A computer-implemented method to determine whether to introduce latency into an audio stream from a particular speaker includes an audio stream from a sender device. The method further includes providing, as input to a trained machine-learning model, the audio stream and a speech analysis score, information about one or more voice emotion parameters, and one or more voice emotion scores for a first user associated with the sender device, wherein the trained machine-learning model is iteratively applied to the audio stream and wherein each iteration corresponds to a respective portion of the audio stream. The method further includes generating as output, with the trained machine-learning model, a level of toxicity in the audio stream. The method further includes transmitting the audio stream to a recipient device, wherein the transmitting is performed to introduce a time delay in the audio stream based on the level of toxicity.