Method and apparatus for performing conferencing services and echo suppression
View Patent ↗The present invention provides a method and apparatus for providing conferencing services in a converged services platform operable to perform a variety of enhanced telecommunications services. The advanced conferencing algorithm embodying the invention incorporates techniques in each of four signal processing categories including noise gating, echo suppression, automatic gain control and DTMF clamping are employed to produce a high perceptual quality conference output for a conference of several hundred participants, and which is computationally cost effective.
1. A method of performing an audio conference of multiple attendees, including the steps of
(A) receiving input samples of PCM information from each attendee in the conference;
(B) calculating the energy of each sample for each attendee of the conference;
(C) selecting a predetermined number of samples that exhibit the highest energy;
(D) generating a histogram to determine one or more noise peaks;
(E) selecting one of said noise peaks as a noise floor;
(F) eliminating from the conference incoming samples that fall below said noise floor; and
(G) selecting for inclusion in said conference a predetermined number of the remaining samples whose energies are the largest, and summing a predetermined number of the remaining samples whose energies are the largest to produce an output signal.
2. The method as defined in claim 1 comprising the further step of calculating the energy of the sample using the sum of the squares of the signal over a predetermined time period.
3. The method as defined in claim 1 wherein generating said histogram includes the further step of plotting a total energy level for a predetermined time block to determine said noise peaks.
4. The method as defined in claim 3 comprising the further steps of:
evaluating peaks including ignoring lowest peaks that are close to zero; and
evaluating secondary peaks to find the most occurring energy representation and setting this as said noise floor.
5. The method as defined in claim 4 comprising the further step of applying a logical relationship to said histogram when there are multiple peaks to determine the best noise floor including using a relationship between a ratio of the average histogram value to that of a second peak and determining if said ratio exceeds a predetermined amount, then the sample is determined to be speech and is above the predetermined noise floor.
6. A method of performing an audio conference of multiple attendees, including the steps of
(A) receiving the input samples of PCM information from each attendee in the conference;
(B) calculating the energy of each sample for each attendee of the conference;
(C) suppressing any echo out of said conference output signal by aggregating the energies calculated from each sample during the predetermined time period; and
(1) populating a matrix with the energy aggregates that are calculated;
(2) solving normal equations for said matrix to produce results;
(3) evaluating said results;
(3) applying a moving average to said results across the time dimension;
(4) selecting peaks and a predetermined time lag that represent echo in said energies;
(5) evaluating incoming speech samples and finding those having an energy value is less than the predetermined peak energy at that predetermined time lag;
(6) masking any such samples whose energies are less than or equal to the predetermined peak, thus suppressing the echo in said sample;
(D) selecting a predetermined number of samples that exhibit the highest energy;
(E) summing these energies to produce a conference output signal; and
(F) for each attendee of the conference, subtracting any speech attributable to that attendee from his own returned conference output signal.
7. The method as defined in claim 6 further comprising in said suppressing step:
generating a synthetic echo envelope using calculated energies of samples; and
using said synthetic echo envelope to determine a delay and gain for an echo signal to be used in echo suppression.
8. A method of performing an audio conference of multiple attendees, including the steps of:
(A) receiving input samples of PCM information from each attendee in a conference;
(B) for each attendee of the conference, detecting dual tone multi-frequency tones employing a detector to identify a row tone and a column tone;
(C) adjusting a filter to apply a coefficient that multiplies the signal by a predetermined amount in order to remove each row tone and each column tone to thereby remove the DTMF from the signal; and
(D) sending the signal with the DTMF having been removed to a conferencing process to be summed with other attendee signals.