Disclosed are various embodiments for translation of speech in a video messaging application. A segment of streaming video is decoded to separate the visual component from the audio component. The audio component is then converted to text, which may then be translated and converted to a translation output comprising a new language. In response, the translation output may be encoded with the previously separated visual component. A delay is imposed on the visual component to account for any delays that may arise in translation. The translated video may then be streamed to participants giving the appearance of real-time video conferencing.