Tree structured cohort selection for speaker recognition system

• US 6,006,184 A
• Filed: 01/28/1998
• Issued: 12/21/1999
• Est. Priority Date: 01/28/1997
• Status: Expired due to Fees

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Yamada, Eiko, Hattori, Hiroaki
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Smits, Talivaldis Ivars
• Application Number:
  US09/014,565
• Time in Patent Office: 692 Days
• Field of Search: 704/239, 704/245, 704/246
• US Class Current: 704/246
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 17/12 Score normalisation

Abstract

In a speaker recognition system, a tree-structured reference pattern storing unit has first through M-th node stages each of which has nodes that respectively store a reference pattern of inhibiting speakers. The reference pattern of each node of (N-1)-th node stage represents acoustic features in the reference patterns of predetermined ones of the nodes of the N-th node stage. An analysis unit analyzes input speech and converts the input speech into feature vectors. A similarities calculating unit calculates similarities between the feature vectors and the reference patterns of all of the inhibiting speakers. An inhibiting speaker selecting unit sorts the similarities and selects a predetermined number of inhibiting speakers. The similarities calculating unit calculates the similarity of the node of the first node stage and calculates the similarities of ones of the nodes of the N-th node stage which are connected to a predetermined number of nodes of the (N-1)-th node stage, selected in an order based on highest similarities.

Claims

1. A speaker recognition system comprising:

a tree-structured reference pattern storing unit having first through M-th node stages with the first node stage having one node and with each of the second through M-th node stages having a plurality of nodes, each of said nodes storing a reference pattern of an inhibiting speaker, each of said nodes of an (N)-th node stage being connected to predetermined ones of said nodes of said (N-1)-th node stage, said reference pattern of each of said nodes of said (N-1)-th node stage representing acoustic features in said reference patterns of the predetermined ones of said nodes of said N-th node stage, where M represents an integer greater than three and N represents an integer from one to M;

an analysis unit for analyzing an input verification utterance and for converting said input verification utterance to feature vectors;

a similarities calculating unit connected to said tree-structured reference pattern storing unit and to said analysis unit for calculating similarities between said feature vectors and said reference patterns of all of said inhibiting speakers; and

an inhibiting speaker selecting unit connected to said similarities calculating unit for calculating said similarities to select a predetermined number of said inhibiting speakers.

2. A speaker recognition system as claimed in claim 1, wherein said similarities calculating unit calculates said similarities of ones of said nodes of said N-th node stage which are connected to a predetermined number of ones of said nodes of said (N-1)-th node stage selected in order from one of said nodes of said (N-1)-th node stage having the highest similarities;

and wherein said similarities calculating unit uses said similarities of ones of said nodes of said (N-1)-th node stage for unselected ones of said nodes except said predetermined number of ones.

3. A speaker recognition system as claimed in claim 1, wherein said tree-structured reference pattern storing unit has a plurality of reference patterns of said inhibiting speakers in said M-th node stage.

4. A speaker recognition system as claimed in claim 2, wherein said tree-structured reference pattern storing unit has a plurality of reference patterns of said inhibiting speakers in said M-th node stage.