Francis Tyers

Another idea would be to write up a technical report/paper and post it to ArXiv or somewhere.

Thanks for the report, this is a known issue when testing using transfer learning directly after training. You can stop the test and rerun it independently from the checkpoint and...

[Here](https://github.com/ftyers/commonvoice-docker/blob/main/test.sh) is the command I use. But for further assistance, please get in touch on [Matrix](https://app.element.io/#/room/#machinelearning:mozilla.org) or [Discourse](https://discourse.mozilla.org/c/deepspeech/247).

@DanBmh do you have any indication that this works better than tuning the whole network? In our experiments (see earlier versions of the [Common Voice paper](https://arxiv.org/abs/1912.06670) and [Josh' thesis](http://jrmeyer.github.io/misc/MEYER_dissertation_2019.pdf) --...

@DanBmh So, you freeze all layers apart from the last one, and then train the last layer. Then when that has trained, you train all the layers? I'd be interested...

Perhaps this could be made into a topic on [Discourse](https://discourse.mozilla.org/) under the [Mozilla STT](https://discourse.mozilla.org/c/mozilla-voice-stt/) topic and further information could be gathered as to the effectiveness, you could also give us...

@the-nose-knows it's never to early to start. What is the "Android specific gradle stuff" that you would like it to not rely on? Perhaps start there, and feel free to...

The error looks like something to do with SWIG, there is documentation here: http://www.swig.org/Doc1.3/Java.html

Thanks @soroushhashemifar could you give some details about how you imagine this should be used and how the "reduce on plateau" functionality doesn't solve this already?

Hi @yashsehgal you could start by joining us on [#machinelearning](https://chat.mozilla.org/#/room/#machine-learning:mozilla.org) on Matrix. It will be a bit difficult to guide you through GitHub issues! Thanks for your interest :)