Speech enhancement based on singular value decomposition of ambiguity function

Summary form only given. An approach is proposed for speech enhancement, which expresses a solution at a different angle from the effective methods given by several publications in recent years. The analysis process for noisy speech is based on the ambiguity function (AF) of time-frequency representation (TFD) that is potentially fit for characterization of non-stationarities. After resolving the ambiguity function of the degraded speech by way of singular value decomposition (SVD), the speech improvement is achieved by tuning the singular values using the scaling factor matrix. Finally, some experimental results are produced to elucidate the efficacy of the method.