CONTINUOUS SPEECH PROCESSING
Continuous Speech Processing (CSP) is a feature available with select boards supporting high-performance, speech enabled applications.
This technology enhances existing speech technologies by providing board-level firmware that processes real time voice signals to identify human speech input and present it to the host platform for speech recognition.
The real-time functions include both echo cancellation and Voice Activity Detection (VAD).
FEATURES AND BENEFITS
CSP has the following features and offers these benefits:
Provides low implementation cost and enhanced System performance
Reduces system latency, increases recognition accuracy, and improves overall system response time
CSP can be employed in applications such as:
Voice portals which includes: weather, traffic, movies, restaurant guides, etc.
Speech enabled ...view middle of the document...
Speech recognition provides host-based recognition engines and a variety of tools for developing and implementing robust applications.
There are two different approaches:
1. Traditional Speech Processing
2. Competitive Speech Processing Feature
TRADITIONAL SPEECH PROCESSING
COMPETITIVE SPEECH PROCESSING
HOW CONTINUOUS SPEECH PROCESSING WORKS
The duration of an echo is measured in tens of milliseconds (ms). The number of milliseconds an echo canceller removes is known as the length of the echo canceller.
The echo canceller has two adaptation modes:
1. Fast mode for rapid convergence
2. Slow mode for slower convergence
2. VOICE ACTIVITY DETECTOR:
VAD is the CSP component that examines the incoming signal and determines if the signal contains significant energy to be identified as speech.
VAD detects audio energy and triggers data transmission only when speech is present.
VAD technology in CSP includes a pre-speech buffer that produces better voice recognition using less host processing and enhances the accuracy of speech detection.
CSP supports various application-friendly features, such as the ability to interrupt speech prompts by speaking over them known as “barge-in”.
The combination of echo cancellation and VAD can be used to effect barge-in.
The barge-in feature stops the playing of the prompt upon detection of audio energy exceeding the threshold.
HARDWARE SYSTEM REQUIREMENTS
For boards with Continuous Speech Processing:
• Compact PCI bus computer Single or dual processor PCI or PCI Express bus, or
• to the number of channels being used Operating system hardware requirements vary according
CSP now includes a silence compressed streaming feature, which removes the silence between the caller’s utterances before streaming to the host.
This saves host processing cycles and allows for increased port density. The amount of silence compression and energy thresholds are configurable.