Speech transformation using log energy and orthogonal matrix

• US 6,278,970 B1
• Filed: 03/11/1998
• Issued: 08/21/2001
• Est. Priority Date: 03/29/1996
• Status: Expired due to Fees

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Milner, Benjamin P
• Primary Examiner(s): Knepper, David D.
• Application Number:
  US09/043,060
• Time in Patent Office: 1,259 Days
• Field of Search: 704/250, 704/236, 704/222, 704/203
• US Class Current: 704/203
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit

Abstract

Calculate the log frame energy value of each of a pre-determined number n of frames of an input speech signal and apply a matrix transform to the n log frame energy values to form a temporal matrix representing the input speech signal. The matrix transform may be a discrete cosine transform.

Claims

1. A method of generating features for use with speech responsive apparatus, said method comprising:

calculating the logarithmic frame energy value of each of a sequence of a predetermined number n of frames of an input speech signal; and

applying a predetermined orthogonal transform matrix to the n logarithmic frame every values to form a frame energy vector representing the input speech signal, wherein the predetermined orthogonal transform matrix encodes temporal information such that the elements of the off-diagonal of the covariance matrix of the frame energy vector are substantially zero.

2. A method as in claim 1 wherein successive sequences represent overlapping groups of n frames of the input signal.

3. A method as in claim 1, wherein the predetermined orthogonal transform matrix represents a discrete cosine transform.

4. A method as in claim 1, wherein the temporal matrix is truncated so as to include fewer than n elements.

5. A method of speech recognition comprising:

receiving an input signal representing speech, said input signal being divided into frames;

generating a feature by calculating the logarithmic frame energy value of each of a predetermined number n frames of the input speech signal;

applying a predetermined orthogonal transform matrix to the n logarithmic frame energy values to form a frame energy vector representing the input speech signal, the predetermined orthogonal transform matrix encoding temporal information such that the elements of the off-diagonal of the covariance matrix of the frame energy vector are substantially zero;

comparing the generated feature with recognition data representing allowed utterances, said recognition data relating to the feature; and

indicating recognition or otherwise on the basis of the comparison step.

6. A method of speech recognition as in claim 5 wherein the predetermined orthogonal transform matrix represents a discrete cosine transform.

7. Feature generating apparatus for use with speech responsive apparatus, said feature generating apparatus comprising:

a processor for calculating the logarithm of the energy of each of a predetermined number n of frames of an input speech signal; and

a processor for applying a predetermined orthogonal transform matrix to the n logarithmic energy values so calculated to form a frame energy vector representing the input speech signal, the predetermined orthogonal transform matrix encoding temporal information such that the elements of the off diagonal of the covariance matrix of the frame energy vector are substantially zero.

8. Feature generating apparatus as in claim 7 in which the predetermined orthogonal transform matrix represents a discrete cosine transform.

9. Speech recognition apparatus including feature generating apparatus as in claim 7.