Auto-Auto is a ground-breaking new framework to support the building of languages for use in Automatic Speech Recognition. This technology will enable the rapid creation and continuous improvement of any language or variants of existing languages, to support specific use-cases such as regional dialects, industry domains or accents.
Dr Tom Ash, Director of Speech Recognition at Speechmatics, who has been leading our efforts on Auto-Auto commented, “Building new languages in speech recognition has traditionally been a laborious process, but using our recent research in machine learning, we have created technology that makes the process faster, more accurate and more efficient. We believe this technology will help us train fast and accurate models for many of the world’s languages over the coming months”.
Language database creation has always been a time consuming and prohibitively expensive task in speech recognition and one of the biggest barriers to entry for the industry. Given the difficulty of building new languages, speech recognition efforts have always been focused on the top 5 to 10 languages, where the technology is now achieving good results. However, approximately 7000 languages are in use today, so being able to develop the next 100 most used languages, in an accelerated time-frame, will ensure the world can be better understood. Auto-Auto presents a huge opportunity to Speechmatics and its customers, as Speechmatics will be the only speech company that offers ASR in almost any language at commercial accuracy levels.
Now that we have created the Auto-Auto framework, we will be releasing beta versions of new languages over the coming weeks. You can register your interest in becoming an early tester by joining our “Developers Forum”. Email firstname.lastname@example.org for details.
Dr Tony Robinson, Chief Scientific Officer and founder of Speechmatics, “Utilising Big-Data/Machine-Learning techniques within Auto-Auto enables the rapid creation of Enterprise grade languages, far more quickly and efficiently than has been the recent case”.
Benedikt von Thuengen, CEO, remarks “Through this technology our enterprise and applications customers can ensure the most accurate and efficient transcription of their voice data; thereby aiding discoverability and searchability, and delivering actionable data. The high cost of building a new language has traditionally been a big barrier to entry into the speech recognition market for new players. With this new tool we will be able to build almost any language or variant that our customers require, automatically and with minimal resources. Auto-Auto is a complete game changer, not only for us but for the entire industry. We look forward to working with anyone wanting to build out ASR in any language or application.”