Articles tagged with Speech Recognition
A new chip designed by MIT researchers has the potential to make voice control ubiquitous in electronics, offering significant power savings. The chip's ability to minimize memory bandwidth and compress weights associated with each node enables efficient speech recognition, making it practical for relatively simple electronic devices.
Researchers at MIT's CSAIL have developed a new system that analyzes correspondences between images and spoken descriptions to train speech-recognition systems. The system can potentially provide automatic speech recognition for less-resourced languages, leading to fully automated translation capabilities.
Disney researchers developed a speech technology system that can sort through overlapping speech, social side talk, and creative pronunciations of young children to make it work. The system was 85% accurate in recognizing keywords, outperforming commercial speech recognition systems.
Researchers at Cornell University have developed a new method called 'data smashing' that enables automated discovery without human intervention, opening doors to complex observations and expert-driven analysis.
Researchers have developed improved artificial voice devices that can produce age-appropriate voices for children and women with vocal cord problems. The new technology enhances speech signal analysis by tackling the challenge of identifying glottal excitation signals in higher-pitched voices.
Scientists from UCSF used brain recordings to decode which speaker a subject was listening to, even when multiple voices were present. The study sheds light on the human brain's ability to focus on one voice in noisy environments, with implications for language learning disorders and neuroprosthetic devices.
A new study in Canada found that breast imaging reports generated with automatic speech recognition (ASR) systems have a higher error rate compared to conventional dictation transcription. At least one major error was found in 23% of ASR reports, while 4% of conventional reports contained errors affecting patient management.
Researchers developed a hybrid approach combining bionic wavelet transform and recurrent neural network for improved speech enhancement. The method showed significant noise reduction without compromising intelligibility, achieving a 12 dB increase in signal-to-noise ratio.
Researchers are working to improve automatic speech recognition technology, aiming for a word-error rate of no more than 10 percent. A new multi-language audio database will be developed using publicly available sources like YouTube, featuring hundreds of speakers in English, Spanish, and Mandarin Chinese.
A recent study found significantly higher transcription error rates in women compared to men, with error rates ranging from 0.015 to 0.206 in females and 0.025 to 0.139 in males. The results suggest that women may need to spend more time training on commercial speech recognition systems to achieve accurate results.
Researchers at the University of East Anglia are developing computerized lip-reading systems that can automatically convert video lip-motions into text. The three-year project aims to improve the accuracy of lip-reading, a skill that can be unreliable even for trained experts.
Researchers from Johns Hopkins University are developing mathematical models to represent the safest and most effective ways to perform surgery. By analyzing data from robotic medical tools, they aim to evaluate a surgeon's work and help doctors improve their skills. The goal is also to enable robotic surgical tools to perform with gre...
The next generation of speech recognizers will go beyond conversion to text, using information detectors that provide a machine with digestible information. Researchers are working on new mathematical algorithms and paradigms to tackle the broader problem of human-to-machine and machine-to-human speech.
UCSD researchers have developed a new estimator that outperforms the historic Good-Turing formula under all conditions. The new estimator, called attenuation, evaluates the highest possible ratio between the probability assigned to each symbol in a sequence by any distribution and the corresponding probability assigned by the estimator.
The Journal of Rehabilitation Research and Development covers a range of rehabilitation research disciplines. Researchers have developed techniques for early detection of medication-induced hearing loss and created 3D models of bones using CT scans.
A Johns Hopkins University team is developing a speech recognition system to help historians sift through Holocaust survivor interviews in languages other than English. The system aims to improve access to the archive's vast collection of video interviews.
Researchers at USC have created a machine system that recognizes spoken words better than humans, with potential benefits for Navy sonar and improving interaction between man and computer. The system may eventually enable voice control of computers, help the deaf, and aid air traffic controllers in noisy environments.
Researchers at Johns Hopkins University are developing new tools to improve speech recognition accuracy. The team aims to enable computers to understand any kind of human speech and provide a powerful way to search through hours of recorded speeches and news reports.
A new computer program developed at the University of Rochester enables users to have natural conversations with computers. The program, Phenelope DuJour, uses intention recognition to understand user intent and respond accordingly, improving the efficiency of human-computer interactions.