Method and apparatus for classifying audio signals

• US 4,542,525 A
• Filed: 09/27/1983
• Issued: 09/17/1985
• Est. Priority Date: 09/29/1982
• Status: Expired due to Fees

• Current Assignee: Blaupunkt-Werke GmbH (Aurelius AG)
• Original Assignee: Blaupunkt-Werke GmbH (Aurelius AG)
• Inventors: Hopf, Reinhard
• Primary Examiner(s): Isen, Forester W.
• Application Number:
  US06/536,213
• Time in Patent Office: 721 Days
• Field of Search: 381/41, 381/42, 381/43, 381/46, 381/56, 381/110, 73/584, 455/228
• US Class Current: 381/56
• CPC Class Codes: G10H 1/00 Details of electrophonic musical instruments
• CPC Class Codes: G10H 2210/046 for differentiation between music and non-music signals, based on the identification of musical parameters, e.g. based on tempo detection
• CPC Class Codes: G10L 25/00 Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 muting semiconductor-based amplifiers when some special characteristics of a signal are sensed by a speech detector, e.g. sensing when no signal is present, H03G3/34
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10

Abstract

The null transitions of an audio frequency signal are converted by Schmitt trigger circuits, one of which has a small hysteresis range centered on the null value and the other of which has a much larger hysteresis range likewise centered on the null value, into two binary pulse sequences of variable pulse lengths. The Schmitt trigger circuits are so constituted that a positive pulse length is produced by a negative null transition of the audio signal and vice versa and, moreover, the Schmitt trigger circuits return to their quiescent state 2 milliseconds after a positive null transition of the signal, also producing a positive pulse length, in this case beginning the indication of the pause. The pauses in the two binary pulse sequences thus produced, which exceed predetermined length (60 milliseconds in both cases and, additionally, 30 milliseconds in the case of the pulses formed by the Schmitt trigger with the narrower hysteresis range) and from the three different pause detection operations logic circuits derive either a speech recognition signal, a music recognition signal or an indication of an unidentifiable signal. The logic circuit uses as criteria the number of pauses and the time span of simultaneous or alternating appearance of signal pauses derived from the two different pulse sequences.

Claims

1. Method of automatic classification of audio signals based on conversion of the null transitions of an analog audio frequency signal into at least one pulse sequence by reference to voltage thresholds determined by an absolute value of voltage difference from the null value of the analog signal, comprising the steps of:

converting said analog audio frequency signal into a first binary pulse sequence by use of first voltage thresholds determined by a first absolute value of voltage;

converting said analog audio frequency signal into a second binary pulse sequence by use of second voltage thresholds determined by a second absolute value of voltage substantially higher than said first absolute value of voltage;

detecting the pauses of said first binary pulse sequence which exceed a predetermined first time lapse magnitude and thereby producing a first derived pulse sequence;

detecting the pauses of said first binary pulse sequence which exceed a predetermined second time lapse magnitude which is substantially greater than said first time lapse magnitude and thereby producing a second derived pulse sequence;

detecting the pauses of said second binary pulse sequence which exceed a predetermined third time lapse magnitude which is at least about the same magnitude as said second time lapse magnitude and thereby producing a third derived pulse sequence;

determined whether said audio-frequency signal is a speech signal, a music signal or an unidentifiable kind of signal from said derived pulse sequences, by pause count and by simultaneity and/or alternation of pauses detected by said pulse sequences respectively derived from said first and second binary pulse sequences, and

preparing readiness for repetition of said method when said determining step is completed. PG,21

2. Method according to claim 1 in which both said signal conversion steps are combined with provision of return of the binary pulse sequence to the quiescent signal state after a short time interval of at least one millisecond following the last previous change of binary value away from the signal state corresponding to the quiescent state.

3. Method according to claim 2 in which said binary pulse sequences are so produced that every negative pulse flank of said first and second binary pulse sequence represents a positive null transition of said audio-frequency signal and every positive pulse flank represents either a negative null transition of said audio-frequency signal or the beginning of a pause, and in which the duration of positive pulses of said first and second binary pulse sequences is used to produce, by comparison with reference values of time lapse magnitude, the derived pulses of said derived pulse sequences.

4. Method according to claim 2 in which said second time lapse magnitude is about twice said first time lapse magnitude.

5. Method according to claim 4 in which said second and third time lapse magnitudes are substantially equal.

6. Method according to claim 2, in which the classification determining step includes the substep of determining that said audio-frequency signal is a speech signal when the number of pauses represented by pulses of said first derived pulse sequence is greater than three and less than twelve while the number of pauses represented by pulses of said third derived pulse sequence is greater than four.

7. Method according to claim 2, in which the classification determining step includes the substep of determining that said audio-frequency signal is a music signal when the number of pauses represented by pulses of said first derived pulse sequence is greater than three and the time lapse of a pause represented by said third derived pulse sequence, occurring in the absence of simultaneous representation of a pause by said second derived pulse sequence, exceeds a predetermined fourth time lapse magnitude.

8. Method according to claim 2, in which the classification determining step includes the substep of determining that said audio-frequency signal is a music signal when the number of pauses represented by pulses of said first derived pulse sequence is smaller than 3 and the time lapse of non-detection of pauses represented by said second derived pulse sequence is greater than a predetermined fifth time lapse magnitude, which is substantially greater than said fourth time lapse magnitude.

9. Method according to claim 8 in which said fifth time lapse magnitude is about twice said fourth time lapse magnitude.

10. Method according to claim 2 in which the classification determining step includes the substep of determining that said audio-frequency signal is of an unidentifiable kind when the time lapse during which signal pauses represented by said second derived pulse sequence occur is greater than a sixth time lapse magnitude which is greater than said fourth time lapse magnitude and less than said fifth time lapse magnitude.

11. Method according to claim 2 in which the classification determining step includes the substep of determining that said audio frequency signal is of an unidentifiable kind when the number of signal pauses represented by said third derived pulse sequence occurring during simultaneous non-detection of pauses represented by said second derived pulse sequence is greater than 8.

12. Method according to claim 7 in which the classification determining step includes the substep of determining that said audio-frequency signal is of an unidentifiable kind when the number of pauses represented by pulses of said first derived pulse sequence is at least 3 and the time lapse of non-detection of signal pauses represented by said second derived pulse sequence is greater than said fourth predetermined time lapse value.

13. Method according to claim 10 in which said first absolute voltage value threshold is 0.15 volts, said second voltage value threshold is 1.1 volts, said first predetermined time lapse magnitude is 30 milliseconds, said second predetermined time lapse magnitude and said third predetermined time lapse magnitudes are 60 milliseconds, said fourth predetermined time lapse magnitude is 1.5 seconds, said fifth predetermined time lapse magnitude is 3 seconds and said sixth predetermined time lapse magnitude is 1.6 seconds.

28. Apparatus according to claim 1, in which a filter having a cut-off frequency above its passband located in the neighborhood of 36 Hz is interposed between said source of audio-frequency signals and the inputs of said first and second Schmitt trigger circuits.

14. Apparatus for connection to a source for automatic classification of audio-frequency signals received from a transmission or recording channel for classification of said signals as speech, music or unidentified signals, comprising:

first and second Schmitt trigger circuits having their inputs connected to said source of audio-frequency signals and having their hysteresis thresholds substantially symmetrically disposed about the null potential of said audio frequency signals as supplied by said source, both said Schmitt trigger circuits having two possible states, one of which corresponds to an initial state in absence of said audio-frequency signals and being equipped with means for assuring return of said circuits to said initial state after an interval of at least one millesecond in the other of said states, said first Schmitt trigger circuit having a small hysteresis voltage range and said second Schmitt trigger circuit having a substantially larger hysteresis voltage range than said first Schmitt trigger circuit;

first and second monoflop timing circuits connected to the output of said first Schmitt trigger circuit for respectively detecting pauses in said audio-frequency signal exceeding first and second predetermined time lapse values;

a third monoflop timing circuit connected to the output of said second Schmitt trigger circuit for detecting gaps in higher amplitude portions of said audio signals exceeding a third predetermined time lapse value, and

an evaluation circuit connected to the output of said first, second and third monoflops and containing counters for counting said pauses and gaps detected by said respective monoflop timing circuits, and fourth, fifth and sixth timing circuits, said counters and said fourth, fifth and sixth timing circuits being interconnected for providing signal classification output signals, said evaluation circuit including means for resetting at least said counters promptly after signal classification output signal has been produced.

15. Apparatus according to claim 14, in which the hysteresis range of said first Schmitt trigger circuit is 0.3 V, the hysteresis range of said second Schmitt trigger circuit is 2.2 V, said first predetermined time lapse value is 30 ms and said second and third predetermined time lapse values are both 60 ms.

16. Apparatus according to claim 14, in which said fourth, fifth and sixth timing circuits are incorporated in a time lapse threshold logic circuit having its input connected to the outputs of said monoflop timing circuits, and in which a storage unit and a correlation circuit are located in said evaluation circuit, said storage unit having its inputs connected to the outputs of said time lapse threshold logic circuit and its outputs connected to said correlation circuit, said correlation circuit having outputs providing the respective classification signals.

17. Apparatus according to claim 16, in which said storage unit is composed of an array of RS latch circuits which have their respective Q outputs connected to said correlation circuits.

18. Apparatus according to claim 17, in which said resetting means includes a stop-start circuit (43) constituted as an RS flipflop circuit having a start input and a stop input and an output connected both to the reset inputs of said counters and to said fourth, fifth and sixth timing circuits and to the reset inputs of said RS latch circuits, an OR-gate having its outputs being connected to said stop input and its input connected to said classification signal outputs.

19. Apparatus according to claim 18, in which said counters are pulse counters having counting and reset inputs and said fourth, fifth and sixth timing circuits are constituted as clock pulse counters connected to a source of clock pulses and having counting, enable, and reset inputs.

20. Apparatus according to claim 19, in which said counters have their counting inputs respectively connected to the outputs of said first, second and third monoflop timing circuits and said time lapse threshold logic circuit includes first, second and third count state comparators (47-49) having their inputs connected to the output of the said counter which responds to the output of said first monoflop and their outputs connected respectively to the S inputs of a corresponding number of said RS latch circuits, said first count state comparator providing an output for a count exceeding 2, said second count state comparator providing an output for a count state not less than 4 nor more than 12, and said third count state comparator provides an output for a count state exceeding 3, the outputs of said count state comparators being respectively connected to corresponding S inputs of latch circuits of said RS latch circuits.

21. Apparatus according to claim 20, in which a fourth count state comparator is connected to the output of the said counter which responds to said third monoflop for producing an output in response to a count state exceeding 4 and supplying said output to the S input of one of said RS latch circuits.

22. Apparatus according to claim 21, in which said correlation circuit includes a first AND-gate having its inputs connected to the respective outputs of the said RS latch circuits to which said second and fourth count state comparators are connected and its output connected to one of said classification signal outputs serving to provide speech classification signals.

23. Apparatus according to claim 21, in which said correlation circuit includes a second AND-gate having one input connected for receiving a negated output of said third monoflop timing circuit and another output connected for receiving a normal output of said second monoflop timing circuit, said AND-gate having its output connected to the counting circuit of said third counter, and in which a fifth count state comparator is connected to said third counter which fifth count state comparator is constituted to provide an output to the S input of one of said RS latch circuits for a count state exceeding 8.

24. Apparatus according to claim 20, in which first, second and third threshold value integrators are connected to the respective counters of said fourth, fifth and sixth timing circuits for respectively producing signals when time lapses of 1.6 S and 1.5 S and 3 S are detected furnishing said signals to S inputs of respective latch circuits of said array of RS latch circuits.

25. Apparatus according to claim 24, in which said time lapse threshold logic circuit includes means for connecting the enable input of said fourth timing circuit with an inverting output of said second monoflop timing circuit, means for connecting the enable input of said sixth timing circuit with a normal output of said second monoflop, and means for connecting the enable input of said fifth timing circuit in parallel with the counting input of said sixth timing circuit.

26. Apparatus according to claim 25, in which said correlation circuit includes a third AND-gate (60) having its inputs connected respectively to the outputs of said RS latch circuit connected to said first count state comparator and to said RS latch circuit connected to the output of said third threshold value indicator and an OR-gate (61) having inputs connected respectively to the output of said third AND-gate and to the outputs of said RS latch circuits connected respectively to said fifth count state comparator and said first threshold value integrator, the output of said OR-gate being connected to one of said classification signal outputs which serves to supply signals indicating said unidentifiable signal classification.

27. Apparatus according to claim 26, in which said correlation circuit includes a fourth AND-gate (62) having its inputs connected respectively to the said RS latch circuits connected to said third count state comparator and to said second threshold value integrator, a fifth AND-gate (64) having its inputs connected for respectively receiving a negated output of said RS latch circuit connected to said first count state comparator and a normal output from said RS latch circuit connected to said third threshold value integrator, and an OR-gate (65) having its inputs connected to the outputs of said fourth and fifth AND-gates, said OR-gate having its output connected to one of said signal classification outputs serving to provide music classification signals.