Speech adaptation device suitable for speech recognition device and word spotting device

• US 5,819,223 A
• Filed: 01/26/1996
• Issued: 10/06/1998
• Est. Priority Date: 01/26/1995
• Status: Expired due to Fees

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Takagi, Keizaburo
• Primary Examiner(s): Tung, Kee M.
• Application Number:
  US08/592,174
• Time in Patent Office: 984 Days
• Field of Search: 395/2.65, 395/2.6, 395/2.61, 395/2.63, 395/2.64, 395/2.45, 395/2.52-2.54, 382/181, 382/184, 382/190, 382/209, 382/215, 382/224, 382/225, 382/228
• US Class Current: 704/256.1
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

A speech adaptation device comprises a vocabulary independent reference pattern memory for memorizing a plurality of vocabulary independent reference patterns having one or more categories. Each category has one or more acoustic units, and has such a connection relation of the acoustic units that allows reception of any sequence of the acoustic units appearing in the input speech. A preliminary matching unit is for use in making time-alignment between the time series of the feature vectors of the input speech obtained from the analysis unit and the vocabulary independent reference pattern to obtain mean vectors for individual categories of the input speech and the vocabulary independent reference pattern from the aligned portion for the individual categories of the feature vectors of the input speech and the vocabulary independent reference pattern. An adaptation unit is for use in making correction of at least one of the time series of the feature vectors of the input speech and the vocabulary independent reference pattern by using the mean vectors for each category calculated by the preliminary matching unit.

Claims

1. A speech adaptation device comprising:

an analysis unit for use in converting an input speech into a time series of feature vectors;

a vocabulary independent reference pattern memory for memorizing a plurality of vocabulary independent reference patterns having one or more categories, each category having one or more acoustic units, and having such a connection relation of the acoustic units that allows reception of any sequence of the acoustic units appearing in said input speech;

a preliminary matching unit connected to said analysis unit and said vocabulary independent reference pattern memory for use in making time-alignment between the time series of the feature vectors of said input speech obtained from said analysis unit and said vocabulary independent reference pattern to obtain mean vectors for individual categories of the input speech and the vocabulary independent reference pattern from the aligned portion for the individual categories of the feature vectors of said input speech and said vocabulary independent reference pattern; and

an adaptation unit for use in making correction of at least one of the time series of the feature vectors of said input speech and said vocabulary independent reference pattern by using the mean vectors for each category calculated by said preliminary matching unit,

wherein the category of said vocabulary independent reference pattern comprises one or more noise categories and one or more speech categories, said adaptation unit replacing a noise portion of either one of the time series of the feature vectors of said input speech or said vocabulary independent reference pattern with a mean vector Ni of the noise portion of the other, subtracting a mean vector Nt of the noise portion of either one from a speech portion of the same, then multiplying it by (Si-Ni)/(St-Nt), (where a slash / indicates division for each channel of the vectors, Si indicates a mean vector of the speech portion of the other, and St indicates a mean vector of the speech portion of the same), and further adding Ni thereto.

3. A speech adaptation device as claimed in claim 1, wherein the acoustic unit of said vocabulary independent reference pattern is one of a state of HMM, each centroid obtained after the vector quantization, and distribution of HMM.

4. A speech adaptation device as claimed in claim 1, wherein said vocabulary independent reference pattern is formed by using an acoustic unit of the reference pattern for recognition or for word spotting.

2. A speech adaptation device as claimed in claim 2, wherein the acoustic unit of said vocabulary independent reference pattern is based on a consonant (C), a vowel (V) and a noise (N) appearing in said input speech, or based on a sequence of two or more of these (such as CV, VC, CVC, VN).

5. A speech adaptation device comprising:

an analysis unit for use in converting an input speech into a time series of feature vectors;

a vocabulary independent reference pattern memory for memorizing a plurality of vocabulary independent reference patterns having one or more categories, each category having one or more acoustic units, and having such a connection relation of the acoustic units that allows reception of any sequence of the acoustic units appearing in said input speech;

a preliminary matching unit connected to said analysis unit and said vocabulary independent reference pattern memory for use in making time-alignment between the time series of the feature vectors of said input speech obtained from said analysis unit and said vocabulary independent reference pattern to obtain mean vectors for individual categories of the input speech and the vocabulary independent reference pattern from the aligned portion for the individual categories of the feature vectors of said input speech and said vocabulary independent reference pattern; and

an adaptation unit for use in making correction of at least one of the time series of the feature vectors of said input speech and said vocabulary independent reference pattern by using the mean vectors for each category calculated by said preliminary matching unit,

wherein the acoustic unit of said vocabulary independent reference pattern is based on a consonant (C), a vowel (V) and a noise (N) appearing in said input speech, or based on a sequence of two or more of these (such as CV, VC, CVC, VN).

6. A speech adaptation device as claimed in claim 5, wherein the acoustic unit of said vocabulary independent reference pattern is formed by using one of HMM (Hidden Markov Model), a part of a time series feature vectors of a training speech, and each centroid obtained after vector quantization.

7. A speech adaptation device as claimed in claim 6, wherein said vocabulary independent reference pattern is formed by using an acoustic unit of the reference pattern for recognition or for word spotting.

8. A speech adaptation device as claimed in claim 5, wherein said vocabulary independent reference pattern is formed by using an acoustic unit of the reference pattern for recognition or for word spotting.

9. A speech adaptation device comprising:

an analysis unit for use in converting an input speech into a time series of feature vectors;

a vocabulary independent reference pattern memory for memorizing a plurality of vocabulary independent reference patterns having one or more categories, each category having one or more acoustic units, and having such a connection relation of the acoustic units that allows reception of any sequence of the acoustic units appearing in said input speech;

a preliminary matching unit connected to said analysis unit and said vocabulary independent reference pattern memory for use in making time-alignment between the time series of the feature vectors of said input speech obtained from said analysis unit and said vocabulary independent reference pattern to obtain mean vectors for individual categories of the input speech and the vocabulary independent reference pattern from the aligned portion for the individual categories of the feature vectors of said input speech and said vocabulary independent reference pattern; and

an adaptation unit for use in making correction of at least one of the time series of the feature vectors of said input speech and said vocabulary independent reference pattern by using the mean vectors for each category calculated by said preliminary matching unit,

wherein said adaptation unit carries out correction for at least one of the time series of the feature vectors of said input speech and said vocabulary independent reference pattern by using a difference between the means vectors of each category, and

wherein the acoustic unit of said vocabulary independent reference pattern is based on a consonant (C), a vowel (V) and a noise (N) appearing in said input speech, or based on a sequence of two or more of these (such as CV, VC, CVC, VN).