Mozilla’s DeepSpeech project is the launch of an open source speech recognition system.
Accurate speech recognition technology is not easy to come by. There are countless factors to take into account so that the identification of words is adequate. Not only that. The system also has to be able to recognize expressions, differentiate accents and tones of voice.
The complexity makes the developers of this technology jealously guard it. This is not the case with Mozilla, which has published its voice recognition platform in open source. The system is called DeepSpeech and consists of machine learning algorithms capable of processing language and transforming it into text. The error rate is 6.5%good brand.
DeepSpeech is nourished by a database of 400,000 recordings, from 20,000 different people. In total there are 500 hours of audio that have served to train the algorithms. At the moment, the platform only works with English, but Mozilla plans to launch the product in multilingual mode in 2018.
Mozilla’s is not the only one in the open source community. From John Hopkins University came the Kaldi platform, while from Carnegie Mellon University emerged the CMUSphinx voice recognition system. There are other names, but the truth is that, to date, there is no real alternative to proprietary software.
At the moment, the main speech recognition systems are closed. Google, Amazon, Apple and Microsoft they keep the rights to their technology. This is not to say that a developer, even a speaker manufacturer, cannot use Alexa. In fact, you can, as this has been Amazon’s strategy for advancing the personal assistant market. But Jeff Bezo’s company reserves the development of the software to itself.
Even the company Nuance, which specializes in the field of speech recognition and markets its technology for use by third parties, also does not offer an open source platform. However, Mozilla’s DeepSpeech project could be the first open source initiative to have enough size to compete in this field.
Images: rawpixel, goranmx