Method of and apparatus for determining start-point and end-point of isolated utterances in a speech signal

• US 4,945,566 A
• Filed: 11/18/1988
• Issued: 07/31/1990
• Est. Priority Date: 11/24/1987
• Status: Expired due to Fees

• Current Assignee: US Philips Corporation (Koninklijke Philips N.V.)
• Original Assignee: US Philips Corporation (Koninklijke Philips N.V.)
• Inventors: Mergel, Dieter, Tomaschewski, Horst H., Ney, Hermann
• Primary Examiner(s): Kemeny, Emanuel S.
• Application Number:
  US07/274,093
• Time in Patent Office: 620 Days
• Field of Search: 381/41, 381/46
• US Class Current: 704/253
• CPC Class Codes: G10L 25/87 Detection of discrete points within a voice signal

Abstract

In a method of and an arrangment for determining the start-point and end-point of a word signal in a speech signal consisting of isolated utterances, three adjacent windows are determined at each new digital value for the last arrived stored digital values, in which the central window contains the actual word signal. The length of this central window is varied for each digital value between a minimum and a maximum value, and a threshold value is determined from the two adjacent windows and is subtracted from the energy contained in the central window. Thus, the method and the apparatus always takes the overall speech signal into account instead of individual isolated portions so that a reliable end-point determination then is possible.

Claims

1. In a method of determining a start-point and an end-point of a word signal corresponding to an isolated utterance in a speech signal by establishing an extreme value in a sequence of digital values derived from the speech signal, taking into account those values surrounding the extreme value of the signal variation and a threshold value, the improvement comprising:

assigning a plurality of previously, sequentially received digital values to three adjacent time windows, a first window (end-window) including a predetermined first number (B_R) of the digital values which arrived last, a second window (signal window) including a second number (B_S1) of digital values, said second number varying between a predetermined first value and a predetermined higher second value, and a third window (start-window) including a predetermined third number (B_F) of digital values, forming for each new digital value a threshold value (thr) from the digital values in the first window and, consecutively for each value (1) of the second number (B_S1), from the digital values of the third window, decreasing each digital value of the second window by said threshold value, comparing the sum of the digital values thus decreased for each of said second number to the highest previous sum similarly produced and, depending on the result of said comparison, storing said sum together with positional data indicating the position of the second window in the sequence of digital values, and wherein the positional data stored last indicate the start-point and the end-point of the word signal.

2. A method as claimed in claim 1, wherein only positional data which remained unchanged for a predetermined number of consecutively arrived digital values are used as start-point and end-point.

3. A method as claimed in claim 2, wherein the threshold value is formed by adding the digital values in the first window and in the third window and a correction value.

4. A method as claimed in claim 3, wherein for each new digital value, using the lowest value of the second number (B_S0), the sum of the digital values of the second window is formed and stored if a previously stored second window sum is smaller than the present sum and the sum of the digital values of the third window is formed and stored if a previously stored third window sum is larger than the present sum, and the correction value is produced by taking the difference between the two stored window sums.

5. A method as claimed in claim 4, wherein the correction value is the difference between the two window sums, divided by a constant predetermined signal-to-noise ratio value.

6. A method as claimed in claim 1, wherein the lowest of always three consecutive digitized sampling values of the speech signal is used as the digital value.

7. An arrangement for performing the method as claimed in claim 1 comprising:

a first store for storing digital values derived from a speech signal, a second store for storing intermediate results, an arithmetic unit which receives the digital values from the first store and also the intermediate results from the second store and determines the energy in one of the windows and also further intermediate results, a comparator for comparing intermediate results from the second store to the values produced by the arithmetic unit and for controlling entry of the arithmetic unit values into the second store, a control unit for addressing, in accordance with the steps of the method, the first and the second store and the arithmetic unit, and a counting device for counting the different second numbers of digital values in the second window and for applying an end-of-loop signal to the control unit after a predetermined number of different second numbers of values.

8. An arrangement as claimed in claim 7, wherein at least the arithmetic unit and the control unit comprise a microprocessor.

9. A method as claimed in claim 1, wherein the threshold value is formed by adding the digital values in the first window and in the third window and a correction value.

10. A method as claimed in claim 9, wherein for each new digital value, using the lowest value of the second number (B_S0), the sum of the digital values of the second window is formed and stored if a previously stored second window sum is smaller than the present sum and the sum of the digital values of the third window is formed and stored if a previously stored third window sum is larger than the present sum, and the correction value is produced by finding the difference between the two stored window sums.

11. A method as claimed in claim 10, wherein the correction value is determined by taking the difference between the two window sums, divided by a constant predetermined signal-to-noise ratio value.

12. A method as claimed in claim 1, wherein the threshold value is determined by:

deriving a first signal indicative of average energy in the first window,

deriving a second signal indicative of average energy in the third window,

deriving a third signal indicative of a threshold correction value, and

adding said first, second and third signals to derive a further signal indicative of said threshold value.

13. A method as claimed in claim 12, wherein said third signal is derived by:

deriving a fourth signal indicative of average energy in the second window,

subtracting said second signal from the fourth signal, and

dividing the result of the subtracting step by a given signal/noise ratio value to derive said third signal.

14. A method as claimed in claim 9, wherein the digital value comprises a lowest value of three consecutive digitized sampling values of the speech signal.