We have developed a Japanese Lombard speech corpus suitable for a wide range of applications such as those for improving the performance of noisy speech and speaker recognition systems and analyzing acoustic changes introduced by the Lombard effect. The corpus contains clean speech data with neutral and Lombard talking styles and noisy speech data with the Lombard talking style. Its development was based on a sample size of 40 people speaking while being subjected to various types of noise at different sound pressure levels (SPLs). The evaluations were primarily targeted at speech and speaker recognition systems, but they also relate to the accuracy analysis of simulation-based assessments of noisy speech recognition systems. The impact of the Lombard effect on the recognition system performance is also discussed, as is the extent to which the impact varies across different types and SPLs of noise.