Assistive robotic devices -such as robotic wheelchairs- need environment information to ensure safe navigation. In the field of envi- ronment recognition, range sensors (such as Li- DAR and ultrasonic systems) and artificial vi- sion devices are widely used; however, these sensors depend on environment constraints (such as lighting variability or color of objects). In this work we propose a sound-based ap- proach to enhance the environment recognition process. Our proposal is based on a neural net- work implementation which is able to classify up to 15 different environments, with accuracy rates ranging from 84% to 93%. This classification can later be used to constrain assistive vehicle navigation in order to protect the user. In this work, we include real experimentation (carried out at UFES's campus -Brazil-) and statistical validation of our proposal.