Phoneme-level low-power consumption spoken language assessment and defect diagnosis method

An evaluation method and low-power technology, applied in voice analysis, instruments, etc., can solve problems such as switching troubles, inability to objectively score the quality of pronunciation, and students not knowing how to improve, so as to reduce load and power consumption, and avoid network problems. Traffic consumption, the effect of reducing hardware requirements

CN103985392AInactive Publication Date: 2014-08-13柳超
11 Cites 15 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Publication Date
2014-08-13
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a phoneme-level low-power consumption spoken language assessment and defect diagnosis method. The method comprises the steps of (1) conducting acoustic feature extraction on user pronunciation and standard pronunciation to obtain feature vector sequences; (2) decoding the feature vector sequence of the user pronunciation and the feature vector sequence of the standard pronunciation with the Viterbi algorithm based on a weighted finite state transducer Q to obtain the mapping relations between the feature vector sequences and a phoneme sequence, the alignment alpha of the user pronunciation and the alignment (as described in the specification) of the standard pronunciation; (3) comparing the alignment alpha and the feature vector sequence of the user pronunciation with the alignment (as described in the specification) and the feature vector sequence of the standard pronunciation so that user pronunciation quality can be assessed, wherein the higher the goodness of fit is, the better the pronunciation quality is. The method has the advantages that complicated operation is executed in a server, and therefore the load and power consumption of a terminal are effectively reduced; the terminal can be used without networking, and therefore network flow consumption is avoided; phonemes with pronunciation flaws can be recognized and targeted practice can be provided.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention relates to the technical field of computer-aided language learning and speech recognition, in particular to a phonetic-level low-power spoken language evaluation and defect diagnosis method. The phoneme-level technology enables scoring and feedback to be refined to the phoneme level: after the user reads a series of texts, the core phonemes that hinder the user's pure pronunciation can be identified, and corresponding training materials and targeted exercises can be provided. It can be adapted to the learning of English, Chinese, Spanish and other languages, as well as the diagnosis and evaluation of patients with language barriers. Background technique

[0002] Language learning is about imitation, especially phonetics. Taking English as an example, in order to practice pure oral English, the best way is to follow the pure pronunciation of native English speakers. Many existing tutorials or tutoring materials are based on this. In gene...

Examples

Embodiment Construction

[0040] First, the relationship between the first embodiment and the second embodiment will be described. In the first embodiment, the standard voice is not used, so only the user's voice is processed, and in the second embodiment, the standard voice is used, so it is necessary to increase the step of processing the standard voice (this step can be done in the server terminal can also be performed at the user terminal). Due to the addition of standard voices, the evaluation methods for user voices are also different.

[0041] When introducing the embodiments of the present invention, it is necessary to introduce related technologies in the field of speech processing, in order to facilitate a clearer understanding of the present invention.

[0042] Such as figure 1 As shown, the process of processing speech processing and extracting feature vectors is roughly divided into the following three steps:

[0043]i, the sound wave data (waveform) of the original collected sound abou...