Related Experiment Videos
State of the art in continuous speech recognition
1BBN Systems and Technologies, Cambridge, MA 02138, USA.
Summary
Recent advances in automatic speech recognition (ASR) have significantly reduced word errors and increased speed. Improved mathematical speech modeling now enables real-time, accurate, speaker-independent ASR on standard computers.
Area of Science:
- Computer Science
- Artificial Intelligence
- Signal Processing
Background:
- Significant progress in machine-based automatic speech recognition (ASR) over the last decade.
- Increased computational power and improved algorithms have driven ASR advancements.
Purpose of the Study:
- To highlight key advancements in ASR technology.
- To focus on the impact of enhanced speech modeling techniques on ASR performance.
Main Methods:
- Development of faster recognition search algorithms.
- Utilization of more powerful computing hardware.
- Focus on improved mathematical modeling of speech sounds.
Main Results:
- Word error rate reduced by over a factor of 5.
- Recognition speeds increased by several orders of magnitude.
- Achieved high-accuracy, speaker-independent, continuous ASR in real-time on standard workstations.
Conclusions:
- ASR technology is now accessible to the general public.
- Advances in speech modeling are crucial for high-performance ASR.
- Future ASR development relies on sophisticated mathematical speech sound modeling.