Packet loss concealment based on statistical
View Patent ↗A method for performing packet loss concealment of lost packets in Voice over IP (Internet Protocol) speech transmission. Statistical n-gram models are initially created with use of a training speech corpus, and then, packets lost during transmission are advantageously replaced based on these models. In particular, the existence of statistical patterns in successive voice over IP (VoIP) packets is advantageously exploited by first using conventional vector quantization (VQ) techniques to quantize the parameter data for each packet with use of a corresponding VQ index, and then determining statistical correlations between consecutive sequences of such VQ indices representative of the corresponding sequences of n packets. The statistic n-gram predictive models so created are then used to predict parameter data for use in representing lost data packets.
1. A method for performing packet loss concealment in a packet-based speech communication system, the method comprising the steps of:
receiving a sequence of one or more speech packets each comprising a set of speech parameters representative of speech data;
identifying a failure to receive a speech packet comprising a set of speech parameters representative of speech data subsequent to the speech data represented by said sets of speech parameters comprised in said received sequence of one or more speech packets;
determining a set of replacement speech parameters based on said sets of speech parameters comprised in said received sequence of one or more speech packets and further based on a statistical model, the statistical model comprising data identifying speech data represented by said set of replacement speech parameters as being likely to follow said speech data represented by said sets of speech parameters comprised in said received sequence of one or more speech packets;
generating speech data for each speech packet in said received sequence of one or more speech packets based on said set of speech parameters comprised therein; and
generating speech data for said speech packet whose failure to be received was identified, based on said determined set of replacement speech parameters.
2. The method of claim 1 wherein said speech parameters comprise line spectral frequency parameters.
3. The method of claim 2 wherein said speech parameters comprise tenth order line spectral frequency parameters.
4. The method of claim 1 wherein said statistical model comprises a set of n-gram predictive models derived based on a training speech corpus.
5. The method of claim 4 wherein said statistical model is based on a Vector Quantization codebook comprising a plurality of sets of speech parameters and a corresponding Vector Quantization index associated therewith.
6. The method of claim 5 wherein each of said n-gram predictive models comprises a sequence of n−1 of said Vector Quantization indices representative of speech data comprised in a given sequence of n−1 speech packets, and one or more corresponding predicted Vector Quantization indices, wherein said one or more corresponding predicted Vector Quantization indices represent speech data which has been determined to be likely to follow said speech data comprised in said given sequence of n−1 speech packets.
7. The method of claim 6 wherein said step of determining said set of replacement speech parameters comprises
converting each of said sets of speech parameters comprised in said received sequence of speech packets to a corresponding Vector Quantization index to thereby construct a sequence of n−1 Vector Quantization Indices based thereupon,
identifying in said set of n-gram predictive models the one or more predicted Vector Quantization indices which correspond to said constructed sequence of n−1 Vector Quantization Indices, and
determining said set of replacement speech parameters based upon said one or more identified predicted Vector Quantization indices.
8. The method of claim 7 wherein said set of replacement speech parameters is determined by identifying the predicted Vector Quantization index which corresponds to said constructed sequence of n−1 Vector Quantization Indices and which represents speech data which has been determined to be most likely to follow said speech data comprised in said given sequence of n−1 speech packets.
9. The method of claim 7 wherein said set of replacement speech parameters is determined by identifying a plural number γ of more predicted Vector Quantization indices which correspond to said constructed sequence of n−1 Vector Quantization Indices, where the plural number γ comprises the perplexity of said statistical model, and by deriving said set of replacement speech parameters based upon a combination of said γ predicted Vector Quantization indices.
10. The method of claim 9 wherein said set of replacement speech parameters is derived by calculating mean values of said speech parameters in said plural number γ of sets of speech parameters associated with said γ predicted Vector Quantization indices.