Patent US5745874A successfully refreshed from Google Patents.
US5745874A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Preprocessor for automatic speech recognition system
Assignee: National Semiconductor Corp
•
Filing: 1996-03-04
•
Granted: 1998-04-28
•
Expires: 2016-03-04
Abstract
Abstract A preprocessor for automatic speech recognition based upon auditory modeling includes a tapped delay line and a neural network in the form of a multilayer perceptron. The tapped delay line receives an analog speech signal and provides multiple time delayed samples thereof in parallel as inputs for the neural network. The single analog output of the neural network is suitable for interfacing with a signal processor for further processing of the speech information using spectral signal analysis so as to provide a speech representation with desirable characteristics of an auditory based spectral analysis model while simultaneously maintaining a standard analog signal interface.
Inventors
- 👤 William Shields Neely
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-31 18:28
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/02; G10L25/00; G10L25/27; G10L25/30
Claims Analysis
Total Claims: 22
Claim 1
Independent Claim
106 words
An apparatus including a preprocessor for automatic speech recognition based upon auditory modeling, said preprocessor comprising: a first signal processor for receiving a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith providing a plurality of sequential time interval samples of said speech signal; and a second signal processor, coupled to said first signal processor, for receiving and processing said plurality of sequential time interval samples of said speech signal together simultaneously and in accordance therewith providing a single processed signal which represents a sequence of preprocessed acoustic feature vectors corresponding to said speech spectrum envelope.
Claim 2
Independent Claim
106 words
An apparatus including a preprocessor for automatic speech recognition based upon auditory modeling, said preprocessor comprising: a first signal processor for receiving a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith providing a plurality of sequential time interval samples of said speech signal; and a second signal processor, coupled to said first signal processor, for receiving and processing said plurality of sequential time interval samples of said speech signal together simultaneously and in accordance therewith providing a single processed signal which represents a sequence of preprocessed acoustic feature vectors corresponding to said speech spectrum envelope.
Claim 3
Dependent Claim
35 words
The apparatus of claim 1, wherein said first signal processor comprises a tapped delay line and said plurality of sequential time interval samples of said speech signal comprises a plurality of sequentially time delayed signals.
Claim 4
Dependent Claim
76 words
The apparatus of claim 1, wherein said speech signal comprises an analog speech signal and said first signal processor comprises: an analog-to-digital converter for receiving and digitizing said analog speech signal to provide a digital speech signal; and a shift register, which includes a plurality of outputs and is coupled to said analog-to-digital converter, for receiving and selectively shifting said digital speech signal to provide said plurality of sequential time interval samples of said speech signal.
Claim 5
Dependent Claim
44 words
The apparatus of claim 1, wherein said second signal processor comprises a first neural network which includes a plurality of inputs for receiving said plurality of sequential time interval samples of said speech signal and a single output for providing said single processed signal.
Claim 6
Dependent Claim
15 words
The apparatus of claim 4, wherein said first neural network comprises a multiple layer perceptron.
Claim 7
Dependent Claim
37 words
The apparatus of claim 4, further comprising a second neural network, coupled to said first neural network, for receiving and further processing said single processed signal to provide a processed output which represents said speech spectrum envelope.
Claim 8
Dependent Claim
36 words
The apparatus of claim 6, wherein said first and second neural networks together receive and cooperatively process a plurality of training signals by back-propagating errors in said processed output through said second and first neural networks.
Claim 9
Dependent Claim
45 words
The apparatus of claim 1, further comprising a third signal processor, coupled to said second signal processor, for receiving and further processing said single processed signal to provide a processed output which includes a sequence of processed acoustic feature vectors representing said speech spectrum envelope.
Claim 10
Dependent Claim
16 words
The apparatus of claim 8, wherein said third signal processor comprises a mel-frequency cepstral coefficients generator.
Claim 11
Dependent Claim
28 words
The apparatus of claim 9, wherein said mel-frequency cepstral coefficients generator comprises a computer programmed to generate a plurality of auditory coefficients based upon said single processed signal.
Claim 12
Independent Claim
94 words
A method of preprocessing a signal for automatic speech recognition based upon auditory modeling, said method comprising the steps of: receiving and processing a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith generating a plurality of sequential time interval samples of said speech signal; and processing said plurality of sequential time interval samples of said speech signal together simultaneously and in accordance therewith generating a single processed signal which represents a sequence of preprocessed acoustic feature vectors corresponding to said speech spectrum envelope.
Claim 13
Independent Claim
94 words
A method of preprocessing a signal for automatic speech recognition based upon auditory modeling, said method comprising the steps of: receiving and processing a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith generating a plurality of sequential time interval samples of said speech signal; and processing said plurality of sequential time interval samples of said speech signal together simultaneously and in accordance therewith generating a single processed signal which represents a sequence of preprocessed acoustic feature vectors corresponding to said speech spectrum envelope.
Claim 14
Dependent Claim
79 words
The method of claim 11, wherein said step of receiving and processing a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith generating a plurality of sequential time interval samples of said speech signal comprises propagating said speech signal via a delay line and tapping said delay line to generate a plurality of sequentially time delayed signals as said plurality of sequential time interval samples of said speech signal.
Claim 15
Dependent Claim
94 words
The method of claim 11, wherein said speech signal comprises an analog speech signal and said step of receiving and processing a speech signal corresponding to an acoustic stimulus and having a speech spectrum envelope associated therewith and in accordance therewith generating a plurality of sequential time interval samples of said speech signal comprises the steps of: digitizing said analog speech signal to generate a digital speech signal; and selectively shifting said digital speech signal to generate a plurality of output signals as said plurality of sequential time interval samples of said speech signal.
Claim 16
Dependent Claim
81 words
The method of claim 11, wherein said step of processing said plurality of sequential time interval samples of said speech signal together simultaneously and in accordance therewith generating a single processed signal which represents a sequence of preprocessed acoustic feature vectors corresponding to said speech spectrum envelope comprises inputting said plurality of sequential time interval samples of said speech signal as a plurality of inputs to a first neural network and outputting a single output therefrom as said single processed signal.
Claim 17
Dependent Claim
76 words
The method of claim 14, wherein said step of inputting said plurality of sequential time interval samples of said speech signal as a plurality of inputs to a first neural network and outputting a single output therefrom as said single processed signal comprises inputting said plurality of sequential time interval samples of said speech signal as a plurality of inputs to a multiple layer perceptron and outputting a single output therefrom as said single processed signal.
Claim 18
Dependent Claim
32 words
The method of claim 14, further comprising the step of further processing said single processed signal with a second neural network to provide a processed output which represents said speech spectrum envelope.
Claim 19
Dependent Claim
41 words
The method of claim 16, further comprising the step of receiving and cooperatively processing with said first and second neural networks together a plurality of training signals by back-propagating errors in said processed output through said second and first neural networks.
Claim 20
Dependent Claim
35 words
The method of claim 11, further comprising the step of further processing said single processed signal to provide a processed output which includes a sequence of processed acoustic feature vectors representing said speech spectrum envelope.
Claim 21
Dependent Claim
47 words
The method of claim 18, wherein said step of further processing said single processed signal to provide a processed output which includes a sequence of processed acoustic feature vectors representing said speech spectrum envelope comprises further processing said single processed signal with a mel-frequency cepstral coefficients generator.
Claim 22
Dependent Claim
45 words
The method of claim 19, wherein said step of further processing said single processed signal with a mel-frequency cepstral coefficients generator comprises further processing said single processed signal with a computer programmed to generate a plurality of auditory coefficients based upon said single processed signal.
References Cited (Backward Citations)
US5315704A
Cited
US5479563A
Cited
US5586215A
Cited
US5461696A
Cited
US5581658A
Cited
US5509103A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2001-04-27 | 2001-10-27 | 2002-04-28 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2005-04-27 | 2005-10-27 | 2006-04-28 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2009-04-27 | 2009-10-27 | 2010-04-28 | $7700.00 | Window Closed / Lapsed |