Foundry Tools

Generative AI service that extracts structured insights from unstructured documents, images, audio, and video using a single API — built on Bedrock foundation models, with customisable blueprints for domain-specific extraction. Launched at re:Invent 2024

Automated image and video analysis with deep learning, offering object and scene detection, facial analysis, text in image (OCR), celebrity recognition, and content moderation

Natural language processing service that extracts entities, key phrases, sentiment, syntax and PII from unstructured text, with custom classification and entity models and a separate Comprehend Medical for clinical text.

Neural text-to-speech service that converts text into lifelike speech in 40+ languages with dozens of voices, including expressive long-form, generative, and newscaster speaking styles plus SSML markup and phoneme control

Automatic speech recognition service with real-time streaming and batch transcription across 100+ languages, speaker diarisation, custom vocabulary, PII redaction, and specialised Transcribe Medical and Transcribe Call Analytics flavours

Neural machine translation across 75 languages with optional active custom translation (style transfer, terminology lists, parallel-data fine-tuning) — real-time string translation for chat or APIs, plus async batch translation for document corpora

Image analysis and recognition with pre-trained models for label detection, OCR, face and landmark recognition, explicit content filtering, and product search

Managed natural language understanding with entity extraction, sentiment analysis, syntactic analysis, content classification across 700+ categories, and PII detection — plus AutoML Natural Language for custom classification, entity extraction, and sentiment models

Automatic speech recognition with Chirp universal speech model supporting 125+ languages, real-time streaming and batch APIs, speaker diarisation, word-level timestamps, and phrase hints to bias recognition toward domain-specific vocabulary

Neural text-to-speech with 380+ voices in 50+ languages, including premium Journey voices, Studio voices for long-form narration, and Custom Voice for cloning an organisation's brand voice, plus full SSML and phoneme control

Neural machine translation between 135+ languages with per-document translation that preserves layout, glossary support for domain-specific terms, AutoML Translation for custom models, and an Advanced edition with batch mode and adaptive translation

Neural machine translation API with strong Korean ↔ English / Japanese / Chinese / Vietnamese performance — text, document, and image (OCR + translate) endpoints

Pre-trained natural-language services — sentiment analysis, named-entity recognition, key-phrase extraction, language detection, text classification, and translation across 100+ languages.

Pre-trained computer-vision services — image classification, object detection, OCR, document image analysis, and custom-model training on customer data. REST API plus async batch processing.

Speech-to-text transcription with word-level timestamps, speaker diarisation, profanity filtering, and custom-vocabulary support. Synchronous and asynchronous-batch modes; 30+ languages.