 7

They can be up to 2 hours long

Question 8

How does image resolution affect Rekognition results?

Answer 8

While it accepts images that are at least 80 pixels in both dimensions, a VGA (640x480) or higher resolution is recommended

Question 9

How big should objects be for Rekognition to reliably identify them?

Answer 9

As a rule of thumb, at least 5% of the image size

Question 10

How can Rekognition results be reviewed by a human?

Answer 10

Using Amazon Augmented AI

Question 11

How is Rekognition used with Amazon Augmented AI?

Answer 11

Results below a threshold, or as part of a random sample, can be sent for human review

Question 12

Besides resolution, what can affect Rekognition results?

Answer 12

Heavy blur and lighting etc,

Question 13

What kinds of labels does Rekognition use when classifying?

Answer 13

• Objects e.g. person
• Scenes e.g. beach
• Concepts e.g. outdoors

It uses a hierarchical system so parent labels are provided if they exist

Question 14

How is object and scene detection different with Rekognition?

Answer 14

It uses multiple frames to better understand motion etc.

Question 15

How can you tell if the Rekognition model has been updated?

Answer 15

Every API call returns a [*]ModelVersion based on the kind of model e.g. LabelModelVersion

Question 16

What information does Rekognition provide about a detected face?

Answer 16

Its pose, gender, age, emotions and facial landmarks etc.

Question 17

What kinds of facial recognition does Rekognition support?

Answer 17

• Face comparison - are two people the same?

- True facial recognition based on a face collection

Question 18

Does Rekognition work for S3 objects stored in other regions?

Answer 18

No

Question 19

How can custom pronunciations be used with Polly?

Answer 19

Either SSML inline or custom lexicons

Question 20

How does Polly encode which words are spoken when?

Answer 20

Using Speech Marks, which are delivered as a JSON stream separately from the audio

Question 21

What information do Speech Marks include?

Answer 21

• Sentences
• Words
• Visemes - the shape of the lips corresponding to that sound
• SSML

Question 22

How does Polly return the synthesised speech?

Answer 22

Either to an S3 bucket or as a stream

Question 23

Where can Amazon Lex bots be deployed?

Answer 23

Alexa, Connect, Facebook Messenger, Slack and Twillo SMS etc.

Question 24

At a high-level, how does Amazon Fraud Detector work?

Answer 24

You upload historical fraud data which is used to train a model. This model is used along with a model based on Amazon’s experiences

Question 25

How can Amazon Fraud Detector be customised?

Answer 25

You can add basic rules e.g. IF X and Y then A

Question 26

What is Amazon Personalise used for?

Answer 26

Making recommendations to uses and finding similar items

Question 27

What formats does Textract support?

Answer 27

PNG, JPEG and PDF

Question 28

What is Amazon Kendra?

Answer 28

An enterprise search service which can answer questions using ML

Question 29

What is Amazon Forecast?

Answer 29

A service to make predictions based on time-series data

Question 30

How can Amazon Forecast be customised?

Answer 30

By configuring the HPO (hyper parameter optimisation) parameters

Question 31

What are HPO parameters?

Answer 31

Hyper-parameter optimisation parameters

Question 32

How does Transcribe support receiving data?

Answer 32

Either as files in an S3 bucket or with a bidirectional stream over HTTP2

Question 33

What are the duration limits for Transcribe?

Answer 33

Up to 4 hours for streaming or batch

Question 34

How can domain specific language be captured by Transcribe?

Answer 34

By configuring a custom vocabulary using IPA or SoundsLike

Question 35

What factors are likely to impact the performance of Transcribe?

Answer 35

Background noise, strong accents and switching between multiple languages

Question 36

How can personal information be removed from transcriptions?

Answer 36

With Automatic Content Redaction, which is only support for batch transcriptions in English

Question 37

What does Automatic Content Redaction do in Transcribe?

Answer 37

Filters out personally identifiable information (PII)