The latest model of Amazon Transcribe has undergone training using millions of hours of unlabeled audio data sourced from over 100 languages.
Amazon Web Services (AWS) has expanded the speech recognition abilities of its transcription tool. Amazon Transcribe, an automatic speech recognition (ASR) service, now offers support for over 100 languages, thanks to a new speech foundation model. The AWS service utilises machine learning techniques to automatically convert speech to text for businesses. Amazon Transcribe is used to transcribe customer calls, create subtitles and meeting notes, and detect harmful content in audio.
Announced at the AWS re: Invent event on Sunday, where the company showcased a slew of AI services now enhanced with foundation model (FM)-powered capabilities, Amazon Transcribe’s new model is trained using millions of hours of unlabelled audio data from over 100 languages. According to AWS, its transcription tool can now learn inherent patterns of human speech across different languages and accents. The tool will now also ensure that traditionally under-represented languages are also recognised and transcribed accurately.
According to AWS, the new model increases the accuracy of the transcription service between 20 percent and 50 percent for most languages. “In addition to substantial accuracy improvement, this large ASR model also delivers improvements in readability with more accurate punctuation and capitalization,” the AWS blog announcing the new speech model said.
The new ASR model brings increased ease of use, customization, user safety, and privacy across all the 100+ languages, AWS said. “These include features such as automatic punctuation, custom vocabulary, automatic language identification, speaker diarization, word-level confidence scores, and custom vocabulary filter,” it added. AWS customers can embed the speech-to-text tool, which now offers expanded support for different accents, noise environments, and acoustic conditions, in their applications.
AWS customers using Amazon Transcribe service in batch mode can access the FM-powered speech recognition without having to change either the API endpoint or input parameters, the blog said.
In addition to an improved Amazon Transcribe, AWS has also brought enhancements to its other AI services. Amazon Personalize, which brings machine learning-powered personalization for customers, can now generate more compelling content and product recommendations. And Amazon Lex AI chatbot now provides accurate and conversational responses to FAQs.
In June, reports indicated that AWS was exploring the possibility of utilizing new AI chips from Advanced Micro Devices (AMD). Dave Brown, the Vice President of Elastic Compute Cloud at Amazon, mentioned, “We’re still working together on where exactly that will land between AWS and AMD, but it’s something that our teams are working together on.