Method and system for speech recognition of short words and sentences based on dtw and gmm
A speech recognition and speech technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problem that the Gaussian mixture model GMM cannot be applied to short-term speech recognition, the discrete vector set Gaussian mixture model GMM modeling, and the inability to use DTW distance, etc. problem, to achieve the effect of reducing the amount of calculation, speeding up the real-time performance, and high recognition efficiency
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0054] The embodiment of the present invention provides a short-word speech recognition method based on DTW and GMM, which enables the Gaussian mixture model GMM to use the DTW distance and directly process data of different lengths in the time domain, with high recognition efficiency and high recognition accuracy. The advantages.
[0055] like figure 1 As shown, a DTW and GMM-based speech recognition method for short words and sentences provided by an embodiment of the present invention includes the following steps.
[0056] Step 100: Acquire the voice and audio data collected at the current stage.
[0057] Step 110: Determine a recognition result corresponding to the voice and audio data collected at the current stage according to the voice and audio data collected at the current stage, the Gaussian mixture model with built-in DTW distance, and the voice and audio database.
[0058] Wherein, the voice and audio database includes n short words and sentences, the Gaussian mi...
Embodiment 2
[0076] In order to achieve the above purpose, the embodiment of the present invention provides a short phrase speech recognition system based on DTW and GMM, such as figure 2 shown, including:
[0077] The data acquisition module 200 is used for acquiring the voice and audio data collected in the current stage.
[0078] The recognition module 210 is configured to determine the recognition result corresponding to the voice and audio data collected at the current stage according to the voice and audio data collected at the current stage, the Gaussian mixture model with built-in DTW distance, and the voice and audio database.
[0079] Wherein, the voice and audio database includes n short words and sentences, the Gaussian mixture model of the built-in DTW distance includes K Gaussian models, and K=n; the Gaussian mixture model of the built-in DTW distance is based on the voice and audio database, DTW algorithm and Gaussian mixture model are constructed.
[0080] Wherein, the i...
Embodiment 3
[0086] The embodiment of the present invention provides a short-word speech recognition method using a Gaussian mixture model in the time domain, and the method has the advantages of high recognition accuracy and high computational efficiency.
[0087] Now, machine learning based on large-scale multi-level neural networks can be applied to the field of speech recognition, and has achieved successful accuracy that traditional hidden Markov models (HMMs) cannot achieve, but only for continuous speech recognition. In the field of short-word and short-sentence speech recognition, the traditional DTW algorithm (Dynamic Time Warping, DTW) still occupies a place, because compared with the deep learning method, the DTW algorithm has the advantages of simplicity, less dependence on data accumulation, and recognition accuracy theory. It can also meet the needs and other characteristics.
[0088] The short-word speech recognition method based on DTW algorithm is briefly described as foll...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More