Commit ca3d302f authored by Ben Milde's avatar Ben Milde

Added README. This commit reflects a version of the code that is useful for...

Added README. This commit reflects a version of the code that is useful for reproducing the results in the paper "Unspeech: Unsupervised Speech Context Embeddings", Benjamin Milde, Chris Biemann, Proceedings of Interspeech 2018, Hyderabad, India. There might be a newer and better version of this repository available at http://unspeech.net and/or https://gitlab.com/milde/unspeech
parent 51622e31
Unspeech training code for “Unsupervised Speech Context Embeddings”
If you use our code or models in your academic work, please cite this paper:
“Unspeech: Unsupervised Speech Context Embeddings”, Benjamin Milde, Chris Biemann, Proceedings of Interspeech 2018, Hyderabad, India
Visit http://unspeech.net for more information, examples on training models, using them to generate features and clustering them. There are also pretrained models available for some of the models that were evaluated in our paper.
Short overview of the main programs:
unsup_model_neg.py – Main training and feature generation code, using a discriminative objective function. Works with Tensorflow 1.5+, tested with 1.8.
unsup_model.py – Some first experiments with other objective functions and a generative model of speech. Not used in the paper, Pre-Tensorflow 1.0 code.
unsup_model_10.py – Similar to unsup_model.py but updated to Tensorflow 1.0 (will not work with newer versions)
show_feats.py – can be used to visualize features in Kaldi ark,scp format (FBANK, MFCC, unspeech…)
cluster.py - cluster features with HDBSCAN, evaluate with ARI / NMI also visualize clusters with TSNE.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment