It has long been known that Facebook developers have been working on creating a specific speech recognition engine for their platform.
Since the beginning of this year, money offers have been made by their research department to collect voice recordings worldwide in different languages.
Now it’s official: they have developed what they claim is the largest Automatic Speech Recognition (ASR) model of its kind, a model that has learned to understand words in 51 languages after training on over 16,000 hours of voice recordings.
In an article published on the Arxiv.org prepress server, the co-authors say that the system, which contains about a billion parameters, improves speech recognition performance by up to 28.8% compared to the current average reference values.
Designing a single model to recognize speech in multiple languages could be revolutionary. First of all, it simplifies the production line of the backend in voice technologies and studies have shown that the formation of multilingual models on languages of the same group can reduce the global Word Error Rate (WER).
#Voicebot #VocalMultichannel #chatbot