Recipe: Acoustic Word Embeddings for Switchboard
This is a recipe for extracting acoustic word embeddings for a subset of the Switchboard corpus. The models are described in detail in Kamper et al., 2015:
- H. Kamper, W. Wang, and K. Livescu, "Deep convolutional acoustic word embeddings using word-pair side information," in Proc. ICASSP, 2016.
Please cite this paper if you use this code. All the neural networks are implemented in the package couscous.
Install all dependencies (below).
Clone couscous into the appropriate directory:
mkdir ../src git clone https://github.com/kamperh/couscous.git ../src/couscous
Run the steps in kaldi_features/run.sh.
Run the steps in cnn_wordembeds/readme.md.
If you run the steps correctly above, then if you execute the following:
cd cnn_wordembeds/ ./apply_layers.py models/siamese_triplets_cnn.1/ test ./eval_samediff.py \ models/siamese_triplets_cnn.1/swbd.test.layer_-1.npz
Then the evaluation should show the following output:
Average precision: 0.537404372048 Precision-recall breakeven: 0.542724052097
The average precision (AP) of 0.537 is used for the number reported in Table 1, row 9 of Kamper et al., 2015.
- Theano and all its dependencies.
- couscous: should be cloned into the