The future of Mozilla's Speech to Text project DeepSpeech is uncertain

[harlan4096]

Following the layoffs at Mozilla and rumor that the organization would be able to extend its search deal with Google for another three years, it became clear quickly that things needed time to settle down.

Employees who were fired would provide insight on how the letting go of employees would affect certain projects at Mozilla. A week later, Reuben Morais published an article on the official Discourse site of Mozilla about the future of DeepSpeech.
 
DeepSpeech, or Mozilla Voice STT (Speech To Text) is an open source Speech-to-Text engine that is trained using machine learning techniques. It is designed as a counterweight to closed source services operated by Google, Amazon and other companies.

DeepSpeech uses machine learning techniques that are based on Baidu's Deep Speech research paper and Google TensorFlow for its implementation. The service can be run on a wide range of devices in real time including Raspberry Pi 4, devices that run Windows, OS X or Linux, Android, and iOS.

Morais admits in the first paragraph of the post that he does not know how the layoffs and the restructuring of the entire company affects the DeepSpeech project.

Unfortunately, as of this moment we don’t have concrete answers to give. We’re working to find out if the project will have a new home in the restructured Mozilla, and what changes would be necessary for a successful transition.

Morais notes that the team will release DeepSpeech 1.0 soon as "most of the technical changes" landed already and that the team sees "no reason not to ship it". The team will continue its work on DeepSpeech until an official decision has been made and the team is informed about it.

Closing WordsNot all is lost if Mozilla would decide to end support for DeepSpeech. The project is open source and it is possible that another organization or individual developers will fork it to continue development. DeepSpeech is not the only VTT project that is open source either, e.g. Kaldi, Espresso, or Nvidia OpenSeq2Seq, are also open source.

Common Voice, designed to "help teach machines how real people speak", has also been moved into maintenance mode as the future of the project is unclear.

Now You: What would you do, as Mozilla?
...

Continue Reading