New Cognitive Services capabilities are now generally available
Published date: May 19, 2020
New Cognitive Services capabilities are now generally available:
- Computer Vision—Advanced text extraction: The most advanced text extraction capability for Computer Vision, Read 3.0, is now generally available and expanding its language coverage beyond English and Spanish to include French, German, Portuguese, Italian, and Dutch. Read 3.0 in containers is also available in preview.
- Containers: Deploy Cognitive Services anywhere from the cloud to the edge with containers. Language Understanding and Text Analytics sentiment analysis in containers are now generally available.
- Language Understanding—Enhanced portal experience: The Language Understanding service has revamped the labeling experience, making it easier to build apps and bots that can understand the complex language structures people tend to use. For example, in this order: “I want a large chicken pizza without sauce and a medium pizza with olives,” there are two different language structures within the same order. This new portal makes it easier to break apart complex requests into related parts.
- QnA Maker—Improved collaboration and text editing capabilities: QnA Maker intelligently parses existing content such as FAQ pages or other document types into Q&A pairs. The service produces a ‘knowledge base’ which is a set of questions and answers. Inevitably, multiple users will want to edit and curate the knowledge base. Use Role-based Access Control (RBAC), to enable users to collaborate on knowledge bases, making it even easier to collaborate on rapid bot development. Additionally, QnA Maker is now giving more control to content managers with a rich text editor they can use to control the formatting of responses to users.
- Speech—Expanded language coverage, accuracy improvements:
- Speech to Text—Quickly transcribe audio to text. Speech to Text is expanding to 27 new locales (coming soon), with 30 percent improvement in speech transcription accuracy.
- Neural Text to Speech—Converts text to lifelike speech for more natural interfaces. Neural TTS is extending support to 11 new locales with 15 new voices, with pronunciation error rate reduced by 50 percent for 13 locales, enabling more customers to benefit from a broad range of natural-sounding voices.
- Text Analytics—Enhanced capabilities: Text Analytics v3 in general availability includes sentiment analysis, key phrase extraction, language detection, and named entity recognition capabilities, as well as model version controls. Sentiment analysis v3 provides improved accuracy in addition to document-level and sentence-level confidence scores. Named entity recognition v3 includes 5 new categories and 10 subcategories including Product, Event, Skill, and Address with improved accuracy across all categories. In addition, the enhanced Text Analytics v3 SDK provides easy-to-consume and robust coding interfaces to build client applications that use Text Analytics’ capabilities. Text Analytics now has a data limitation update, enabling the API to become more efficient at handling large payloads and provide better consistency on the response time across multiple requests.