| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech retrieval device, a speech retrieval method, and a program capable of accurately retrieving a retrieval word from among a speech signal of different speech speed.SOLUTION: A conversion unit 112 converts a retrieval character string into a phoneme string, a speech speed information acquisition unit 114 acquires speech speed information according to speech speed of a speech signal of a retrieval object, a time length change unit 115 changes an average duration length of a phoneme according to the speech speed information and a time length derivation unit 116 derives an utterance time length of speech corresponding to the retrieval character string by using the changed duration length. A section designation unit 117 designates a plurality of likelihood acquisition sections in the speech signal of the retrieval object, and a likelihood acquisition unit 121 acquires likelihood indicating a degree of likelihood that the likelihood acquisition section is a section in which a speech corresponding to the retrieval character string is uttered. On the basis of likelihood acquired for each selected likelihood acquisition section, an identification unit 127 identifies an estimation section in which it is estimated that speech corresponding to the retrieval character string is uttered from among the speech signal of the retrieval object.SELECTED DRAWING: Figure 2 |
