Methods and apparatus for non-acoustic speech characterization and recognition

• US 6,006,175 A
• Filed: 02/06/1996
• Issued: 12/21/1999
• Est. Priority Date: 02/06/1996
• Status: Expired due to Term

• Current Assignee: Lawrence Livermore National Security LLC (Government of the United States of America)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Holzrichter, John F.
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/597,596
• Time in Patent Office: 1,414 Days
• Field of Search: 395/2.1, 395/2.16, 395/2.27, 395/2.37, 395/2.67, 395/2.17, 395/2.15, 395/2.14, 704/205-208, 704/201, 704/218, 704/228, 704/258
• US Class Current: 704/208
• CPC Class Codes: A61B 5/0507 using microwaves or terahertz waves
• CPC Class Codes: G01N 2291/02491 Materials with nonlinear acoustic properties
• CPC Class Codes: G01N 2291/02872 Pressure
• CPC Class Codes: G06F 18/256 of results relating to different input data, e.g. multimodal recognition
• CPC Class Codes: G06V 40/10 Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
• CPC Class Codes: G10L 15/24 Speech recognition using non-acoustical features

Abstract

By simultaneously recording EM wave reflections and acoustic speech information, the positions and velocities of the speech organs as speech is articulated can be defined for each acoustic speech unit. Well defined time frames and feature vectors describing the speech, to the degree required, can be formed. Such feature vectors can uniquely characterize the speech unit being articulated each time frame. The onset of speech, rejection of external noise, vocalized pitch periods, articulator conditions, accurate timing, the identification of the speaker, acoustic speech unit recognition, and organ mechanical parameters can be determined.

Claims

1. A method of speech characterization of speech by a speaker, comprising:

directing electromagnetic (EM) radiation toward speech organs of the speaker;

detecting electromagnetic (EM) radiation scattered from the speech organs to measure speech organ conditions to obtain EM speech information;

detecting acoustic speech output from the speaker to obtain acoustic speech information;

combining the EM speech information with the acoustic speech information using a speech characterization algorithm.

2. The method of claim 1 wherein the speech is selected from normal sounded speech, whispered speech, and non-sounded speech.

3. The method of claim 1 wherein acoustic speech output of the speaker is detected using at least one acoustic microphone.

4. The method of claim 3 further comprising measuring acoustic pressure or sound intensity over a plurality of sampling times to obtain amplitude vs. time, frequency, zero crossing times, energy per time interval, and LPC or cepstral coefficients of acoustic speech.

5. The method of claim 1 wherein acoustic speech output of the speaker is detected using at least one EM wave microphone to detect acoustic vibrations.

6. The method of claim 1 wherein EM radiation is directed to and detected from the speech organs using an EM wave transmitting and receiving system.

7. The method of claim 6 wherein the EM wave generating, transmitting and detecting system is an RF, microwave, millimeter wave, infrared, or visible wave EM sensor.

8. The method of claim 7 wherein the EM sensor is operated in a time of flight, non-coherent mode.

9. The method of claim 8 wherein the EM sensor is range-gated.

10. The method of claim 7 wherein the EM sensor is operated in a coherent mode.

11. The method of claim 10 wherein the EM sensor is operated in a homodyne, heterodyne or other interferometric coherent detection mode.

12. The method of claim 7 where the EM sensor is operated in a field disturbance mode, with or without a range gate, with time filtered output.

13. The method of claim 1 further comprising controlling the time of generation, transmission and detection of the EM radiation and of substantially simultaneous reception of the acoustic speech output.

14. The method of claim 1 further comprising creating feature vectors describing features of the acoustic speech output and EM sensor measured speech organ conditions during a defined time frame of speech.

15. The method of claim 14 further comprising storing, in the feature vector, the start time, duration time, and end time of the defined time frame of each feature vector.

16. The method of claim 14 further comprising associating information contained in the feature vector with information from other instruments or devices for the purposes of synchronization of timing.

17. The method of claim 14 further comprising storing the feature vectors in an electronic library.

18. The method of claim 17 further comprising performing time alignment of the feature vector of a particular speaker, and comparing the time aligned feature vector of the particular speaker with feature vectors in the library.

19. The method of claim 14 further comprising producing a feature vector for one or more speakers, averaging the feature vectors of the one or more speakers, and storing the averaged feature vector in a library.

20. The method of claim 14 further comprising normalizing and quantizing the speaker'"'"'s feature vectors to those of a reference speaker or group of speakers.

21. The method of claim 14 further comprising producing feature vectors for at least one of position and velocity of at least one of the velum, jaw, tongue, glottal tissue, and lips.

22. The method of claim 14 further comprising forming a single or multiple speech frame feature vector which defines a syllable like unit, a phoneme, a PLU, a diphone, a triphone, an acoustic unit, a word, or a word sequence.

23. The method of claim 14 further comprising applying a statistical technique or pattern matching technique to the feature vector to identify a syllable like unit, a phoneme, a PLU, a diphone, a triphone, an acoustic unit, a word, or a word sequence.

24. The method of claim 23 wherein the statistical technique is a Hidden Markov Model technique or a neural network technique.

25. The method of claim 23 wherein the pattern matching technique is a phonetic-template matching technique.

26. The method of claim 23 wherein the algorithm uses a joining or excluding method of identification by comparing feature vectors identified using conventional acoustic techniques to those identified using non-acoustic techniques to obtain a higher overall probability of identification.

27. The method of claim 14 further comprising forming the feature vector by first forming separate acoustic and EM feature vectors, and then combining the separate acoustic and EM feature vectors.

28. The method of claim 14 further comprising identifying sound changes and EM signal changes to define a new feature vector defined by changes from a referenced feature vector.

29. The method of claim 14 further comprising identifying acoustic changes and EM signal changes, compared to those of the last time frame, to define a new speech time frame.

30. The method of claim 14 further comprising automatically forming the feature vector.

31. The method of claim 14 further comprising creating a feature vector that describes a defined condition and change from the defined condition of position and/or velocity of at least one speech organ during a plurality of speech time frames.

32. The method of claim 14 further comprising forming a feature vector for velocity and acceleration over a plurality of time frames.

33. The method of claim 14 further comprising identifying a speaker from patterns of feature vectors formed by that speaker, over a sequential series of speech time frames.

34. The method of claim 1 further comprising obtaining organ velocity or acceleration information from the detected EM radiation.

35. The method of claim 1 further comprising measuring other speech information than the EM speech information and acoustic speech information and combining the other speech information with the EM speech information and acoustic speech information.

36. The method of claim 1 further comprising determining a set of mechanical parameters of the vocal system from the EM speech information and acoustic speech information for vocal system modeling.

37. The method of claim 1 wherein the algorithm determines the onset of speech, end of speech, speech period, pauses, speech rate, and extraneous noise.

38. The method of claim 1 wherein the algorithm determines the presence of voiced or unvoiced speech.

39. The method of claim 1 further comprising measuring organ contact as one organ touches another and strongly changes the EM wave reflecting condition because of changing resonator or boundary condition effects.

40. The method of claim 1 further comprising generating and transmitting a series of known wavelengths to detect organ interface spacings using coherent reflections and transmissions from the tissue and tissue interfaces.

41. Apparatus for speech characterization of speech by a speaker, comprising:

at least one electromagnetic (EM) wave generating, transmitting and detecting unit for directing EM waves toward and detecting EM waves scattered from speech organs of the speaker to obtain EM speech information;

at least one microphone for detecting acoustic speech output from the speaker to obtain acoustic speech information;

means for combining the EM speech information with the acoustic speech information using a speech characterization algorithm.

42. The apparatus of claim 41 wherein each EM wave generating, transmitting and receiving unit is a RF, microwave, millimeter wave, infrared, or visible wave radar.

43. The apparatus of claim 41 wherein each microphone is an acoustic microphone or an EM microphone.

44. The apparatus of claim 41 further comprising a structure for mounting the at least one EM wave generating, transmitting and detecting unit and at least one microphone such that they can detect the conditions of the speech organs of the speaker.

45. The apparatus of claim 41 further comprising means for controlling the time of generation, transmission, and reception of the EM waves and of substantially simultaneous reception of the acoustic speech output.

46. The apparatus of claim 42 wherein the EM unit is a time of flight, non-coherent radar, or a field disturbance sensor, with or without a range gate, with time filtered output, or a coherent radar.

47. The apparatus of claim 42 wherein the EM unit is a range-gated radar.

48. The apparatus of claim 42 wherein the EM unit is a homodyne, heterodyne or other interferometric coherent detection EM sensor.