|Mozilla Updates Voice Recognition Project|
|Written by Kay Ewbank|
|Tuesday, 14 July 2020|
Mozilla has released an updated dataset for its Common Voice project, along with a major update to its DeepSpeech speech-to-text and text-to-speech engines.
Mozilla's Common Voice project aims to provide a free database of recordings of people speaking sample sentences. The database is open-source and free for anyone can use. DeepSpeech is an automatic speech recognition (ASR) engine that aims to make speech recognition technology and trained models openly available to developers. DeepSpeech is a deep learning-based ASR engine with a simple API, and a collection of pre-trained English models.
The Common Voice project now has an updated dataset with 7,226 total hours of contributed voice data, of which 5,591 of have been confirmed as valid. The release comprises over 5.5million clips, with over 5,000 unique speakers. The new release includes voice recordings in 54 languages, of which 14 are new to the platform and dataset. The update also now has voice data for single words including the digits zero through nine, as well as the words yes, no, hey and Firefox. The team says this single word datawill help Mozilla benchmark the accuracy of its open source voice recognition engine, Deep Speech 243, in multiple languages for a similar task.
DeepSpeech has also been upgraded, and the developers say it now offers faster speech recognition and support for Google’s TensorFlow Lite framework. DeepSpeech consists of two main subsystems: an acoustic model and a decoder. The acoustic model is a deep neural network that receives audio features as inputs, and outputs character probabilities. The decoder uses a beam search algorithm to transform the character probabilities into textual transcripts that are then returned by the system.
or email your comment to: firstname.lastname@example.org