QRP08-3: Speech Repetition Method for Compensating Delay Spike Gaps in Voice Packet Stream

Voice over Internet Protocol (VoIP) utilizes the Internet for voice communication. VoIP has a problem that a voice gap occasionally happens according to a delay spike. A voice gap caused by a delay spike is longer than 500 milliseconds. Voice gap compensation methods have been proposed before now, but they are all only for short term gaps. No compensation method for long term voice gaps was proposed, because it was too difficult to compensate several phonemes. Our proposed method is for long term voice gaps. When a spike delay happens and next packet does not arrive, the receiver stops replaying voice and repeats the talk-spurt from the beginning instead of replaying next packet after the voice gap. This method can transmit the meaning of a sentence clearly, because automatic speech repetition simulates speech repair which often occurs in normal conversation. Humans are accustomed to this speech repair. This method can preserve naturalness as well, because it changes voice gaps into speech repetition. And the listener has little awareness of speech repetition compensation. We implemented the IP telephony system equipped with speech repetition compensation and also evaluated this method. As the result, we could ensure the usefulness of this method.