Abstract: Systems and methods for recording multiple channels of incoming data. Shared data may be centrally recorded, whereas unique data may be recorded at an endpoint, such as a workstation. A central recording system may request the unique data recorded at the endpoint and correlate it to the shared data. In some implementations, the data may be audio, where the shared data is audio played over speakers, whereas the unique data is audio at a handset of a workstation.

Abstract: According to one embodiment, a prosody editing apparatus includes a storage, a first selection unit, a search unit, a normalization unit, a mapping unit, a display, a second selection unit, a restoring unit and a replacing unit. The search unit searches the storage for one or more second prosodic patterns corresponding to attribute information that matches attribute information of the selected phrase. The mapping maps each of the normalized second prosodic patterns on a low-dimensional space. The restoring unit restores a restored prosodic pattern according to the selected coordinates. The replacing unit replaces prosody of synthetic speech generated based on the selected phrase by the restored prosodic pattern.

Abstract: A corpus of textual listings is received and main concept words and attribute words therein are identified via an iterative process of parsing listings and expanding a semantic model. During the parsing phase, the corpus of textual listings is parsed to tag one or more head noun words and/or one or more identifier words in each listing based on previously identified main concept words or using a head noun identification rule. Once substantially each listing in the corpus has been parsed in this manner, the expansion phase assigns head noun words as main concept words and modifier words as attribute words, where possible. During the next iteration, the newly identified main concept words and/or attribute words are used to further parse the listings. These iterations are repeated until a termination condition is reached. Remaining words in the corpus are clustered based on the main concept words and attribute words.

Abstract: The embodiments of the present invention improves conventional attenuation schemes by replacing constant attenuation with an adaptive attenuation scheme that allows more aggressive attenuation, without introducing audible change of signal frequency characteristics.

Abstract: Topics are determined for short text messages using an unsupervised topic model. In a training corpus created from a number of short text messages, a vocabulary of words is identified, and for each word a distributed vector representation is obtained by processing windows of the corpus having a fixed length. The corpus is modeled as a Gaussian mixture model in which Gaussian components represent topics. To determine a topic of a sample short text message, a posterior distribution over the corpus topics is obtained using the Gaussian mixture model.

Abstract: Devices, systems and methods are disclosed for reducing noise in input data by performing a hysteresis operation followed by a lateral excitation smoothing operation. For example, an audio signal may be represented as a sequence of feature vectors. A row of the sequence of feature vectors may, for example, be associated with the same harmonic of the audio signal at different points in time. To determine portions of the row that correspond to the harmonic being present, the system may compare an amplitude to a low threshold and a high threshold and select a series of data points that are above the low threshold and include at least one data point above the high threshold. The system may iteratively perform a spreading technique, spreading a center value of a center data point in a kernel to neighboring data points in the kernel, to further reduce noise.

Abstract: Approaches presented herein enable assignment of translated work to an agent in a customer support environment based on a confidence factor that measures accuracy of translation and an agent's language skill. Specifically, agent proficiencies in a set of natural languages are measured and scored. An incoming customer communication is translated into one or more natural languages and each language translation is assigned a translation score based on a confidence of translation. The skill score and translation score are utilized to calculate a confidence factor for each language. In one approach, the customer communication is assigned to an agent that has a confidence factor greater than a predetermined threshold confidence factor. In another approach, the communication is only assigned if a rule optimizing agent availability and risk of constrained resources is satisfied.

Abstract: The present disclosure describes a teleconferencing system that may use a virtual participant processor to translate language content of the teleconference into each participant's spoken language without additional user inputs. The virtual participant processor may connect to the teleconference as do the other participants. The virtual participant processor may intercept all text or audio data that was previously exchanged between the participants may now be intercepted by the virtual participant processor. Upon obtaining a partial or complete language recognition result or making a language preference determination, the virtual participant processor may call a translation engine appropriate for each of the participants. The virtual participant processor may send the resulting translation to a teleconference management processor. The teleconference management processor may deliver the respective translated text or audio data to the appropriate participant.

Abstract: A method includes separating, at a device, an input audio signal into at least a low-band signal and a high-band signal. The low-band signal corresponds to a low-band frequency range and the high-band signal corresponds to a high-band frequency range. The method also includes selecting a non-linear processing function of a plurality of non-linear processing functions. The method further includes generating a first extended signal based on the low-band signal and the non-linear processing function. The method also includes generating at least one adjustment parameter based on the first extended signal, the high-band signal, or both.

Abstract: An information processing system for input prediction that displays written character string candidates of words or phrases corresponding to an input character string in a terminal device. The terminal device acquires internal candidates by performing a search based on the input character string and transmits a candidate transmission request signal including identification information of a terminal dictionary stored in the terminal device and the input character string to a server. The server performs a search through the terminal dictionary corresponding to the identification information in the received signal and through a server dictionary dedicated to the server. From dictionary data extracted from the server dictionary, the server returns dictionary data of external candidates different from dictionary data extracted from the terminal dictionary as external candidates to the terminal device.

Abstract: A language setting system, for use by a driver having a preferred language and a smartphone, when entering a vehicle to set an active language of the vehicle for vehicle communications with the driver. The vehicle connects with and scans the smartphone. When the smartphone has a user selected language setting, the user selected language setting is used to set the active language of the vehicle. When scan of the smartphone detects examples of language usage in text and email communications, and such examples meet a threshold usage, the active language may be set to the language of the usage examples. When examples of multiple languages are detected on the smartphone, the driver may be prompted to select the preferred language from a list of the detected examples.

Abstract: The present document provides a scene recognition method and device based on ambient sound and a mobile terminal. The device includes: a sound collection module, a preprocessing module, a feature extraction module, a scene recognition module and a database. The method includes: collecting a sound signal; processing the sound signal into a frequency domain signal; extracting sound feature information from the frequency domain signal; inputting the sound feature information under a preset model, matching a model output result with weight values of sound sample models of scenes, and determining a scene corresponding to the sound feature information. The present document implements locating based on background sound information as the feature of the scene, so that the mobile terminal quickly and correctly recognizes the current scene under the maintaining low-consumption state.

Abstract: A dialog system that includes a dialog manager to manage a conversation between the dialog system and a user, and to associate the conversation with a complex activity, and a plan engine to execute a plan script in connection with the complex activity, the plan script including a set of atomic dialog activities and logic to control a data and sequence flow of the atomic dialog activities, the set of atomic dialog activities being sub-activities of the complex activity, the complex activity being specified via a declarative activity specification language that connects the atomic dialog activities with a process.

Abstract: Provided are, among other things, systems, methods and techniques for detecting whether a transient exists within an audio signal. According to one representative embodiment, a segment of a digital audio signal is divided into blocks, and a norm value is calculated for each of a number of the blocks, resulting in a set of norm values for such blocks, each such norm value representing a measure of signal strength within a corresponding block. A maximum norm value is then identified across such blocks, and a test criterion is applied to the norm values. If the test criterion is not satisfied, a first signal indicating that the segment does not include any transient is output, and if the test criterion is satisfied, a second signal indicating that the segment includes a transient is output. According to this embodiment, the test criterion involves a comparison of the maximum norm value to a different second maximum norm value, subject to a specified constraint, within the segment.

Abstract: A speech recognition and control system including a receiver for receiving an audio input, an event detector for analyzing the audio input and identifying at least one event of the audio input, a recognizer for interpreting at least a portion of the audio input, a database including a plurality of rules, and a controller for generating a control command based on the at least one event and at least one rule.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing speaker identification. In some implementations, an utterance vector that is derived from an utterance is obtained. Hash values are determined for the utterance vector according to multiple different hash functions. A set of speaker vectors from a plurality of hash tables is determined using the hash values, where each speaker vector was derived from one or more utterances of a respective speaker. The speaker vectors in the set are compared with the utterance vector. A speaker vector is selected based on comparing the speaker vectors in the set with the utterance vector.

Abstract: Disclosed herein are systems, computer-implemented methods, and tangible computer-readable storage media for captioning a media presentation. The method includes receiving automatic speech recognition (ASR) output from a media presentation and a transcription of the media presentation. The method includes selecting via a processor a pair of anchor words in the media presentation based on the ASR output and transcription and generating captions by aligning the transcription with the ASR output between the selected pair of anchor words. The transcription can be human-generated. Selecting pairs of anchor words can be based on a similarity threshold between the ASR output and the transcription. In one variation, commonly used words on a stop list are ineligible as anchor words. The method includes outputting the media presentation with the generated captions. The presentation can be a recording of a live event.

Abstract: A device and a method to improve quality of a signal in a lossy communication system are disclosed. One or more samples of the signal are received from a first and second microphone transducers. The received samples are processed and filtered to obtain a processed signal. A voice activity detector is provided for iteratively identifying speech regions and non-speech regions of the signal. All samples received by the microphones are continuously monitored and quality of each sample is improved by reducing or eliminating the noise detected in the non-speech regions of the processed signal.

Abstract: The disclosed embodiments illustrate methods and systems for processing multimedia content. The method includes extracting one or more words from an audio stream associated with multimedia content. Each word has associated one or more timestamps indicative of temporal occurrences of said word in said multimedia content. The method further includes creating a word cloud of said one or more words in said multimedia content based on a measure of emphasis laid on each word in said multimedia content and said one or more timestamps associated with said one or more words. The method further includes presenting one or more multimedia snippets, of said multimedia content, associated with a word selected by a user from said word cloud. Each of said one or more multimedia snippets corresponds to said one or more timestamps associated with occurrences of said word in said multimedia content.

Abstract: A voice retrieval device includes a processor; and a memory which stores a plurality of instructions, which when executed by the processor, cause the processor to execute: setting detection criteria for a retrieval word, based on a characteristic of the retrieval word, such that the higher the detection accuracy of the retrieval word or the lower the pronunciation difficulty of the retrieval word or the lower the appearance probability of the retrieval word, the stricter the detection criteria; performing first voice retrieval processing on voice data according to the detection criteria and detecting a section that possibly includes the retrieval word as a candidate section from the voice data; and performing second voice retrieval processing different from the first voice retrieval processing on each candidate section and determining whether or not the retrieval word is included in each candidate section.