The world of language and speech recognition is growing, often the solutions offered are proprietary and the data are stored in companies servers.
Since 2014 Mozilla Research is been involved in the creation of an opensource version that can be used by anyone.
The project DeepSpeech is an opensource Speech-To-Text engine, using a model trained by machine learning techniques based on Baidu’s Deep Speech research paper. The model is generated by deep learning neural networks that need to be trained for each language of choice, it this way the algorithm learns the pattern for that precise language.
This is why Mozilla created the online project Common Voice, where you can donate your voice by reading sentences or validate the recordings available.
The Italian community is working on both sides, activating people to contribute to Common Voice, and training the DeepSpeech algorithms for the Italian language. Such training requires high computational power, the collaboration with the HPC4AI project will help the Italian community in the training and release of the model.
Contacts: Alessandro Mazzei, Stefania Delprete