Today's vending machines lack the personalized analysis technology for user interaction. This technology will attract users to engage in consumption via the vending machine. Therefore, the technology has been filed for patent applications in countries with a large market share for consumption via vending machines such as Taiwan, United States and Japan. Hopefully, this technology will be applied to the domestic trial product delivery market and profits can be gained via patent licensing. Meanwhile, for the Web delivery platform industrialists, such as Digwow and iTRY Before I Buy!, the current goal of technology transfer is to facilitate industrial development and enhance output services for relevant industrialists. Patent application: 1. Commodity selection system and methods and relevant computer program products (Taiwan / US / Japan) 2. Trial delivery system, methods and recording media that can be read by the computer (Taiwan / US / Japan)
技術現況敘述-英文: Today's vending machines lack the personalized analysis technology for user interaction. This technology will attract users to engage in consumption via the vending machine. Therefore, the technology has been filed for patent applications in countries with a large market share for consumption via vending machines such as Taiwan, United States and Japan. Hopefully, this technology will be applied to the domestic trial product delivery market and profits can be gained via patent licensing. Meanwhile, for the Web delivery platform industrialists, such as Digwow and iTRY Before I Buy!, the current goal of technology transfer is to facilitate industrial development and enhance output services for relevant industrialists. Patent application: 1. Commodity selection system and methods and relevant computer program products (Taiwan / US / Japan) 2. Trial delivery system, methods and recording media that can be read by the computer (Taiwan / US / Japan)
An intelligent information retrieval system for finding information components corresponding to an input query word. The system includes a synonym block for finding synonymous indexed keywords of the query word; an information indexing block for finding corresponding information components based on the indexed keyword; a component ranking-filtering block for ranking and filtering the found information components and outputting the desired information components being selected; a synonym adjusting block for adjusting the fuzzy mechanism of the synonym block based on the found information components; and a filtering adjusting block for adjusting the fuzzy mechanism of the ranking-filtering block based on the found information components. The aforementioned synonym block and information-indexing block are implemented by neuro-fuzzy networks for accelerating parallel processing and automatic learning. Further, the synonym block can tolerate input errors by way of query word encoding and position shift compensation.
技術摘要-英文: An intelligent information retrieval system for finding information components corresponding to an input query word. The system includes a synonym block for finding synonymous indexed keywords of the query word; an information indexing block for finding corresponding information components based on the indexed keyword; a component ranking-filtering block for ranking and filtering the found information components and outputting the desired information components being selected; a synonym adjusting block for adjusting the fuzzy mechanism of the synonym block based on the found information components; and a filtering adjusting block for adjusting the fuzzy mechanism of the ranking-filtering block based on the found information components. The aforementioned synonym block and information-indexing block are implemented by neuro-fuzzy networks for accelerating parallel processing and automatic learning. Further, the synonym block can tolerate input errors by way of query word encoding and position shift compensation.
本發明係有關一種詞語驗證方法及系統,其首先抽取出語音信號中的特徵參數向量序列,再經過語音辨識後可取得至少一候選詞,依照候選詞的詞彙內容所對應的驗證單元將該語音信號切割為對應於驗證單元的語音音段,並求出這些語音音段的驗證用特徵參數向量序列,之後依序使用這些驗證用特徵參數向量序列進行驗證,以產生驗證分數;這個驗證方法係使用語音音段所相對應之驗證單元的類神經網路來計算驗證分數,而這個類神經網路係為一多層感知,驗證分數係使用驗證用特徵參數向量序列對此多層感知進行前饋動作求得;在合併所有語音音段之驗證分數以取得詞語驗證分數後,即可根據一預先定義好的門檻值來決定接受或拒絕該候選詞。A method and system for utterance verification is disclosed. It can extract a sequence of feature vectors from speech signal. At least one candidate string is obtained after the process of speech recognition by using the sequence of feature vectors. Then, the speech signal is segmented into speech segments according to the verification-unit-specified structure of candidate string, and that makes each speech segment corresponding to a verification-unit. After calculating the sequences of verification-specific feature vectors for these speech segments, sequentially use these sequences to generate the verification scores of speech segments in verification process. This invention uses neural networks for the calculations of verification scores. Where each neural network is a Multi-Layer Perceptron and was developed for each verification-unit. Verification score is obtained through using the feed-forward process of MLP. Finally, utterance verification score is obtained by combining all of the verification scores of speech segments and is used to compare with a pre-defined threshold for the decision of acceptance or rejection of the candidate string.
技術摘要-中文: 本發明係有關一種詞語驗證方法及系統,其首先抽取出語音信號中的特徵參數向量序列,再經過語音辨識後可取得至少一候選詞,依照候選詞的詞彙內容所對應的驗證單元將該語音信號切割為對應於驗證單元的語音音段,並求出這些語音音段的驗證用特徵參數向量序列,之後依序使用這些驗證用特徵參數向量序列進行驗證,以產生驗證分數;這個驗證方法係使用語音音段所相對應之驗證單元的類神經網路來計算驗證分數,而這個類神經網路係為一多層感知,驗證分數係使用驗證用特徵參數向量序列對此多層感知進行前饋動作求得;在合併所有語音音段之驗證分數以取得詞語驗證分數後,即可根據一預先定義好的門檻值來決定接受或拒絕該候選詞。A method and system for utterance verification is disclosed. It can extract a sequence of feature vectors from speech signal. At least one candidate string is obtained after the process of speech recognition by using the sequence of feature vectors. Then, the speech signal is segmented into speech segments according to the verification-unit-specified structure of candidate string, and that makes each speech segment corresponding to a verification-unit. After calculating the sequences of verification-specific feature vectors for these speech segments, sequentially use these sequences to generate the verification scores of speech segments in verification process. This invention uses neural networks for the calculations of verification scores. Where each neural network is a Multi-Layer Perceptron and was developed for each verification-unit. Verification score is obtained through using the feed-forward process of MLP. Finally, utterance verification score is obtained by combining all of the verification scores of speech segments and is used to compare with a pre-defined threshold for the decision of acceptance or rejection of the candidate string.