add description for sid

link ot thesis

README.md

@@ -60,5 +60,5 @@ This was part of my Bachelor Thesis.
 
 - [Sphereface](https://github.com/wy1iu/sphereface): The original implementation of the Angular margin based softmax implementation for face recognition.
 - [Speech Brain](https://speechbrain.github.io/) An all in one PyTorch speech recognition framework.
-- [pyannote.metric(https://pyannote.github.io/pyannote-metrics): A framework for diarization evaluation and error analysis.
+- [pyannote.metric](https://pyannote.github.io/pyannote-metrics): A framework for diarization evaluation and error analysis.
 - [kaldi with tensorflow dnn](https://github.com/hsn-zeinali/x-vector-kaldi-tf): A Tensorflow implementation of x-vector topology on top of kaldi.

sid/README.md

@@ -1,3 +1,44 @@
 ## Speaker identification
 
-*Implement me*
+This file will guide you through the usage of the sid implementation.
+[The scripts and their meaning](#the_scripts_and_their_meaning) will
+tell you what scripts there are wand what they are used for.
+[Steps](#steps) tells what you need to do to get it running successfully.
+
+
+### The scripts and their meaning
+
+First of all, you will find all scripts belonging to the execution of the model in the `/src` directory. `/local` does contain helper scripts.
+
+- **diary_extract.py** is used for extracting xvectors for the diarization task. This script is used by `run.sh` script of the diarization implementation.
+- **extract.py** is used for extracting xvectors when these should be saved and further processed by e.g. PLDA.
+- **loss_functions.py** contains the implementation of the two loss functions used in this work.
+- **model.py** declares the model for the deep neural network.
+- **train.py** is used to control and train the model.
+- **utils.py** contains handy functions. (read more about each function in the script it self).
+- **validate.py** This will generate x-vectors but not save them in order to directly compare different extracted x-vectors with cosine distance.
+- **get_epoch_loss.py** calculates the average loss of each epoch
+- **plda-score.sh** runns plda on the x-vectors and saves the outcome as score files.
+- **run-parralel.sh** is a helper script for executing `validate.py` and `extract.py` in parallel on multiple GPUs.
+- **train-plda.sh** will train the PLDA.
+
+### Steps
+
+First run the [`run.sh`](https://github.com/kaldi-asr/kaldi/blob/master/egs/voxceleb/v2/run.sh) script from the kaldi voxceleb v2 eg **fully or at least until stage 6**.
+Then the `run.sh` script from this project can be executed.
+`run.sh` takes three arguments.
+The first parameter tells the flavour of the model. The flavour can either be 
+ - `stat` for statistical polling and regular softmax loss
+ - `attent` for attention and regular softmax loss, or
+ - `asoft` for statistial polling and angular softmax loss.
+
+The second parameter tells if PLDA should be used or not. It can either be
+- `plda` for plda enabled, or
+- `nonplda` for plda disabled.
+
+The third parameter is a just a word you can add in the end to identify your run.
+So an example can be:
+`$ ./run.sh asoft nonplda first_test_run`
+Runns angular softmax without plda and calls the resulting log and eer file `asfot_nonplda_first_test_run.*`
+
+Read more in the publication of my thesis (link <here>, if not ask me in an issue).