To generate conversation record with an index without depending upon the vocal characteristic of a user at a low cost.
A narration data generating part M28 generates narration data obtained by dividing voiced data from silent data together with their required times on the basis of a voice inputted from a voice inputting part M26 inputting only the voice of a user carrying a conversation recorder 20 having strong directivity. Narration data similarly generated by a conversation recorder 20 carried by another speaker is inputted through a beam light receiving and emitting part M16 and a beam communication controlling part 18, and a conversation data generating part M34 regards agreement representation from the narration data, etc., of a self-speaker and the other speaker as silent data, processes the representation, etc., to generate conversation data. Then, an index data storing part M40 sets an index to voiced data other than the agreement representation, etc., from the conversation data.