(19) |
 |
|
(11) |
EP 1 860 646 A3 |
(12) |
EUROPEAN PATENT APPLICATION |
(88) |
Date of publication A3: |
|
03.09.2008 Bulletin 2008/36 |
(43) |
Date of publication A2: |
|
28.11.2007 Bulletin 2007/48 |
(22) |
Date of filing: 27.03.2003 |
|
(51) |
International Patent Classification (IPC):
|
|
(84) |
Designated Contracting States: |
|
DE FI FR GB NL |
(30) |
Priority: |
29.03.2002 US 369043 14.01.2003 US 341869
|
(62) |
Application number of the earlier application in accordance with Art. 76 EPC: |
|
03100795.8 / 1394769 |
(71) |
Applicant: AT&T Corp. |
|
New York, NY 10013-2412 (US) |
|
(72) |
Inventors: |
|
- Conkie, Alistair, D.
Morris County, NJ 07960 (US)
- Kim, Yeon-Jun
Morris County, NJ 07981 (US)
|
(74) |
Representative: Suckling, Andrew Michael |
|
Marks & Clerk
4220 Nash Court Oxford Business Park South
Oxford
Oxfordshire OX4 2RU Oxford Business Park South
Oxford
Oxfordshire OX4 2RU (GB) |
|
|
|
(54) |
Automatic segmentaion in speech synthesis |
(57) A method for segmenting phone labels to reduce misalignments in order to improve
synthetic speech when the phone labels are concatenated comprises:
training a set of HMMs using one of a specific speaker's hand-labeled speech data
and speaker-independent speech data;
segmenting the trained set of HMMs using an alignment to produce phone labels, wherein
each phone label has a spectral boundary;
using a weighted slope metric to identify bending points of spectral transitions,
wherein each bending point corresponds to a spectral boundary; and
correcting a particular spectral boundary of a particular phone label if the particular
spectral boundary does not coincide with a particular bending point.