Embodiments of the present disclosure disclose a method, apparatus and system for testing an intelligent voice device. A specific implementation of the method includes: acquiring test data, the test data comprising: to-be-executed field control data and to-be-played voice data; sending, based on the to-be-executed field control data, a first motion instruction to the omnidirectional mobile mechanism connected to the intelligent voice device and/or the voice playing device; sending a playing instruction, in response to determining that a relative pose of the intelligent voice device and the voice playing device is the target relative pose; receiving predefined device information, associated with the voice recognition, of the intelligent voice device; and generating a test result based on the device information.