Noise Reduction Gen 3: Third Generation Voice Quality Enhancement Algorithm

Telephony by Adaptive Digital

ADT NOISE REDUCTION G3

Noise Cancellation: Includes single frequency and siren suppression

DESCRIPTION

The objective of noise reduction algorithms is to improve one or more perceptual aspects of noisy speech, most notably, quality and clarity. If it were always the same, background noise would be relatively easy to eliminate. Since background noise can originate from varying sources at varying degrees of intensity, noise reduction is a challenge. The sound generated from a fan or HVAC unit, the rumble of auto and air traffic, and blaring radios all have different characteristics. A good Noise Reduction algorithm needs to be able to reduce a wide variety of noises without degradation in speech quality.

Adaptive Digital’s third generation Noise Reduction algorithm analyses the speech signal for three distinct types of degradation noise, tone, and siren.

In a situation where white noise is present in conjunction with speech, the noise reduction algorithm can achieve up to 12 dB of signal to noise ratio improvement without significant degradation to the desired speech signal.

In a situation where a single frequency tone is present in conjunction with speech, the noise reduction algorithm can reduce the undesired tone 30 dB or more without significant degradation to the desired speech signal.

In a situation where a siren is present in conjunction with speech, the noise reduction algorithm can reduce the undesired siren by up to 20 dB or more without significant degradation to the desired speech.

Data sheet: .pdf

FEATURES

  1. Functions are C-callable
  2. Multiple channel operation
  3. Parameters are user configurable
  4. Reduces background noise whether or not speech is present thereby increasing the signal-to-noise ratio.
  5. Reduces amplitude of tonal signals whether or not speech is present.
  6. Reduces amplitude of most siren signals with moving tonal content.

AVAILABILITY

ADT Noise Reduction is available on the TMS320™ DSP Family

C55x™DSP, & C64x™DSP, C64x+™DSP, C674x™DSP Generations

SPECIFICATIONS

Noise Reduction Gen 3

CPU Utilization

All Memory usage is given in units of byte.

 - C64x C64x+ C674x C55x
Sampling Rate (Hz) Peak MIPS Average MIPS Peak MIPS Average MIPS Peak MIPS Average MIPS Peak MIPS Average MIPS
8000 30 15 30 15 30 15 90 45

 

Please Note: MIPS

*C6000 MIPS were characterized on with all program and data residing in external RAM, but with cache enabled.

The peak MIPS are double compared to the average MIPS due to the fixed internal processing frame size, which is 64 samples. If the users frame size is not equal to an integral multiple of 64 samples, the number of internal frames processed will not be consistent. For example, if the user’s frame size is 96, the algorithm will run one internal frame the first time it is called, and save 32 samples in a FIFO since there aren’t enough samples yet to run an internal frame. The second time the algorithm is called with 96 more samples, there will be 128 samples to process. The algorithm will therefore run two internal frames, consuming twice the MIPS as it did during the first frame – hence a doubling of MIPS.

If the user can operate at a frame size of 64 or an integral multiple of 64, the peak MIPS will be equal to the average MIPS listed in the table.

Memory Utilization
All Memory usage is given in units of byte.

Processor Program Data Constants Per Channel Common Scratch
C64x 22272 389 355 11228 2068 1544
C64x+ 20416 389 382 11228 2068 1544
C674x 22336 389 355 11228 2068 1544
C55x 8095 0 446 11228 2068 1544

FUNCTION

NCAN_ADT_config( . . .) Configures Noise Canceller

NCAN_ADT_init( . . .) Initiates a channel of Noise Cancellation

NCAN_ADT_reduce( . . . ) Executes Noise Cancellation