Speech recognition based captioning system

A method of displaying text information corresponding to a speech portion of audio signals of a television program as a closed caption on an video display device, the method comprising the steps of:

• decoding the audio signals of the television program;
• filtering the audio signals by using a spectral subtraction method to extract the speech portion;
• receiving a training text as a part of the television signal, the training text corresponding to a part of the speech portion of the audio signals;
• generating a hidden Markov model from the training text and the part of the speech portion of the audio signals;
• parsing the audio speech signals into phonemes based on the generated Hidden Markov model;
• identifying words in a database corresponding to grouped phonemes;
• and
• converting the identified words into text data for presentation on the display of the audio-visual device as closed captioned textual data.

Apparatus for displaying text information corresponding to a speech portion of audio signals of television program signals as a closed caption on an video display device, the apparatus comprising:

• a decoder which separates the audio signals from the television program signals;
• a spectral subtraction speech filter which identifies portions of the audio signals that include speech components and separates the identified speech component signals from the audio signals;
• a phoneme generator including a speaker-dependent speech recognition system which parses the speech portion into phonemes in accordance with a speech model which include s a hidden Markov model, the phoneme generator including:
• means for receiving a training text as a part of the television signal, the training text corresponding to a part of the speech portion of the audio signals;
• means for adapting the hidden Markov model based on the training text and the part of the speech portion of the audio signals corresponding to the training text;
• and
• means for applying the updated hidden Markov model to parse the speech portion of the audio signals to provide the phonemes;
• a database of words, each word being identified as corresponding to a discrete set of phonemes;
• a word matcher which groups the phonemes provided by the phoneme generator and identifies words In the database corresponding to the grouped phonemes;
• and
• a formatting processor that converts the identified words into text data for display on the display device as the closed caption.

A computer readable carrier including computer program instructions tangibly embodied thereon that cause a computer to implement a method for displaying text information corresponding to a speech portion of audio signals of television program signals as a closed caption on an video display device, the method comprising the steps of:

• decoding the audio signals of the television program;
• filtering the audio signals by using a spectral subtraction method to extract the speech portion;
• parsing the speech portion into discrete speech components in accordance with a speech model and grouping the parsed speech components employing a speaker dependent model to provide phonemes as the parsed speech components wherein the speaker dependent model employs a hidden Markov model;
• receiving a training text as a part of the television signal, the training text corresponding to a part of the speech portion of the audio signals:
• updating the hidden Markov model based on the training text and the part of the speech portion of the audio signals corresponding to the training text;
• applying the updated hidden Markov model to parse the speech portion of the audio signals to provide the phonemes;
• identifying words in a database corresponding to the grouped speech components;
• and
• converting the identified words into text data for display on the display device as the closed caption.