Recently there has been rapidly increasing use of biometrics as ways of authentication of individuals in various gateways such as airports, ATM machines, electronic devices and more. Biometric modalities include voice, iris, face, fingerprint, DNA, handwriting, etc. They are attached to a person's identity and may not be changed should they ever fall in the hands of unauthorized individuals. Thus with the abundance and promise of biometric usage in each person's life, it is of great importance to preserve the privacy of an individual's biometric data. Secure computation and outsourcing, on the other hand, have been constantly improving and becoming more practical in the last few years. The goal of secure computation is to provide privacy of sensitive data while the data is being used in computation involving not fully trusted parties. The task of protecting the privacy of biometric data is quite challenging. One challenge in this regard is the variety of types of computation that different biometrics require. Based on the biometric type and the goal of computation, different computational techniques are required. Providing security in each case requires its own tools and techniques. In addition, usually these computations are carried out on a large number of inputs, and they demand high computational power. Thus such computations are outsourced to powerful yet untrustworthy servers. Providing secure solutions for outsourcing biometric computations often proves to be very difficult. In this work we explore solutions and techniques to provide secure computation over iris images and voice data. In particular, we provide the first secure solutions for outsourcing of iris matching computation to either a single server or multiple servers. We evaluate the accuracy of the proposed approaches experimentally. In addition, we provide the first solutions for secure multi-party computation and secure outsourcing to multiple servers of voice data. To achieve this goal, we develop the first secure multi-party computation techniques for floating point operations. We show experimentally that floating point operations enjoy rather fast performance and in some cases outperform integer operations. We then extend our solution to secure two-party joint computation on voice data. Our solutions work both in passive and active security models.