Abstract: A system for selecting a rich advertisement for display to a user is provided. The system may include an advertisement engine with a first selection module configured to select a list of text advertisements for a text slate based on a query entered by the user and determine a first expected revenue of the text slate according to a first auction of text advertisements. The advertisement engine may also include a second selection module configured to select a rich advertisement for a mixed slate based on the query entered by the user and determine a second expected revenue of the mixed slate. Further, the advertisement engine may determine whether to display the text slate or the mixed slate based on the first expected revenue and the second expected revenue.

Abstract: Embodiments are directed towards employing long and short term historical user click propensity behaviors to adapt or filter a number of advertisements displayed and their location on a search results' page. A network device tracks a user's short and long term historical click behaviors. For a given search query for the user, a variety of candidate advertisements are selected. A normalized click-through rate (COEC) is estimated for each advertisement. The COEC and the user's short and long term click behavior, represented by User Click Propensity (UCP), is used to generate a User effective Cost Per Thousand (UeCPM) value. Candidate advertisements are filtered based on a minimum threshold value for UeCPMs. Page placement for the remaining advertisements is determined based on a user expected revenue for an advertisement determined from the UCP. Advertisements having a user expected revenue above another threshold are placed in a north page location.

Abstract: A method for executing multiple, cascading ad auctions, including receiving a request to deliver ads in response to a search query for display on a search results page having a plurality of ad sections; executing at least first and second sequential auctions for the plurality of ad sections by: ranking the ads in the first auction and identifying which of the ads placeable in the North section to deliver according to a first utility function that considers a relevance of the search results page with the ads placed; ranking a remainder of the ads not identified for delivery to the North section in the second auction and identifying which of the remainder of the ranked ads to deliver to a second section based on a second utility function; and delivering the ads identified by the first utility function for display in the North section and the ads identified by the second utility function for display in the second section.

Abstract: A method for reducing ad impact on users in a search results page includes receiving a request to deliver ads in response to a search query for display on a search results page; receiving relevance scores for a plurality of ranked web results that are to be served to the search results page; ranking a plurality of ads identified as relevant to the search query according to a position-normalized, click-through-rate metric and bid values, wherein a predetermined number of the top-ranked ads are placeable in a plurality of North ad slots; incrementally and additively placing the placeable ads sequentially according to rank (k) in their respective North ad slots until a utility score generated by a utility function for a current iteration of ads fails to exceed a threshold value; and delivering to the search results page the ads placed in the North ad slots.

Abstract: A method of speech recognition that uses hierarchical data structures that include a top level grammar and various related subgrammars, such as word, phone, and state subgrammars. A speech signal is acquired, and a probabilistic search is performed using the speech signal as an input, and using the (unexpanded) grammars and subgrammars as possible inputs. Memory is allocated to a subgrammar when a transition to that subgrammar is made during the probabilistic search. The subgrammar may then be expanded and evaluated, and the probability of a match between the speech signal and an element of the subgrammar for which memory has been allocated may be computed. Because unexpanded grammars and subgrammars take up very little memory, this method enables systems to recognize and process a larger vocabulary that would otherwise be possible.

Abstract: Described herein is a system that enables service provider's to integrate speech functionality into their applications. A service provider maintains a set of application servers. To provide a particular speech service to a client of the application server, the application server causes the client to request the speech service from another set of servers. This set of servers is responsible for providing this speech service as well as others. Such speech services include recording digital speech data at the client, and storing the recordings. Later, the application servers may retrieve the recordings, and even more, retrieve data derived from the recordings, such as data generated through speech recognition processes.

Abstract: A method of speech recognition that uses hierarchical data structures that include a top level grammar and various related subgrammars, such as word, phone, and state subgrammars. A speech signal is acquired, and a probabilistic search is performed using the speech signal as an input, and using the (unexpanded) grammars and subgrammars as possible inputs. Memory is allocated to a subgrammar when a transition to that subgrammar is made during the probabilistic search. The subgrammar may then be expanded and evaluated, and the probability of a match between the speech signal and an element of the subgrammar for which memory has been allocated may be computed. Because unexpanded grammars and subgrammars take up very little memory, this method enables systems to recognize and process a larger vocabulary that would otherwise be possible.

Abstract: A method for language fluency training on a computer system having an audio output device includes invoking a web browser program, receiving a pre-recorded file including a message in a spoken language from a conversation partner, playing the message to a user seeking fluency training in the spoken language from within the web browser program on the audio output device, asynchronously with playing the message, recording a user file including a message in the spoken language from the user in response to the message from within the web browser program, outputting the user file to the conversation partner and to a language instructor, receiving an instruction file including an instruction message in the spoken language from the language instructor in response to the user message and playing the instruction message to the user from within the web browser program on the audio output device.

Abstract: A method and apparatus are described for asynchronously conducting interviews though a user interface executing on a client. The user interface prompts an interviewee for at least one audio response, which is digitally recorded. User interfaces are generated by a client according to code defining the user interfaces downloaded from a server via, for example, the Internet. The server may be remote from the client, thereby allowing interviewees to interact with the user interfaces on their own computers. The user interface queries an interviewee and the interviewee responds, either by entering text or digitally recording a response using controls supplied by the user interface. The responses are down loaded via, for example, the Internet to a server. Evaluators may review an interviewee's response through the use of user interfaces. A user (e.g.

Abstract: A method for language fluency training on a computer system having an audio output device includes invoking a web browser program, receiving a pre-recorded file including a message in a spoken language from a conversation partner, playing the message to a user seeking fluency training in the spoken language from within the web browser program on the audio output device, asynchronously with playing the message, recording a user file including a message in the spoken language from the user in response to the message from within the web browser program, outputting the user file to the conversation partner and to a language instructor, receiving an instruction file including an instruction message in the spoken language from the language instructor in response to the user message and playing the instruction message to the user from within the web browser program on the audio output device.

Abstract: An automatic recognition system and method divides observation vectors into subvectors and determines a quantization index for the subvectors. Subvector indices can then be transmitted or otherwise stored and used to perform recognition. In a further embodiment, recognition probabilities are determined for subvectors separately and these probabilities are combined to generate probabilities for the observed vectors. An automatic system for assigning bits to subvector indices can be used to improve recognition.

Abstract: Pronunciation quality is automatically evaluated for an utterance of speech based on one or more pronunciation scores. One type of pronunciation score is based on duration of acoustic units. Examples of acoustic units include phones and syllables. Another type of pronunciation score is based on a posterior probability that a piece of input speech corresponds to a certain model such as an HMM, given the piece of input speech. Speech may be segmented into phones and syllables for evaluation with respect to the models. The utterance of speech may be an arbitrary utterance made up of a sequence of words which had not been encountered before. Pronunciation scores are converted into grades as would be assigned by human graders. Pronunciation quality may be evaluated in a client-server language instruction environment.

Abstract: Pronunciation quality is automatically evaluated for an utterance of speech based on one or more pronunciation scores. One type of pronunciation score is based on duration of acoustic units. Examples of acoustic units include phones and syllables. Another type of pronunciation score is based on a posterior probability that a piece of input speech corresponds to a certain model, such as a hidden Markov model, given the piece of input speech. Speech may be segmented into phones and syllable for evaluation with respect to the models. The utterance of speech may be an arbitrary utterance made up of a sequence of words which had not been encountered before. Pronunciation scores are converted into grades as would be assigned by human graders. Pronunciation quality may be evaluated in a client-server language instruction environment.

Abstract: A method and apparatus for automatic recognition of speech adapts to a particular speaker by using adaptation data to develop a transformation through which speaker independent models are transformed into speaker adapted models. The speaker adapted models are then used for speaker recognition and achieve better recognition accuracy than non-adapted models. In a further embodiment, the transformation-based adaptation technique is combined with a known Bayesian adaptation technique.