Reducing computation in an i-vector speaker recognition system using a tree-structured universal background model
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- New Mexico State Univ., Las Cruces, NM (United States)
The majority of state-of-the-art speaker recognition systems (SR) utilize speaker models that are derived from an adapted universal background model (UBM) in the form of a Gaussian mixture model (GMM). This is true for GMM supervector systems, joint factor analysis systems, and most recently i-vector systems. In all of the identified systems, the posterior probabilities and sufficient statistics calculations represent a computational bottleneck in both enrollment and testing. We propose a multi-layered hash system, employing a tree-structured GMM–UBM which uses Runnalls’ Gaussian mixture reduction technique, in order to reduce the number of these calculations. Moreover, with this tree-structured hash, we can trade-off reduction in computation with a corresponding degradation of equal error rate (EER). As an example, we also reduce this computation by a factor of 15× while incurring less than 10% relative degradation of EER (or 0.3% absolute EER) when evaluated with NIST 2010 speaker recognition evaluation (SRE) telephone data.
- Research Organization:
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE National Nuclear Security Administration (NNSA)
- Grant/Contract Number:
- AC04-94AL85000
- OSTI ID:
- 1140965
- Report Number(s):
- SAND-2014-2055J; PII: S0167639314000582
- Journal Information:
- Speech Communication, Vol. 66, Issue C; ISSN 0167-6393
- Publisher:
- ElsevierCopyright Statement
- Country of Publication:
- United States
- Language:
- English
Web of Science
Similar Records
What the speaker means: the recognition of speakers plans in discourse
Speaker recognition through NLP and CWT modeling.