VoiceBoost™ Real Time Speech Discrimination in Noise

Real time audio signal processing for ultra-low power processors increases speech clarity in noise

Although mobile devices are increasingly used in noisy environments and users have come to expect high voice call quality in all environments, mobile devices have been unable to adequately separate the user's voice from background noise in adverse conditions. This continues to result in users' dissatisfaction with poor quality voice communication in common noisy environments.

Poor voice clarity in noise is due to the fact that present noise reduction approaches used in mobile devices cannot filter out background noise from the speech of interest when the background noise has the same complex frequencies as the foreground speech, such as the babble in a busy restaurant, street, or a lively restaurant. To achieve better voice clarity, a solution is needed to isolate the speech of interest from all background noise, including overlapping noise with the same frequencies as speech.

Achieve Voice Clarity in Noise

The VoiceBoost Real Time Speech Discrimination solution detects and isolates speech from complex background noise within audio signals using a patent-pending mathematical model of human speech. The VoiceBoost Speech Discrimination solution easily integrates into existing audio processing pipelines for ultra-low power processors used within mobile phones, Bluetooth® headsets and portable two-way radios.

Unlike competing speech processing implementations, Malaspina Labs’ VoiceBoost Speech Discrimination solution does not require two or more microphones, large processing power or lengthy signal latencies to separate speech of interest from background noise, including speech-shaped noise and speech babble. The solution supports mixed audio input signals from single omni-directional or directional microphones, as well as spatially separated input signals from multiple microphones (beamforming).

Hear the Difference

Improvements in speech quality greater than 36% have been shown when VoiceBoost Real Time Speech Discrimination is compared to mobile phones and 3G network operators' existing noise reduction algorithms and narrowband codecs.

Speech Discrimination Processing in noise with near-field microphone

Original speech in +5dB SNR omni-drientional babble noise

VoiceBoost Speech Discrimination enhanced with one microphone near-field: 9.3 MIPS, 12ms latency

Speech Discrimination Processing in extreme noise with microphone at arm's length

Original speech in 0dB SNR omni-drientional babble noise

VoiceBoost Speech Discrimination enhanced with two microphones at arm's length: 21 MIPS, 12ms latency