Abstract: A text-to-speech (TTS) system is configured with multiple voice corpuses used to synthesize speech. An incoming TTS request may be processed by a first, smaller, voice corpus to quickly return results to the user. The text of the request may be stored by the TTS system and then processed in the background using a second, larger, voice corpus. The second corpus takes longer to process but returns higher quality results. Future incoming TTS requests may be compared against the text of the first TTS request. If the text, or portions thereof match, the system may return stored results from the processing by the second corpus, thus returning high quality speech results in a shorter time.