Tone detection using pitch period

Details Tone detection using pitch period Tone detection using pitch period

1999-08-25

2001-03-06

Dorvil, Richemond (Department: 2641)

Data processing: speech signal processing, linguistics, language

Speech signal processing

For storage or transmission

C704S201000, C704S203000

Reexamination Certificate

active

06199036

ABSTRACT:

FIELD OF THE INVENTION
The invention relates to apparatus for detecting a tone, to signal processing apparatus, to echo cancelling apparatus, to speech coding apparatus, to corresponding methods, to methods of using a telecommunications network to make a call, and to corresponding software.
BACKGROUND TO THE INVENTION
Tone detection has a variety of applications in telecommunications systems. Tones can be used for transmitting data, or for signalling purposes for example. They can include pure tones such as fax, modem, dial tone, continuity tone as well as multi-frequency tones such as DTMF (Dual Tone Multi-Frequency), R
1
and R
2
signalling. In this context, the term “tone” is not intended to encompass tones occurring as part of human speech, for example in tonal languages such as Chinese.
Accurate detection of tones may be critical for maintaining low error rates for data transmission, or for proper operation of switches or other network equipment which relies on signalling tones. Particularly where tones are carried on speech circuits, it can be difficult to distinguish the tones because speech may show the same frequencies for short periods. It can be important to recognise tones quickly, particularly in switches containing echo cancellation circuits that need to be switched off when tones are detected. An example is an adaptive echo canceller, which adapts to speech. Tones in the speech band can affect the operation of such echo cancellation circuits since they will try to alter their coefficients to adapt to the tones. This is undesirable since when speech reoccurs, the canceller will take longer to readapt to the speech, and so echoes may be heard.
Conventional DTMF detection methods use bandpass filter banks and envelope detectors to estimate the level of each of the eight possible frequency components. The frequencies with the highest levels are selected as candidates for DTMF signal. Further processing is required to discriminate real DTMF tones from voice signals or other energy in the voice band. An example is shown in proceedings of the IEEE-SP International Symposium 1994 ‘Detection of multi-tone signals based on energy operators’, Edgar F. Velez.
It is also known to use digital signal processors (DSP) to perform evaluation of the discrete Fourier transform (DFT) of the signal using algorithms such as the Goertzel algorithm. Conventional methods may be insufficiently reliable, or take too long, or use too much computational resource.
In many cases, there is a limited amount of computing resource available to carry out tone detection operations, particularly in switches where many signals or channels are handled simultaneously. In such cases, reductions in processing requirements per channel can enable greater channel density, which may be commercially very valuable.
In the book “digital processing of speech signals” by Raminer & Schafer published in 1978, ISBN 0-13-213603-1, there is discussion of pitch period detection at Pages 314-319, 372-379 and 150-158. It is used for estimating fundamental frequency in voice signals for speech recognition, to determine whether the speech is voiced or unvoiced, and to enable the speech to be compared to models. At Page 135 it is indicated that pitch period detectors are used in vocoders, in speaker identification and verification systems, and as aids to the handicapped.
Various ways of determining pitch period are known, including an impulse train algorithm shown at Page 136, which is very computationally intensive, a Fourier representation technique shown at Page 314 onwards, and an auto correlation function approach using centre clipping, shown at Pages 150-158.
It is known from U.S. Pat. No. 5,678,221 to detect and use pitch period of a signal to replace a noisy portion of a voice signal with a stored section of the signal before the noise, the stored section being repeated at pitch period intervals. However, there is no suggestion of using pitch period to detect tones as distinct from speech.
SUMMARY OF THE INVENTION
According to a first aspect of the invention there is provided apparatus for detecting a tone in an input signal, the apparatus comprising:
a pitch period detector for detecting a pitch period in the input signal, and
a discriminator for determining the presence of the tone according to the detection of the pitch period.
An advantage of using pitch period is that it can be more reliable, or work faster or can use less computation resources than other methods. The reduced computational load in particular can enable channel density to be increased for systems handling many channels. Channel density is a key metric that is commercially significant. The increased speed of detection can be vital in applications such as speech coding, since implementation of some coding standards involves treating tones differently to speech. The increased reliability can be useful for applications such as echo cancellation
Preferred Features
Preferably the pitch period detector comprises an auto correlator for outputting an auto correlation function of the input signal. This can be less computationally intensive than other methods such as frequency domain methods such as Fourier analysis. It is often easier to implement, simpler to program, and gives reliable results.
Preferably, the pitch period detector comprises a peak detector for detecting a peak in the auto correlation function and a comparator for comparing the peak with a total signal power of the input signal, to detect a pitch period. This gives a good indication of periodicity, independent of power level, is relatively easy to implement, and avoids complex recalculation of thresholds.
Preferably the auto correlation function is calculated for selected offsets which have durations corresponding to a range of pitch periods for a human voice. The calculation of the auto correlation function is typically the most computationally intensive part. If the calculation is not made for every possible offset, a great reduction in the amount of calculation can be achieved, without affecting the quality of detection.
Preferably, the discriminator compares a detected pitch period with a predetermined pitch period corresponding to a given tone, to discriminate the presence of that tone. This enables processing to be simplified and rapid.
Preferably the discriminator determines the pitch periods of a sequence of samples, and discriminates from the sequence, the presence of the tone.
Preferably the detected pitch period is used both for noise detection and for tone detection.
Other Aspects of the Invention
Another aspect of the invention provides an echo canceller comprising the tone detector.
Another aspect of the invention provides a speech coder comprising the tone detector.
Another aspect of the invention provides a method of detecting a tone.
Another aspect of the invention provides software for carrying out the method of detecting a tone.
Any of the preferred features may be combined with any of the aspects set out above as would be apparent to a skilled person.
Other advantages will be apparent to a skilled person, particularly in relation to any further prior art other than that discussed above.


REFERENCES:
patent: 4045620 (1977-08-01), Westbrook
patent: 4386239 (1983-05-01), Chien
patent: 4539435 (1985-09-01), Eckmann
patent: 4653098 (1987-03-01), Nakata et al.
patent: 5907793 (1999-05-01), Reams
patent: 5937060 (1999-08-01), Oh
patent: 5971854 (1999-10-01), Pearson et al.
Digital Processing of Speech Signals by Lawrence R Rabiner & Ronald W Schafer—pp. Nos.: Cover/Inside Cover, 126-141, 150-158, 314-319, 372-379.
Detection of Multi-tone Signals based on Energy Operators—Edger F Velez.
Tone Detection using Wavelet Transforms—Glenn A Shelby.
Dual Tone Detection by Goertzel Algorithm.

Affiliated with

Also associated with

Rating

Tone detection using pitch period does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Tone detection using pitch period, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Tone detection using pitch period will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2457390