We present a mobile application using two smartphones as drumsticks. Based on accelerometer data the type of the beat (left, middle, or right) is derived. Because acceleration sensors are device specific, a device-specific support vector machine on each device is used to detect beats and type of beats. To attain good recognition on a variety of devices more quickly, the training process is simplified and backed by a server. For the sound generation the two devices are connected and a configurable drum sound for each beat on either device is played on one of the devices.