Computing with Hypervectors for Efficient Speaker Identification
We introduce a method to identify speakers by computing with high-dimensional random vectors. Its strengths are simplicity and speed. With only 1.02k active parameters and a 128-minute pass through the training data we achieve Top-1 and Top-5 scores of 31 in contrast to CNN models requiring several million parameters and orders of magnitude higher computational complexity for only a 2× gain in discriminative power as measured in mutual information. An additional 92 seconds of training with Generalized Learning Vector Quantization (GLVQ) raises the scores to 48 in 5.7 ms. All processing was done on standard CPU-based machines.
READ FULL TEXT