Qualcomm incorporated (20240282327). SPEECH ENHANCEMENT USING PREDICTED NOISE simplified abstract
Contents
SPEECH ENHANCEMENT USING PREDICTED NOISE
Organization Name
Inventor(s)
Asif Mohammad of San Diego CA (US)
Fatemeh Alishahi of Los Angeles CA (US)
Maliha Arif Zain of Orlando FL (US)
Sajeer Fazil of San Diego CA (US)
SPEECH ENHANCEMENT USING PREDICTED NOISE - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240282327 titled 'SPEECH ENHANCEMENT USING PREDICTED NOISE
Simplified Explanation:
This patent application describes a device that can remove noise from an input audio signal, specifically focusing on speech signals.
- The device includes processors that analyze the input audio signal to predict noise based on a trained model.
- The predicted noise signal is then subtracted from the input audio signal to produce a cleaner output audio signal.
Key Features and Innovation:
- Utilizes trained models to predict noise in audio signals.
- Removes predicted noise from input audio signals to enhance speech clarity.
Potential Applications:
- Speech enhancement in audio recordings.
- Noise reduction in communication devices.
- Improving speech recognition systems.
Problems Solved:
- Enhances speech intelligibility in noisy environments.
- Improves the quality of audio recordings in challenging acoustic conditions.
Benefits:
- Clearer speech communication.
- Enhanced audio quality in various applications.
- Improved user experience in noisy environments.
Commercial Applications:
- Audio recording devices.
- Communication systems.
- Speech recognition technology.
Questions about Noise Reduction Technology:
1. How does the device predict noise in the input audio signal?
- The device uses trained models to analyze the input audio signal and generate a predicted noise signal.
2. What are the potential benefits of removing noise from audio signals?
- Removing noise can improve speech clarity and overall audio quality in various applications.
Original Abstract Submitted
a device includes one or more processors configured to obtain an input audio signal including at least first speech of a first person. the one or more processors are configured to generate a predicted noise signal based on processing of the input audio signal by a trained model. the one or more processors are configured to subtract the predicted noise signal from the input audio signal to generate an output audio signal.
