A speech understanding system includes a language model comprising a combination of an N-gram language model and a context-free grammar language model. The language model stores information related to words and semantic information to be recognized. A module is adapted to receive input from a user and capture the input for processing. The module is further adapted to receive SALT application program interfaces pertaining to recognition of the input. The module is configured to process the SALT application program interfaces and the input to ascertain semantic information pertaining to a first portion of the input and output a semantic object comprising text and semantic information for the first portion by accessing the language model, wherein performing recognition and outputting the semantic object are performed while capturing continues for subsequent portions of the input.

Semantic object synchronous understanding implemented with speech application language tags

A computer-assisted language correction system including spelling correction functionality, misused word correction functionality, grammar correction functionality and vocabulary enhancement functionality utilizing contextual feature-sequence functionality employing an internet corpus.

Automatic context sensitive language correction and enhancement using an internet corpus

A language model (70) is used in a speech recognition system (60) which has access to a first, smaller data store (72) and a second, larger data store (74). The language model (70) is adapted by formulating an information retrieval query based on information contained in the first data store (72) and querying the second data store (74). Information retrieved from the second data store (74) is used in adapting the language model (70). Also, language models are used in retrieving information from the second data store (74). Language models are built based on information in the first data store (72), and based on information in the second data store (74). The perplexity of a document in the second data store (74) is determined, given the first language model, and given the second language model. Relevancy of the document is determined based upon the first and second perplexities. Documents are retrieved which have a relevancy measure that exceeds a threshold level.

Information retrieval and speech recognition based on language models

A method for generating concatenative speech uses a speech synthesis input to populate a triphone-indexed database that is later used for searching and retrieval to create a phoneme string acceptable for a text-to-speech operation. Prior to initiating the “real time” synthesis process, a database is created of all possible triphone contexts by inputting a continuous stream of speech. The speech data is then analyzed to identify all possible triphone sequences in the stream, and the various units chosen for each context. During a later text-to-speech operation, the triphone contexts in the text are identified and the triphone-indexed phonemes in the database are searched to retrieve the best-matched candidates.

Synthesis-based pre-selection of suitable units for concatenative speech

A system and method for improving the response time of text-to-speech synthesis utilizes "triphone contexts" (i.e., triplets comprising a central phoneme and its immediate context) as the basic unit, instead of performing phoneme-by-phoneme synthesis. Prior to initiating the "real time" synthesis, a database is created of all possible triphones (there are approximately 10000 in the English language) and their associated preselection costs. At run time, therefore, only the most likely candidates are selected from the triphone database, significantly reducing the calculations that are required to be performed in real time.

Method and system for preselection of suitable units for concatenative speech

Disclosed are a method and an apparatus for adapting, particularly reducing, the size of a language model, which comprises word n-grams, in a speech recognition system. The invention provides a mechanism to discard those n-grams for which the acoustic part of the system requires less support from the language model to recognize correctly. The proposed method is suitable for identifying those trigrams in a language model for the purpose of discarding during the built-time of the system. Provided is also another automatic classification scheme for words which allows the compression of a language model, but under retention of accuracy. Moreover it allows an efficient usage of sparsely available text corpora because even singleton trigrams are used when they are helpful. No additional software tools are needed to be developed because the main tool, the fast match scoring, is a module readily available in the known recognizers themselves. Further improvement of the method is accomplished by classification of words according to the common text in which they occur as far as they distinguish from each other acoustically. The invention opens the possibility to make speech recognition available in low-cost personal computers (PC's), even in portable computers like Laptops.

Method and apparatus for adapting the language model's size in a speech recognition system

The invention concerns a computer system for speech recognition with a word table and a triphone table in its memory. In response to a new word not contained in the word table and which is encountered by the computer system during recognition of an unknown spoken text, the fenemic baseform of this new word may be synthetically produced with the aid of a triphone table. For this purpose, the new word is decomposed into triphones, and the fenemic sequences associated with the individual triphones are read from the triphone table. The fenemic sequences are concatenated according to the phonetic baseform of the new word such that the fenemic baseform of the new word is obtained.

Computer system for speech recognition

An experimental speech recognition system was developed for German using an already existing technology reported elsewhere (1). The system recognizes complete sentences when the words are spoken with a small pause in between. The user has to train the system in advance by uttering 110 short sentences. The size of the system's vocabulary is presently limited to about 1300 words, with a coverage being 58% of the running words in a newspaper text from the commercial discourse domain. The system uses 60 allophones and a statistical trigram model made out of a text corpus of 14 million words. The recognition accuracy is over 95%.

The SPRING speech recognition system for German.

Es ist ein Computersystem zur Spracherkennung beschrieben, bei dem eine Wort-Tabelle und eine Triphone-Tabelle im Speicher vorhanden ist. Stost das Computersystem bei der Erkennung eines unbekannten gesprochenen Textes auf ein neues Wort, das in der Wort-Tabelle nicht enthalten ist, so kann die fenemische Grundform dieses neuen Wortes mit Hilfe der Triphone-Tabelle synthetisch erzeugt werden. Zu diesem Zweck wird das neue Wort in Triphones zerlegt und es werden die zu den einzelnen Triphones zugehorigen fenemischen Folgen aus der Triphone-Tabelle ausgelesen. Die fenemischen Folgen werden dann entsprechend der phonetischen Grundform des neuen Wortes aneinandergefugt, wodurch die fenemische Grundform des neuen Wortes entsteht.

Karlheinz Mohr

Papers

Method and apparatus for adapting the language model's size in a speech recognition system

Computer system for speech recognition

The SPRING speech recognition system for German.

Computersystem zur Spracherkennung

Der IBM Spracherkennungsprototyp TANGORA - Anpassung an die deutsche Sprache