|
Although industrial standards (such as speech and audio compression standards) are developed with 16-bit embedded devices in mind, those standards are now extensively used in desktop applications such as Internet telephony, desktop conferencing, speech recognition, etc. For example, an Internet phone client application must include:
- Voice compression algorithm (such as the G.723.1 standard) to reduce the data sent to the network and to decode the incoming bit steam
- Noise Reduction Algorithm (NRA) to reduce the amount of noise (computer fan, noise from cheap sound card / loudspeaker / microphone, back-ground noise) sent to the communication channel
- Acoustic Echo Canceller (AEC) if the computer (or notebook) speakers and microphone to be used in the communication to avoid that the person at the other side of the line would hear a delayed version of his own voice.
- Voice Activity Detector (VAD) to detect when the user is not speaking, and taking an appropriate action
- Comfort Noise Generator (CNG) to simulate the noise we are used to hear during silent intervals when talking through the telephone
Similarly, a speech recognition software package would necessarily include NRA and probably AEC as well.
DSP ALGORITHMS provides PC implementations of all the above mentioned algorithms for the developers of PC applications. All algorithms are directly available in the form of dynamic link libraries for several operating systems including all versions of Windows and Linux. The libraries automatically identify the processor features and uses the most efficient code. For instance the SIMD instructions (available in Intel Pentium III processors and higher) are automatically used if the processor running the code supports those instructions. Besides the libraries, API documentation, integration code examples, and a demo application are provided.
Read more about
|
