Sound Software Employs Blind Source Separation

Sound Software Employs Blind Source Separation

Conexant Systems' is well known for their audio hardware (see “HD Audio Codec Boasts Integrated Class-D Amplifier”). Of course, many of their chips also incorporate software to improve audio quality. Dedicated chips can simplify a designer's chores but often a larger, host processor is available to assist with audio requirements as in smart phones or tablets. Some of Conexant Systems' latest software offerings provide significant improvements in audio quality allow for improved speech recognition.

One application area is voice command processing for home control applications (Fig. 1). Here applications need to address far-field voice processing to reduce background noise thereby allowing improved speech recognition. Speech recognition with lots of noise in the audio source is more difficult and less accurate than operating on a source with very little background noise. The trick is to deliver the best quality voice stream to the recognition software. Conexant Systems' Linux-based far-field voice processing software solution address this and other applications.

Figure 1. Home control applications using voice recognition have a range of noise sources as well as command sources.

The Linux support employs blind source separation (BSS) by using spatial representation of target speech and noise sources and then reducing interference. It can handle voice sources up to five meters away from the pick up microphones.

BSS has advantages over other technologies such as beamforming. The challenge for beamforming is that speech and noise can arrive at any angle relative to the microphones. Beamforming solutions cannot optimized for all angles of the noise source. This is especially true when the noise comes from the same direction as the headset user. BSS does not care where the noise is coming from.

BSS employs a pair of microphones to detect and suppress stationary noise sources. This includes fixed sources such as white noise or noise from air conditioners as well as non-stationary noise like a slamming door. One microphone, the primary voice transmit microphone, is placed near the user’s mouth while the other is farther away detecting environmental sounds.

Conexant is also delivering its new Smart Voice Pickup (SVP) software is part of its AudioSmart solutions. It targets the input from a headset microphone for applications like collaborative, online gaming and unified communications (UC) found in enterprise call centers.

SVP is based on Conexant Systems' SSP noise suppression algorithm. SSP also uses BSS to achieve up to 45 dB noise suppression and deliver up to 19 dB signal-to-noise ratio improvement. The technology is very good at at reducing noise from keystrokes during conference calls.

The last bit of news from Conexant is related to hardware. Their CX2072x I2S audio codec with integrated AudioSmart amplifier targets tablets. This solution addresses speaker output as well as voice processing. It uses AudioSmart voice and speech processing software to provide noise cancellation and full duplex acoustic echo cancellation (AEC). This is handy for Skype call and other voice oriented applications and it utilizes BSS technology.

On the output side, the AudioSmart amplifier support can improve loudness by a factor of two even using the small speakers found in tablets. The chip is compatible with tablet platforms using ARM and Intel-based SOCs, Windows and Android OS.

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.