Acoustic scene interpretation systems and related methods
An acoustic-scene interpretation apparatus can have a transducer configured to convert an acoustic signal to a corresponding electrical signal. A feature extractor can receive a sequence of frames representing the electrical signal and extract a plurality of acoustic features corresponding to each frame. An acoustic-scene classifier can be configured to determine a most-likely acoustic state for each frame in the sequence of frames in correspondence with the respective plurality of acoustic features corresponding to the frame and a selected probability distribution of duration of an acoustic state for each of one or more classes of acoustic scenes. Each respective probability distribution of duration can correspond to a selected class of acoustic scenes. The correspondence between acoustic state and probability distribution of duration can be learned from training data corresponding to each of a plurality of classes of acoustic scenes. Related methods also are disclosed.
1. An acoustic-scene interpretation apparatus, comprising:
a transducer configured to convert an incoming acoustic signal to a corresponding electrical signal;
an acoustic-feature extractor configured to receive a representation of the corresponding electrical signal comprising a sequence of frames and to extract a plurality of acoustic features corresponding to each respective frame;
an acoustic-scene classifier configured to determine a most-likely acoustic state for each current frame in the sequence of frames in correspondence with the plurality of acoustic features corresponding to a current frame and a selected probability distribution of duration of an acoustic state for each of one or more classes of acoustic scenes; and
a recognition engine configured to determine a transcription of the incoming acoustic signal in at least partial correspondence with each respective most-likely acoustic state determined for each frame in the sequence of frames.
2. The acoustic-scene interpretation apparatus according to claim 1 , wherein the plurality of acoustic features comprises at least one linear combination of constituent acoustic features correlated to a selected class of acoustic signal.
3. The acoustic-scene interpretation apparatus according to claim 2 , wherein the class of acoustic signal comprises one or more of speech, music, and silence.
4. The acoustic-scene interpretation apparatus according to claim 1 , further comprising a detector configured to detect a beginning or a terminus of an acoustic event corresponding to a change in the most-likely acoustic state.
5. The acoustic-scene interpretation apparatus according to claim 1 , wherein the one or more classes of acoustic scenes comprises at least one of human speech, music, vehicle traffic and animal activity.
6. The acoustic-scene interpretation apparatus according to claim 1 , wherein the acoustic-scene classifier comprises a voice-activity detector.
7. The acoustic-scene interpretation apparatus according to claim 1 , wherein the recognition engine comprises at least one of a speech-recognition engine, a music-recognition engine, a vehicle-recognition engine and an animal-recognition engine.
8. The acoustic-scene interpretation apparatus according to claim 1 , wherein each respective probability distribution of duration of an acoustic state corresponds to a selected class of acoustic scenes.
9. The acoustic-scene interpretation apparatus according to claim 1 , wherein the correspondence between the most-likely acoustic state for each current frame and each respective probability distribution of duration comprises learned information gathered from one or more observed acoustic signals emitted from each of a plurality of classes of acoustic scenes.
10. A method of interpreting an observed acoustic scene, the method comprising:
converting an incoming acoustic signal to a corresponding electrical signal;
receiving a representation of the corresponding electrical signal comprising a sequence of frames and extracting from the representation a plurality of acoustic features corresponding to each respective frame;
determining a most-likely acoustic state for each current frame in the sequence of frames in correspondence with the respective plurality of acoustic features corresponding to a current frame and a selected probability distribution of duration of an acoustic state for each of one or more classes of acoustic scenes; and
determining a transcription of the incoming acoustic signal in at least partial correspondence with each respective most-likely acoustic state determined for each frame in the sequence of frames.
11. The method according to claim 10 , wherein the plurality of acoustic features comprises at least one linear combination of constituent acoustic features correlated to a selected class of acoustic signal.
12. The method according to claim 11 , wherein the class of acoustic signal comprises one or more of speech, music, and silence.
13. The method according to claim 10 , further comprising detecting a beginning or a terminus of an acoustic event corresponding to a change in the most-likely acoustic state.
14. The method according to claim 10 , wherein the one or more classes of acoustic scenes comprises at least one of human speech, music, vehicle traffic and animal noise.
15. The method according to claim 10 , wherein determining a most-likely acoustic state comprises discriminating between or among at least two of speech, music, and silence.
16. The method according to claim 10 , wherein determining a transcription comprises at least one of identifying a human- or a machine-recognizable representation of the incoming acoustic signal, identifying an attribute characteristic of a class of music, identifying an attribute characteristic of a class of vehicle, and identifying an attribute characteristic of a class of animal.
17. The method according to claim 10 , wherein each respective probability distribution of an acoustic state corresponds to a given class of acoustic scenes.
18. The method according to claim 10 , wherein correspondence between the most-likely acoustic state for each current frame and each respective probability distribution of duration comprises learned information gathered from one or more observed acoustic signals corresponding to each of a plurality of classes of acoustic scenes.