Amazon Transcribe Unleashes Its Generative AI to Support 100 Languages

Amazon Web Services (AWS) has announced a major update to its Amazon Transcribe product, which provides generative AI-based transcription for speech-to-text applications. Amazon Transcribe can now recognize more than 100 spoken languages and offer new AI features for customers.

Amazon Transcribe: A generative AI-based transcription service

Amazon Transcribe is a cloud-based service that allows AWS customers to add speech-to-text capabilities to their apps. It uses self-supervised algorithms to learn patterns of human speech in different languages and accents. It can transcribe speech from audio and video files, as well as from live streams and phone calls.

According to AWS, the Transcribe has trained on millions of hours of unlabeled audio data from over 100 languages. It has ensured that some languages were not over-represented in the training data to achieve high accuracy for lesser-used languages as well as more frequently spoken ones.

The Transcribe claims to have 20 percent to 50 percent accuracy across many languages. It also offers features such as automatic punctuation, custom vocabulary, automatic language identification, and custom vocabulary filters. It can handle noisy environments and multiple speakers.

New languages and AI capabilities for Amazon Transcribe

The latest update to Amazon Transcribe adds 21 new languages to its existing 79 languages, bringing the total to 100 languages. The new languages include Afrikaans, Albanian, Amharic, Armenian, Azerbaijani, Bengali, Bosnian, Bulgarian, Burmese, Croatian, Dari, Estonian, Georgian, Hausa, Kannada, Khmer, Kurdish, Latvian, Lithuanian, Macedonian, Malayalam, Marathi, Mongolian, Nepali, Pashto, Persian, Serbian, Sinhala, Slovak, Slovenian, and Tamil.

The update also introduces new AI capabilities for Transcribe customers. One of them is Transcribe Call Analytics, which is a platform that summarizes interactions between an agent and a customer in a call center. It uses generative AI models to extract key information from the call transcript, such as sentiment, intent, issues, and resolutions. AWS said this feature reduces the after-call work for agents and managers and allows them to quickly access insights without reading the entire transcript.

Another new AI capability is Transcribe Content Generation, which is a feature that helps customers create engaging titles or email subject lines for their recommendation lists. It uses generative AI models to write catchy and relevant phrases that match the theme of the list, such as “Top 10 movies to watch this weekend” or “How to boost your productivity with these apps”. AWS said this feature helps customers increase click-through rates and conversions.

Amazon Transcribe vs other AI transcription services

AWS is not the only company that offers AI-powered transcription services. There are other competitors in the market, such as Otter, which provides AI transcriptions to consumers and enterprises. Otter also has a summarization tool that condenses long transcripts into key points.

Another company that is working on AI-powered language services is Meta, formerly known as Facebook. Meta recently announced that it is developing a generative AI-powered translation model that can recognize nearly 100 spoken languages and translate them into text or speech.

AWS also has other AI products that complement Amazon Transcribe, such as Amazon Personalize, which allows customers to offer personalized recommendations to their users, based on their previous activity. AWS has added a new feature to Amazon Personalize, called Content Generation, which can write titles or email subject lines to thematically connect recommendation lists.

What is the accuracy of Amazon Transcribe?

The accuracy of the Transcribe varies depending on the source, the language, the domain, and the features used. The Transcribe claims to have 20% to 50% accuracy across a wide range of languages, and it also provides ways to improve accuracy through the use of custom vocabularies and custom language models.