- [Jostein] Hi, my name is Jostein, and I'm the team lead for NLB's development team. I will give you an overview of how NLB is using Pipeline 2 in our production workflow. Source files. We get the source files, generally, either as EPUB or PDF from the publishers. We'll then use external producers to get the book marked-up according to the Nordic EPUB 3 guidelines. These EPUBs are the input for our production system. Validation of source files. The first thing that happens is that we validate EPUBs with the Nordic EPUB 3 Validator script in Pipeline 2, this is a script that is not part of the official Pipeline 2 distribution, because it is not a generic script. If the EPUB is valid, then we store it in our master archive. If it is not valid, we return it to the producers. Pre-production. From the master archive, we start with pre-production. This is format specific. So we do pre-processing separately for braille, TTS narration, human narration, and E-book production. At some point we might look into adding sentence detection using Pipeline 2 here, if that's possible, it would be useful both for TTS narration and human narration. Currently we don't use Pipeline 2 during pre-production, so I won't be going into any more details here. Braille conversion. We have been using Pipeline 2 for braille production for a few years now, and we are quite happy with the results, but we also would like some improvements. We are currently using, an NLB specific version of Pipeline 2 that meets our requirements for braille production. However, we have not had time to update this since 2018. So we are stuck with an old version of Pipeline 2 for braille production. There is currently an effort to recombine our version as well as other agency specific versions back into the official DAISY braille scripts, which we are very much looking forward to. TTS narration. For TTS narration, we are currently using Pipeline 1 and its TTS narration scrips. We are about to start testing Pipeline 2 for TTS narration, and it will make it easier to maintain our servers. And with more configuration options, we hope that we'll be able to improve the quality of our TTS productions significantly. TTS post-production. The results from TTS narration is a DAISY 2.02 book. We use Pipeline 1, to validate our DAISY 2.02 books at this point, before making them available for distribution. We are considering switching to Pipeline 2 for DAISY 2.02 validation as well, as it will make it easier to maintain our systems. Human narration. For human narration we use Hindenburg. Pipeline 2 is not involved in this step. Human narration post-production. Post-production of our human narrated audio books starts with a producer that manually listens to parts of the book and makes small corrections if necessary. Most of our human narration is now made onto full textbooks and the granular centralization of text and audio on paragraph or sentence level is done automatically in post-production. Pipeline 2 seems to support sentence detection. So we might want to use that algorithm instead of maintaining our own. We could possibly enrol Pipeline 2 to do sentence detection. At this point, however, since sentence detection could be interesting for other formats as well, we might instead do it during pre-processing, as I mentioned earlier. And finally, the resulting DAISY 2.02 book is validated using Pipeline 1 in the same way as is done for TTS narration. Hybrid, TTS and human narration. We are currently investigating different ways of producing audiobooks, which are partly human narrated and part generated with TTS. The typical use case is for footnotes or other auxiliary contents to pair that with TTS, while the most important text is run by a human narrator. We hope that the TTS narration in Pipeline 2, either now or in the future, can be used for this purpose. E-book conversion. For E-book conversion we do not currently use Pipeline 2 for anything. Manual production and Pipeline 2 instances. In some cases, our producers wants to run Pipeline 2 scripts directly without going through our production system. For this we are using the Pipeline 2 Web UI. In production, we have a one web interface for Nordic EPUB 3 Validation, and one for braille production. In our test environment, we also have two web interfaces, which we use to try out new versions of the Pipeline 2. The production system had used multiple instances of Pipeline 2, and currently we have it connected to two instances of the Nordic EPUB 3 Validator, and one instance of NLB's braille script. These are not the same instances as are used by the web interfaces. So manual production does not interfere with automatic production. The Pipeline 2 Web UI is a separate project from the main Pipeline 2 engine. So it has not been maintained for a while. It has some problems. So it would be nice to have replacement in the future, but it mostly works fine and it's very useful for manual production. So that's it. We're using Pipeline 2 every day in our production system over there, for validation of EPUBs and braille production. We will soon be using it for TTS narration, and we are looking into using it for some other things as well. It is very important for NLB that Pipeline 2 development continues. Thanks for listening!