Speech devices overview
The Speech service works with a wide variety of devices and audio sources. You can use the default audio processing available on a device. Otherwise, the Speech SDK has an option for you to use our advanced audio processing algorithms that are designed to work well with the Speech service. It provides accurate far-field speech recognition via noise suppression, echo cancellation, beamforming, and dereverberation.
Audio processing is enhancements applied to a stream of audio to improve the audio quality. Examples of common enhancements include automatic gain control (AGC), noise suppression, and acoustic echo cancellation (AEC). The Speech SDK integrates Microsoft Audio Stack (MAS), allowing any application or product to use its audio processing capabilities on input audio.
Microphone array recommendations
The Speech SDK works best with a microphone array that has been designed according to our recommended guidelines. For details, see Microphone array recommendations.
Device development kits
The Speech SDK is designed to work with purpose-built development kits, and varying microphone array configurations. For example, you can use one of these Azure development kits.
- Azure Percept DK contains a preconfigured audio processor and a four-microphone linear array. You can use voice commands, keyword spotting, and far field speech with the help of Azure Cognitive Services.
- Azure Kinect DK is a spatial computing developer kit with advanced AI sensors that provide sophisticated computer vision and speech models. As an all-in-one small device with multiple modes, it contains a depth sensor, spatial microphone array with a video camera, and orientation sensor.