Mozilla has set up a project to collect voice samples in order to improve speech recognition accuracy for open source ASR that it is building. Here are some useful links to the project.
https://github.com/mozilla/DeepSpeech
https://voice.mozilla.org/