My thoughts on Mozilla's Project Common Voice
About Project Common Voice
- A part where you can submit audio recordings of you saying a sentence that the website displays.
- A part where you can validate the recordings sent by others.
The aim of the project is to use the user-submitted, and validated recordings to create a voice recognition engine that is on par with proprietary engines. Most cutting-edge voice recognition engines are proprietary, and thus, the public doesn't get access to technology that they often helped improve. Project Common Voice aims to solve that problem.
The project is accepting recordings right now, and will release an open-source database containing these voices in late 2017. There are no restrictions on the quality of voice recording needed for submission, as the engine needs to be able to understand voice in real-world scenarios, where background noise from vehicles and machinery is common. This means almost anybody can submit their voice for the project. The project aims to collect 10000 hours of recordings, as that is the amount of data needed to create a production grade Speech-to-Text System.
Common Voice Homepage
My thoughts on it
I think Common Voice will be a transformative thing for the following reasons:
1. It will create a cutting-edge,but open-source voice recognition engine
Anyone who has used a personal assistant such as Siri or Google Now will know the scarily accurate voice recognition capabilities they have. Ultimately, if open source is to be spread to the masses, it will have to keep up with advancements in technology, and projects like these, and others, for example Mycroft, the open-source personal assistant should be commended for pushing the boundaries of open-source.
The fact that it is open source also means that other open source software projects, which often cannot use commercial voice recognition due to license or financial issues, will have access to high quality voice recognition, which can be used in all sorts of cases, like for improving accessibility.
2. The open source dataset can be used by other people to train their engines on.
If you want to make a new voice recognition algorithm, you will have to have a dataset that you can train your engine on. This is a big limiting factor for the creation of many open source projects, because this means that you have to purchase, often at great cost, a dataset that someone else collected and is selling.
The fact that the dataset will be released publicly means that any developer can use it to train his/her engine.
3. It will make it normal for open source projects to collect user data
Now, before you write angry comments, please here me out. I think this is a great thing.
Anyone who knows the basics of machine learning/AI/Neural Networks (and i admit, i only really know the basics) will know that a lot of the foundational knowledge of this field was developed decades ago, and the main reason this field has exploded in recent times is the vast amounts of user data that these mega-companies have.
AI will increasingly be the most important part of any software going forward, and if open source software is to be competitive with commercial software, it must have the same, high quality AI. Now sure, most open source software, individually, doesn't really have the userbase that these tech companies do, but as a whole, open source software is used by almost every computer user. If we can create huge open source datasets, that open source developers can then use to train their software on, then we can ensure that open source software remains competitive with proprietary software going into the future.
If you liked this post, please upvote and resteem it, and don't forget to follow me for more content like this :-)