[Kde-accessibility] Presenting: Live Transcriber

[Peter Grasch]

Hi,

On 07/29/2013 09:32 PM, Simon Eichhammer wrote:
> Since our last meeting I have worked on a small application called "Live
> Transcriber".
> 
> LiveTranscriber is an open-source, semi-automated transcription platform
> combining automated speech recognition (ASR) with human-made corrections
> and reviewing.
> 
> The whole application is writte in Rails 4, using HTML5 EventSource and
> Redis for client-server communication.
> 
> The speech processing is done by a modified version of
> pocketsphinx_continuous ("fatpocketsphinx").
> 
> I made a small video to demonstrate it and to explain future goals and
> tasks:
> 
> http://www.youtube.com/watch?v=EA9yWoyhHvM
> 
> The project is published under the Affaro GPL license (Thanks Peter for the
> advice).
(Affero)

> Any transcriptions made with the application must be shared with the
> community.
> This way the existing speech models can be further improved.
This clause, as it is now, is IMHO tricky.

Instead, I think it would be better to amend the conditions (probably
through adapting the license) of the software to explicitly define the
annotated samples created by using the software as "covered work" (as in
the context of point 2 of the AGPL).

Reasoning:
First, I'm pretty sure that the current situation (AGPL + separate,
additional clause) is not legally enforceable as the AGPL (just like the
GPL) has a clause saying that user rights are limited with the
conditions of the license only - no other restrictions apply.
Secondly, your clause also requires a good definition of "the community"
and that alone is very hard. The AGPL (which would then also cover the
samples) requires the "source" (in this case the samples) to be released
whenever a work based on it is published (which includes being used on a
website).

Again, I'm not a lawyer but I would think this constellation would be a
lot easier to enforce - the GPL has already been upheld in courts
numerous times.
However, again there are drawbacks:
* I'm not sure if we can declare these samples to be covered under the
license as they probably don't constitute a copyright-able work (i.e.
they can not be covered by the license).
* In contrast to your original clause, this constellation would mean we
would have to *ask* for the samples (which they *have to* provide)
instead of being *given* the samples (without prompt) (in German law:
"Holschuld" vs "Bringschuld").

As a third option we may also just introduce another clause in the
license to say something along the lines of:
"All annotated samples [where 'annotated sample' is defined as ...]
created through the use of the software [where 'use of the software' is
defined as ...] are to be made publicly available under the terms and
conditions of the BSD license.
In the event of this corpus [...] being used, incorporated, etc. into
any other work which is subsequently propagated [as defined in the
AGPL], the availability of this corpus must also be announced to the KDE
e.V. or it's legal successor."
(not sure if KDE e.V. makes a lot of sense there but I'd strongly
suggest to name a legal body and not a natural person)

It still allows people to try out the software at home without requiring
them to immediately share all samples of them playing around with the
software (which, again, is not just extremely hard to enforce but also
probably doesn't make a lot of sense) without compromising on the
essence why this clause is there in the first place.
Drawback: It's no longer a standard AGPL (which it might still be in the
second option outlined above) which means that it will be impossible to
link to e.g. other AGPL or GPL code.

> The source code is available at GitHub:
> https://github.com/skpvox/LiveTranscriber
Haven't yet had time to look through it thoroughly but will hopefully
get around to do a bit of code review in the coming days.

> In case that the development of the backend (fatpocketsphinx) becomes more
> active I will add it as a separate project on KDE.
We talked on IRC already but for the record: we should also evaluate
other decoders before we commit to forking pocketsphinx.

> I'm looking forward to any feedback,
Will have more feedback for you once I reviewed the code.

Best regards,
Peter