Techniques for curating audio and IR commands using machine learning may be provided. For example, the system can receive an audio stream that includes a plurality of audio segments provided by a particular user (e.g., as determined through acoustic fingerprinting) and the system can store the audio stream and/or segments in the user's profile. The system can also store a command associated with the audio segment(s). When a portion of the audio is provided in conjunction with a same command over a threshold number of repetitions, the next time that that audio segment is received, the system may provide the command that corresponds with that audio segment to an output device to cause an operation of the output device.