Hi,

I have been thinking about a speech-to-text tool on N-900. One that can transcribe the speech in an almost readable form. Though there is a selfish motive for it which I will be explaining here, but in general, it might prove to be useful for most of the people. Two of the strong reasons to have a working speech-to-text software are:

1. In my case (I am a student), it would help me in transcribing all that happens in a particular session of the lecture, easing my workload of taking notes. In most cases, it would help in assignments and stuffs like that.

2. The same issue of transcribing may be applied with regard to certain meetings/discussions etc., this would ease the task of documentation and things like those.

The major question that people first see is its accuracy. IMO the google's app in iPhone (for voice search) or in android (various apps) is highly effective in finding out the correct words in most cases. Though this is internet dependent, but still, it does the job in an efficient manner. But I am not completely sure if I would be able to port that here (should be doable).

Plus with google announcing their research on Voice-to-Voice translation, my hopes of a good system has certainly gone up.

We can also have native systems to do the job like CMU's PocketSphinx which are optimized for ARM processors. But then, the accuracy of these systems are debatable. There were also a couple of discussions on the OLPC wiki about the speech-to-text warez, but they are still in their incipient stages.

The question now is, how useful will this application be in your view? Are there any better ways to do this? Any significant approaches? Or is there any other project which is following the same route/path, I would certainly like to join and would like to contribute my bit.

---
Kakashi (Pranav)