A voice retrieval apparatus executes processes of: converting a retrieval string into a phoneme string; obtaining, from a time length memory, a continuous time length for each phoneme contained in the converted phoneme string; deriving a plurality of time lengths corresponding to a plurality of utterance rates as candidate utterance time lengths of voices corresponding to the retrieval string based on the obtained continuous time length; specifying, for each of the plurality of time lengths, a plurality of likelihood obtainment segments having the derived time length within a time length of a retrieval sound signal; obtaining a likelihood showing a plausibility that the specified likelihood obtainment segment specified is a segment where the voices are uttered; and identifying, based on the obtained likelihood, for each of the specified likelihood obtainment segments, an estimation segment where utterance of the voices is estimated in the retrieval sound signal.