Microsoft today announced a series of updates to Azure AI, the umbrella brand for its AI products targeting health care, financial, agricultural, and other industries. The company launched Text Analytics for Health, a feature in preview that enables health providers and researchers to generate new insights from electronic medical records. Microsoft also made Custom Commands and Form Recognizer generally available, allowing developers to create task-oriented voice apps and extract information from up to millions of documents.

AI usage in the enterprise is growing at a record rate. According to a NewVantage Partners survey, the percentage of firms investing greater than $50 million in big data and AI initiatives is up to 65% in 2020 from just 40% in 2018. A separate report by IBM found that 39% of companies are ramping up exploratory phases with AI, with 45% of large companies (over 1,000 employees) claiming to have deployed AI successfully.

“As the world adjusts to new ways of working and staying connected, we remain committed to providing Azure AI solutions that help people and organizations invent with purpose,” wrote Azure AI corporate VP Eric Boyd. “We are excited to introduce these new product innovations that empower all developers to build mission-critical AI apps.”

Text Analytics for Health

Text Analytics for Health is in some ways a response to the pandemic, building on Microsoft’s partnerships with and tools for health organizations to deploy conversational experiences. It’s designed to uncover relationships in electronic health records, with an “opinion mining” feature that assigns sentiment to specific features and topics. Opinion mining is also now available in Microsoft’s broader Text Analytics product for social media data, review sites, and more.

As Boyd points out, the health care system generates approximately a zettabyte (a trillion gigabytes) of data annually, an amount that’s doubling every two years. The scale and distributed nature of the data makes identifying patterns a challenge, but it also increases the potential to improve patient outcomes.

As a proof of concept, Microsoft used Text Analytics for Health and its Cognitive Search service to launch a COVID-19 search engine based on the Allen Institute of AI’s COVID-19 Open Research Dataset (CORD-19). It lets users search academic studies about the coronavirus through filters like publication date, contributors, body structures, diagnoses, and medication names, as well as by symptom and variant.

Form Recognizer and Custom Commands

Last May, Microsoft unveiled Form Recognizer, which automates the ingestion of forms to extract key-value pairs as JSON objects. After a year in testing, the feature is now available for all Azure customers, and it supports forms with tables and other elements, in addition to objects.

Microsoft said Capgemini Group’s Sogeti used Form Recognizer to develop a financial document processing platform. Wilson Allen, another partner, is using Form Recognizer to evaluate financial forms, loan applications, and more.

Also now available generally is Custom Commands, a part of Speech in Azure Cognitive Services that allows developers to create apps for command-and-control scenarios with well-defined variables. Custom Commands supports natural input, including “out of order” (if a user asks about a topic like movies, they can ask about an unrelated topic without skipping a beat) and additional input. And it has built-in corrections that factor in variations in response and pronunciations, dozens of commands out of the box, and a context handler that switches between commands automatically.

Neural Text to Speech

Lastly, Microsoft’s Neural Text to Speech (TTS) service has expanded language support with 15 new natural-sounding voices. These employ models that “learned” to replicate the natural pauses, idiosyncrasies, and hesitancy in people’s speech. In addition to the recordings of a target voice talent, Neural TTS uses a source library that contains voice recordings from many different speakers. And because of the way it synthesizes voices, Neural TTS can produce styles of speech that weren’t part of the original recordings, such as changes in tone of voice and affectation.

The new voices and languages include:

  • Salma in Arabic (Egypt)
  • Zariyah in Arabic (Saudi Arabia)
  • Alba in Catalan (Spain)
  • Christel in Danish (Denmark)
  • Neerja in English (India)
  • Swara in Hindi (India)
  • Colette in Dutch (Netherland)
  • Zofia in Polish (Poland)
  • Fernanda in Portuguese (Portugal)
  • Dariya in Russian (Russia)
  • Hillevi in Swedish (Sweden)
  • Achara in Thai (Thailand)
  • HiuGaai in Chinese (Traditional Cantonese)
  • HsiaoYu in Chinese (Taiwanese Mandarin)

Just a few months ago, Microsoft brought new customer service, newscast, and digital assistant voice styles to Azure Cognitive Services. Each promises natural-sounding speech that matches the patterns and intonations of human voices.