Common Voice is Mozilla’s campaign to build an open-source voice dataset filled with diverse voice data (over 40 different languages and counting) that’s accessible to everyone. The hope is that with easy access to better data, better voice related technology can be built.
Mozilla currently use the Common Voice dataset to help train the language models for their open-source Speech-To-Text engine called DeepSpeech.
Signing up for Common Voice is free. All you need is a microphone and a web browser.
You can “donate your voice” by reading very short text prompts or you can “validate voices” by listening to clips other people have recorded and confirm if what was spoken matches the text prompt. In either case the text prompts are short single sentences making it easier to rip through them.
There are global daily targets for each option. The recorded clips is usually 1200 and the listening target is 2400. I try to aim for 20 of each a day but I’ve found it’s actually a lot easier to validate the voices for some reason. Below is my favourite prompt to date 😀
