Abstract: The present invention proposes a method, system and computer program for speech recognition. According to one embodiment, a method is provided wherein, for an expected input string divided into a plurality of expected string segments, a speech segment is received for each expected string segment. Speech recognition is then performed separately on each said speech segment via the generation, for each said speech segment, of a segment n-best list comprising n highest confidence score results. A global n-best list is then generated corresponding to the expected input string utilizing the segment n-best lists and a final global speech recognition result corresponding to said expected input string is determined via the pruning of the results of the global n-best list utilizing a pruning criterion.

Abstract: A system and method for generating synthetic speech, which operates in a computer implemented Text-To-Speech system. The system comprises at least a speaker database that has been previously created from user recordings, a Front-End system to receive an input text and a Text-To-Speech engine. The Front-End system generates multiple phonetic transcriptions for each word of the input text, and the TTS engine uses a cost function to select which phonetic transcription is the more appropriate for searching the speech segments within the speaker database to be concatenated and synthesized.

Abstract: The present invention proposes a method, system and computer program for speech recognition. According to one embodiment, a method is provided wherein, for an expected input string divided into a plurality of expected string segments, a speech segment is received for each expected string segment. Speech recognition is then performed separately on each said speech segment via the generation, for each said speech segment, of a segment n-best list comprising n highest confidence score results. A global n-best list is then generated corresponding to the expected input string utilizing the segment n-best lists and a final global speech recognition result corresponding to said expected input string is determined via the pruning of the results of the global n-best list utilizing a pruning criterion.

Abstract: A system and method for generating synthetic speech, which operates in a computer implemented Text-To-Speech system. The system comprises at least a speaker database that has been previously created from user recordings, a Front-End system to receive an input text and a Text-To-Speech engine. The Front-End system generates multiple phonetic transcriptions for each word of the input text, and the TTS engine uses a cost function to select which phonetic transcription is the more appropriate for searching the speech segments within the speaker database to be concatenated and synthesized.

Abstract: Disclosed herein is a method for automatically filtering a corpus of documents containing textual and non-textual information of a natural language. According to the method, through a first dividing step (101), the document corpus is divided into appropriate portions. At a following determining step (105), for each portion of the document corpus, there is determined a regularity value (VR) measuring the conformity of the portion with respect to character sequences probabilities predetermined for the language considered. At a comparing step (107), each regularity value (VR) is then compared with a threshold value (VT) to decide whether the conformity is sufficient. Finally, at a rejecting step (111), any portion of the document corpus whose conformity is not sufficient is rejected and removed from the corpus. An apparatus for carrying out such a method is also disclosed.

Abstract: A computerized method is provided for adding a new word to a vocabulary of a speech system, the vocabulary comprising words and corresponding acoustic patterns for a language or language domain. Within a determination step for the new word, a regularity value is determined which measures the conformity with respect to the pronunciation in the language or language domain. In a comparison step, the regularity value is compared to a threshold value to decide whether the conformity is insufficient. Only in the affirmative case of insufficient conformity, a prompting step is performed, prompting for additional information on the pronunciation of the new word. Finally, in an extension step, the new word and an acoustic pattern of the new word are added to the vocabulary.

Abstract: Methods and systems for testing speech recognition systems are disclosed in which the speech recognition device to be tested is directly monitored in accordance with a text-to-speech device. The collection of reference texts to be used by the speech recognition device is provided by a text-to-speech device preferably, in one embodiment, implemented within the same computer system. In such an embodiment, a digital audio file stored within a storage area of a computer system is generated from a reference text using a text-to-speech device. The digital audio file is later read using a speech recognition device to generate a decoded (or recognized) text representative of the reference text. The reference text and the decoded text are compared in an alignment operation and an error report representative of the recognition rate of the speech recognition device is finally generated.

Abstract: Disclosed herein is a method for automatically filtering a corpus of documents containing textual and non-textual information of a natural language. According to the method, through a first dividing step (101), the document corpus is divided into appropriate portions. At a following determining step (105), for each portion of the document corpus, there is determined a regularity value (VR) measuring the conformity of the portion with respect to character sequences probabilities predetermined for the language considered. At a comparing step (107), each regularity value (VR) is then compared with a threshold value (VT) to decide whether the conformity is sufficient. Finally, at a rejecting step (111), any portion of the document corpus whose conformity is not sufficient is rejected and removed from the corpus. An apparatus for carrying out such a method is also disclosed.

Abstract: A method for supporting customized tokenization of domain-specific text acomprises the steps of: loading domain-specific tokenization rules corresponding to the customized tokenization of the domain-specific text; tokenizing the domain-specific text using the loaded domain-specific tokenization rules; and, further tokenizing the domain-specific text using general purpose tokenization rules. The loading step of the inventive method can comprise: loading a speech recognition vocabulary; and, loading domain-specific tokenization rules corresponding to the speech recognition vocabulary. In addition, the tokenizing step can comprise identifying each substring in the domain-specific text matching a regular expression having a corresponding replacement pattern in the loaded domain-specific tokenization rules, and replacing each substring identified in the identifying step with the replacement pattern corresponding to the matched regular expression.

Abstract: A pitch detector to adjust long term prediction in a pulse excitation speech coder. A residual signal r(n) is first derived from the speech signal s(n) by short term filtering. Then, r(n) is processed to calculate a prediction error signal e(n) which is subsequently pulse excitation encoded. The processing of e(n) entails prediction of a residual by measuring a pitch related factor M, employing two steps. First calculating a coarse M value through peak clipping and sign transition detection, and then adjusting the M value by autocorrelation--calculations about the roughly spaced peaks.