Easy-to-use AI solutions in the cloud: AI services from AWS

With so-called "Software-as-a-Service" (SaaS) solutions for artificial intelligence, cloud providers offer interfaces that make it very easy to add AI capabilities to your own products. At the same time, the cloud provider takes care of the actual hardware and software maintenance. This creates the possibility of easy-to-use AI services.

AI services from AWS - an overview

One example of such a service is the tagging of faces in images. Here, images can be sent directly to the AI service, which then returns the recognized faces as a response. Amazon is one of the major cloud providers with AWS and offers a variety of such easy-to-use AI services. These are offered by AWS as AI Services labeled. We would like to give you a brief overview of the most interesting AI services offered by AWS. How you can use AI in your products without much effort? We would like to give you some inspiration for ideas.

One of the main application areas where easy-to-use AWS AI services exist is speech processing. Figure 1 shows the entire processing pipeline for speech. If voice recordings are available, they are first converted to text. The actual text analysis can take place in text form before text is converted back into speech, depending on the application.

According to the illustration, we handle the available services in three steps:

1. Speech-to-text
2. Text analysis
3. Text-to-speech

Transcription: Speech to text

For the conversion of voice recordings into text (also known as transcription), AWS offers the AI Service Amazon Transcribe. Transcribe uses a powerful machine learning model to enable the transcription of voice recordings for a variety of languages. The voice recordings are usually stored on AWS in so-called object stores (e.g. Amazon S3) and can be sent to the service from there.

For some languages, such as English, not only can completed recordings be used, but live transcription is also possible. This involves transcribing an audio stream. Unfortunately, streaming transcription (as of July 2020) is not yet supported for the German language.

In addition to the actual transcription functionality, Transcribe can also different interlocutors in one shot mark. This makes it possible, for example, to distinguish between customers and support staff during a call on a customer hotline.

To ensure that Transcribe also understands the technical jargon of your specific use case, it is possible to extend the general machine learning model with your own Expand vocabulary.

Documents for text

Despite advancing processes, some processes in many companies are still paper-driven, or there are large stocks of archived documents. To enable text analysis in these cases, these documents must first be digitized (scanned). But that's not all: after scanning, text must be extracted from these documents. This becomes particularly complicated as soon as the document has a complicated layout with tables and images. This is where Amazon Textract by specializing in extracting text from complicated form layouts in a meaningful way.

Intelligent language processing through text analysis

If all information is available in text form, further services are now available: The central service for intelligent language processing on AWS is called "Comprehend" (English for "understand"). The text input document is sent to Comprehend, which can now answer a variety of useful questions:

• The extraction of keywords and key phrases: The most important words and phrases are extracted for the text to be analyzed, along with an assessment of how reliable the algorithm is.
• Sentiment analysis: Is a text more positive, neutral or negative? This distinction is important wherever customers can leave you feedback. One example of this is product reviews that customers leave in online stores.
• Recognition of entities: Comprehend can extract the most important people, places, company names, dates and other so-called entities from a text. This allows you to automatically better understand what a text is actually about and to relate different texts to each other. You can also enter your own internal organizational entities.
• Recognition of the language of the text: If you receive texts from many different sources, it is important to recognize the dominant language of the text in order to be able to execute downstream processes in the correct localized variant.
• Automatic document classification: If you already have a data set of documents that have been assigned to different categories within the organization, you can use Comprehend to sort new documents according to the existing classification.

In addition to text analysis with Comprehend, you can also Automated translation leave. Amazon offers the Translate-Service on. This uses a complex neural network to translate texts into many different languages. Here too, the service can be adapted to your own specialist domain by creating a user-defined terminology.

Text to speech

Finally, text can also be converted back into speech. For this purpose, AWS offers the Polly-Service. Polly can create synthesized voice recordings for many languages from text documents. For English, particularly impressive neuronal voices which produce an even higher voice quality. In addition to creating the synthesized voice, Polly can generate additional metadata that describes, for example, which words are spoken at which point in the synthesized recording. This information can be very useful for further processing steps.

Extensive image and video processing

In addition to intelligent language processing, AI is also particularly successful in the field of Image recognition. To avoid having to train complex models yourself, Amazon offers the Recognition service on. This service offers a wide range of features that work for both individual images and videos:

• Label recognitionThis includes object recognition, in which the objects in the image are marked. In addition, Rekognition can recognize concepts such as "landscape" or "beach". In video files, even activities ("gets into car") can be recognized. If you would like to use Rekognition in your own application scenarios, you can supplement it with subject-specific images in order to recognize additional labels.
• Face recognitionRecognition marks faces on images, including facial features such as the position of the eyes or nose. In addition, faces from different images can be compared, making it possible, for example, to search a photo archive for a specific person.
• Text-in-image recognition: With the text recognition functionality, text in images can be recognized and extracted. For example, you could use this to search for all images that show the same serial number of a machine.
• Content moderationDetection of unsafe content allows you to moderate images uploaded by users accordingly, for example to automatically prevent the publication of sexual content.
• Route detectionWhen analyzing videos, the routes taken by people can be detected. This opens up completely new possibilities for analyzing your video material. For example, you can better understand how your systems are actually used and plan corresponding optimizations.

(Product) recommendations of all kinds

Personalized product recommendations have been an important part of Amazon's business model since the very beginning. For example, customers are recommended products that other customers who were interested in the same product have bought. Such a recommendation engine is easily available to everyone on AWS: Amazon Personalize. Simple, historicized CSV files ("comma-separated value") are used as input, which contain information about the users such as age or gender, the available products and the interaction of users with the products. For example: User X bought the product "pan". Personalize can use this information to train a machine learning model. The model can then be used to answer various questions:

• Personalized product recommendations: based on a user - what are the most interesting products for them?
• Product similarity: based on a product - which products are similar to this product?

In Europe, the service is currently only available in the Region Ireland offered (as of July 2020).

Further AI services from AWS

In addition to the services presented here, there are other AI services:

• Amazon Kendra: A knowledge search engine - Kendra allows you to ask questions in normal language, for which Kendra searches a large number of connected knowledge databases. The service is currently only available in English and in Ireland in Europe (as of July 2020).
• Fraud DetectorRisk assessment is of central importance, especially in B2C business. Amazon Fraud Detector offers functions to detect cases of fraud: When creating new user accounts, for buyers who shop on your website via the "guest" function without a user account. The risk assessment of online payment transactions has also already been announced.
• Amazon ForecastFrom stock levels to sales figures - data with a time reference can be found everywhere. With Forecast, historical data can be used to make predictions about the future. This AI service is a little more hands-on: data must be uploaded, a model trained and deployed.
• Amazon LexLex is a framework for creating chatbots. With Lex, for example, you can provide an automated support bot for customers. However, as of today (July 2020) with one important limitation: Lex currently only supports English as a language.

Conclusion

You no longer need to be a machine learning expert to add AI functionalities to your apps and programs. With its AI services, AWS offers very easy-to-use services, particularly in the area of text processing and image recognition, which you can set up and integrate quickly. There are also some more specialized services, such as fraud detection with Amazon Fraud or the simple creation of forecasts based on historical time series with Amazon Forecast, which open up new possibilities.

It is worth investing manageable resources here to evaluate in a prototype whether you can also benefit from AWS AI services for your use case.

---

Find Cloud platforms generally fascinating? Then our White paper comparing central functionalities on AWS and Microsoft's Azure platform may be of interest to you.

Do you already know the AI trends of the next decade?