WO1997000515A1 - Method and arrangement for determining a pitch frequency in an acoustic signal - Google Patents
Method and arrangement for determining a pitch frequency in an acoustic signal Download PDFInfo
- Publication number
- WO1997000515A1 WO1997000515A1 PCT/SE1996/000670 SE9600670W WO9700515A1 WO 1997000515 A1 WO1997000515 A1 WO 1997000515A1 SE 9600670 W SE9600670 W SE 9600670W WO 9700515 A1 WO9700515 A1 WO 9700515A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- filter
- frequency
- pitch
- signal
- pitch frequency
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/90—Pitch determination of speech signals
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
Abstract
Method and arrangement for determining a pitch frequency in an analog acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies. The mixed signal is first converted to a digital signal in a sampling unit (22). The signal is filtered in a filtering element (24), which is adapted to the speech frequency range, and then analysed with regard to the pitch frequency. The invention is characterised by the fact that the digital signal from the sampling unit (22) and the filtering element (24) can interact so that adjustment to different pitch frequency ranges is made possible during filtering. By filtering the digital signal in the filtering element (24) a first set of filtered signals is obtained for each respective pitch frequency range. Each of the signals in the said first set of filtered signals for each respective pitch frequency range is half-wave rectified in a half-wave rectifier (26) and bandpass filtered in a bandpass filter (27), by means of which a fundamental frequency component for each respective pitch frequency range is filtered out prior to analysis of the pitch frequency.
Description
M THOD AND ARRANGEMENT FOR DETERMINING A PITCH FREQUENCY IN AN ACOUSTIC SIGNAL
TECHNICAL SCOPE
The present invention relates to a method and an arrangement for determining a pitch frequency, that is a fundamental frequency in the harmonic pattern, in an analog acoustic signal in the audible range, which forms part of a mixed signal with signal components from a plurality of acoustic signals with different pitch frequencies.
The present invention also relates to a method and an arrangement for the separation and reconstruction of an acoustic signal in the audible range, which forms part of a mixed signal comprising signal components from a plurality of acoustic signals with different pitch frequencies.
PRIOR ART
It has long been well-known that the human ear in certain instances can distinguish a specific acoustic signal from a mixed signal comprising a plurality of mixed acoustic signals, despite the fact that the specific acoustic signal is not amplified above any of the other signals in the mixed signal. The reason for this is that the human ear has the capacity to distinguish a fundamental frequency in the signal's harmonic pattern, a so-called pitch frequency. This can be utilised in particular when separating speech signals from simultaneous speakers when the said speech signals at times overlap. Different speakers more often than not have different pitch frequencies which generally lie in the range (100-400 Hz). The pitch frequency may vary both for different dialects and for different languages; a difference can generally also be noted between different persons who speak the same language and dialect.
In the sphere of automatic speech recognition and aids for those with hearing difficulties there is a need for a suitable means of separating speech signals from simultaneous speakers using hearing aids. Another problem with the said hearing aids, in addition to the fact that it must be possible to carry out the separation very reliably, is that it must be possible to carry out analysis ofthe speech signal very rapidly, any delays preferably not being perceivable by the ear. At the same time there are naturally limits to the size of the equipment, which must be easily portable by a person with impaired hearing.
In previously known methods for the separation of overlapping speech signals with subsequent facility for reconstruction ofthe separated vocal signals, so-called comb filtering is generally performed. In this type of filtering it is possible, with knowledge ofthe pitch frequency, to filter out all harmonic components which constitute multiples of the fundamental frequency. A problem with these methods lies partly in being able to determine the fundamental frequency with sufficient accuracy, and partly in the fact that the comb filters must be very narrow banded, which gives far too poor a time resolution for natural speech. That is to say a filter with a small bandwidth has a longer pulse response that a broader filter and therefore gives inferior time resolution.
In methods for simple separation and identification without a requirement for subsequent reconstruction, use has also been made of a filter bank with overlapping filters (for example 20 to 100 in the frequency range 0-7 kHz) with filter bandwidths which increase with the centre frequency, like the filtering in the auditory canal. A problem with this method is that difficulties occur when filtering in any frequency range is performed for more than one signal. Each filter may be regarded as one channel for a frequency in a certain range. Since the said channels contain signal components from more than one signal, the reliability of the filtered signal is reduced.
DESCRIPTION OF THE INVENΗON
An object of the present invention is to solve the above-mentioned problem and to provide a method and an arrangement by means of which a pitch frequency in an analog acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies, can be deteirnined more reliably.
This problem is solved in that the mixed signal is converted to a digital signal, filtered in a filtering element which is adapted to the speech frequency range, and analysed with regard to the pitch frequency. The digital signal and the filtering element can interact in order to permit adjustment to different pitch frequency ranges during filtering. After adjustment to one or more pitch frequency ranges, the digital
signal is filtered in the filtering element, a first set of filtered signals being obtained for each respective pitch frequency range. Each ofthe signals in the said first set of filtered signals for the respective pitch frequency range is then half-wave rectified and bandpass-filtered, a fundamental frequency component for each respective pitch frequency range being filtered out prior to analysis of the pitch frequency in a known way.
The invention also relates to a first method and arrangement for the separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies. In the method and arrangement according to the invention a sampling frequency is established as a function of a pitch frequency range of interest. The mixed analog signal is then converted to a digital signal, sampling being performed with the said sampling frequency. The digital signal is filtered in a filter bank which comprises a predetermined number of overlapping fllters, each with a different centre frequency and different filter bandwidth, which increases with the centre frequency and is adjusted to a highest pitch frequency range of interest. The signal obtained from each filter is half-wave rectified and then bandpass-filtered in bandpass filters, which are designed to filter out the fundamental frequency component of a signal in the highest pitch frequency range of interest and which all have the same centre frequency and bandwidth. The pitch frequency ofthe signals obtained from the bandpass filters is established through analysis in pitch frequency analysing elements. The spectral envelope is established after half-wave rectification and lowpass filtering ofthe bandpass filtered signals. The above-mentioned pitch-modulated acoustic signal is reconstructed from the established pitch frequency and spectral envelope.
The invention also relates to a second method and arrangement for the separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies. A sampling frequency is established as a function of a pitch frequency range of interest. The said sampling frequency is used to convert the mixed analog signal to a digital signal, sampling being performed
with the said sampling frequency. A predeterrnined number of samples of the digitalised signal is stored in a first buffer unit, which is connected to a filter. The digital signal is filtered in the said filter, the coefficients of which are updated after each read-out from the buffer unit.
This updating simulates a filter bank comprising a predetermined number of filters, each with a different centre frequency and different filter bandwidth which increases with the centre frequency and is adjusted to a highest pitch frequency range of interest. The signal obtained at the filter output is half-wave rectified and bandpass- filtered in a bandpass filter which is adapted to filter out a fundamental frequency component of a signal in the highest pitch frequency range of interest. The pitch frequency is established through analysis by a known method of a number of signals obtained from the bandpass filter. The bandpass-filtered signal is half-wave rectified and lowpass filtered, following which the resulting lowpass-filtered signal is stored in a second buffer unit. The values of two successive blocks with samples are averaged, thereby establishing the spectral envelope. The above-mentioned pitch- modulated acoustic signal is reconstructed from the established pitch frequency and spectral envelope.
DESCRIPTION OF FIGURES
Figures la- lb show speech signals in the time domain with low and high pitch frequency;
Figure 2 shows a first embodiment ofthe arrangement according to the invention;
Figure 3 shows a second embodiment of the arrangement according to the invention; and
Figure 4 shows a third embodiment ofthe arrangement according to the invention.
PREFERRED EMBODIMENTS
The invention will be described below with reference to the drawings and in particular to Figures 2-4, which show preferred embodiments ofthe invention.
Figure 1 shows two sequences of speech signals with different pitch frequency, that is the periodicity or fundamental frequency in the harmonic pattern is primarily vocal sound. The speech signal shown in Figure la has a lower pitch frequency than the speech signal shown in Figure lb. As will be seen from the Figure, the pitch structure of the signal is reflected in its amplitude modulation. A longer sequence is required for analysing signals with a low pitch frequency than for analysing signals with a higher pitch frequency. Speech signals with a lower pitch frequency should therefore be filtered in fllters with a narrower bandwidth than signals with a higher pitch frequency, partly in view ofthe different time resolution requirements for the different signals, partly in view ofthe fact that the frequency interval between the harmonic components is smaller for signals with a lower pitch frequency than those with a higher frequency.
Figure 2 shows a first embodiment of an arrangement according to the invention for deteimining the pitch frequency of a speech signal which is mixed with other speech signals with different pitch frequencies. In the arrangement shown a plurality of so¬ called filter banks 3a, 3b is used. Each filter bank consists of a predeterrnined number of filters, each of which is adjusted to a certain centre frequency in the range 0-7 kHz. Each of the filter banks (3a, 3b) is adjusted to a special pitch frequency for the filter bank, primarily in the range 100-400 Hz. The filter banks should be adjusted to both low and high pitch frequencies. This adjustment is achieved in that the bandwidth of filters with a given centre frequency is divided between filter banks which are designed for the different pitch frequency ranges. A filter bank adjusted for lower pitch frequencies comprises, for example, filters with smaller bandwidth than a filter bank which is adjusted for signals with higher pitch frequencies. The reason for this will be seen from Figure 2, in which two sequences of speech signals with different pitch frequency are shown.
In the arrangement according to the invention the signal processing is digital and this digitalisation is performed at the input to the arrangement in a sampling unit 2 with preceding anti-deflection lowpass filter 1. The digitalised signal is then transmitted to each of a plurality of filter banks 3a, 3b corresponding to the different pitch frequency ranges, in which filtering is performed. Each filter in a filter bank can be regarded as forming part of a channel. The number of channels in one branch of a filter bank is thus equal to the number of filters in the filter bank. A half-wave rectifier 4al, 4a2, 4bl, 4b2 downstream ofthe filter banks is used in order to form amplitude modulation products which constitute the data-carrying part downstream of the filter banks. Amplitude modulation products of fundamental tone frequency are formed in half-wave rectification when more than one harmonic frequency component is present in the same channel. The fundamental frequency component of the signal is filtered out by means of a bandpass filter 5al, 5a2, 5bl, 5b2, the passband of which is adjusted to the pitch frequency range for which each filter bank is designed.
The pitch frequency value is established by analysis ofthe signals obtained downstream of the said bandpass filter. In the embodiment shown this is performed through autocorrelation in an autocorrelation unit 6al, 6a2, 6bl, 6b2 in each channel and subsequent summation of the autocorrelation products for a filter bank and analysis ofthe said total autocorrelation product in a pitch analysis unit 7a, 7b.
Information on the spectral envelope can be obtained from the magnitude of the filtered signals downstream ofthe bandpass filters in the various channels. The filtered signals are rectified in a second half-wave rectifier 8al, 8a2, 8bl, 8b2 and then lowpass-filtered in lowpass filters 9al, 9a2, 9bl, 9b2. After further half-wave rectification in half-wave rectifiers lOal, 10a2, lObl, 10b2, which are arranged in each channel, the spectral envelope can be established by transmission of the signals on each channel.
In the embodiment ofthe present invention shown in Figure 3 appropriate separation can be performed with a single filter bank 13. The said filter bank 13 is adjusted to the highest pitch frequency range of interest, that is a frequency range in the upper
part ofthe frequency interval 100-400 Hz. The subsequent bandpass filters 15al, 15a2 are also adjusted to this pitch frequency range. Adjustment to a situation in which signals of lower pitch frequency are analysed is achieved by being able to vary the sampling frequency for digitalisation at the input ofthe arrangement. When signals with a lower pitch frequency are to be analysed, speech signals are sampled in a sampling unit 12 with variable sampling frequency at the input of the arrangement with a sampling frequency lower than that for which the filters are designed. By this adjustment ofthe sampling frequency, both the filters in the filter bank 13 and the subsequent bandpass filters 15al, 15a2 are scaled in both centre frequency and bandwidth. If the sampling frequency, for example, is half the sampling frequency for which the filters are designed, the centre frequencies and bandwidths of all filters have been halved and the system has been optimised for signals with half the maximum pitch frequency. The value ofthe pitch frequency can be obtained in the same way as in the embodiment shown in Figure 3 through analysis ofthe bandpass-filtered signals in elements for performing pitch frequency analysis of the bandpass-filtered signals. The said elements may consist, for example, of autocorrelation units 16al, 16a2 on respective channels and a pitch analysis unit in which the autocorrelation products are added together and the total autocorrelation product analysed.
The arrangement according to this embodiment is also adapted to separation and reconstruction of a pitch-modulated signal. The magnitude of the filtered signals downstream of the bandpass filters in the various channels in fact also provides information on the spectral envelope, that is the relative relationship between the magnitudes of the harmonic frequency components with different frequencies in the same signal, the so-called formant structure. The spectral envelope or the magnitude function for each channel is obtained after rectification in half-wave rectifiers 20al, 20a2, which are arranged downstream ofthe bandpass filters 15al, 15a2 in each channel and filtering in lowpass filters 19al, 19a2, the bandwidth of which is the same for each channel in a filter bank. After fiirther half-wave rectification in half-wave rectifiers 20al, 20a2, arranged in each channel, the spectral envelope can be established.
A desired signal in the time plane can be reconstructed in elements for this reconstruction by summation of a number of sinusoidal signals, the frequency of which corresponds to the measured pitch frequency and multiples thereof, and the amplitudes of which are given by the amplitudes ofthe measured spectral envelope for the corresponding frequency.
Finally Figure 4 shows a third embodiment of an arrangement for deterrriining pitch frequency. Instead of a filter bank, this arrangement uses a filter 24 with variable coefficients, which can be updated in order to simulate filters in a filter bank. The coefficients are updated by reading out from a coefficient bank 25. The filter 24 is adapted to correspond to a filter bank for the highest pitch frequency of interest. Adjustment to a situation in which signals of lower pitch frequency are to be analysed is performed in the same way as in the embodiment shown in Figure 4, in that the sampling frequency for digitalisation at the input ofthe arrangement can be varied in a sampling unit 22. When signals of lower pitch frequency are to be analysed, speech signals are sampled at the input of the arrangement at a lower sampling frequency than that for which the filters are designed. By means ofthe said adjustment ofthe sampling frequency, the said filter 24 and subsequent bandpass filters are scaled both in centre frequency and bandwidth. If the sampling frequency, for example, is half of the sampling frequency for which the filters are designed, the centre frequency and bandwidth ofthe first filter and the bandpass filter are halved, the system thereby being optimised for signals with half the maximum pitch frequency.
The arrangement shown in Figure 5 is adjusted to analysis of incoming signals block by block and therefore comprises a first buffer unit 23 in which a predetermined number of samples can be stored. Each sample remains accessible for a certain time until a new sample is fed in. With analysis block by block it is possible, instead of a number of filter channels corresponding to the number of filters in the filter bank, to construct a channel which can simulate one channel at a time in series. Updating is necessary only for the first filter 24 after the sampling unit, which filter can be adjusted to simulate the different filters in a filter bank. Any updating of subsequent fllters 27, 31 is not necessary since these filters are of a similar kind for all channels
from a common filter bank.
Pitch frequency analysis is performed in the same way as before for a signal half- wave rectified in a half-wave rectifier 26 downstream of the first filter, which signal is bandpass-filtered in a bandpass filter 27. Autocorrelation is then performed in an autocorrelation unit for each new set of filter coefficients, whereupon the autocorrelation products for the different sets of coefficients which serve to represent a filter bank are added together in a unit for deterrnining the pitch frequency. The pitch frequency is determined through analysis ofthe total autocorrelation product in a pitch analysis unit 29.
The magnitude ofthe signals filtered out downstream ofthe bandpass filter provides information on the spectral envelope. The magnitude function for each channel is obtained after rectification in a second half-wave rectifier 26, filtering in a lowpass filter and rectification in a third half-wave rectifier 32 in the same way as previously. The arrangement includes a second buffer unit 33, in which the magnitude functions for all channels are stored. In deterrnining the spectral envelope, the amplitude values for two successive blocks of samples stored in the second buffer unit are averaged in an averaging unit 34 in order to counteract the extra slow amplitude modulation which can occur due to the fact that harmonic components from different speech signals may be present in the same channel.
Reconstruction ofthe desired signal in the time plane is performed in a reconstruction unit 35. Reconstruction involves the summation of a number of sinusoidal signals, the frequency of which is the established pitch frequency or various multiples thereof, the amplitudes ofthe signals being given by the amplitudes ofthe measured spectral envelope for the corresponding frequency.
Claims
1. A method for determining a pitch frequency in an analog acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals of different pitch frequencies, the mixed signal being converted to a digital signal in a sampling unit (1; 11; 22), filtered in a filtering element (3a, 3b; 13; 24) which is adapted to the speech frequency range, and analysed with regard to the pitch frequency, characterised in that,
- the digital signal from the sampling unit (1; 11; 22) and the filtering element (3a, 3b; 13, 24) can interact so that adjustment to the different pitch frequency ranges is possible in filtering;
- the digital signal is filtered in the filtering element (3a, 3b; 13, 24), a first set of filtered signals being obtained for each respective pitch frequency range; and
- each ofthe signals in the said first set of filtered signals for each respective pitch frequency range is half-wave rectified in half-wave rectifiers (4al, 4a2,
4bl, 4b2; 14al, 14a2; 26) and bandpass filtered in bandpass fllters (5al, 5a2, 5bl, 5b2; 15al, 15a2; 27), by means of which a fundamental frequency component for each respective pitch frequency range is filtered out prior to analysis with regard to the pitch frequency.
2. A method according to Claim 1, characterised in ύtat,
- the filtering element consists of a plurality of filter banks (3a, 3b), each comprising a predeteπnined number of overlapping fllters with rising centre frequency, the filter bandwidth for filters with the same centre frequency being variable in different filter banks.
3. A method according to Claim 1, characterised in that,
- the filtering element consists of a filter bank (13), which is adjusted to a highest pitch frequency range of interest and which comprises a predeterrnined number of overlapping filters with rising centre frequency and filter bandwidth which increases with the centre frequency; and - an adjustment to lower pitch frequency ranges is obtained by reducing the sampling frequency for conversion of the mixed signal to a digital signal in the sampling unit (12).
4. A method according to Claim 1, characterised in that
- the said filtering element consists of a filter (24), the filter coefficients of which are updated in order to simulate different filters in a filter bank adjusted to a highest pitch frequency range of interest; and
- an adjustment to lower pitch frequency ranges is obtained by reduction of the sampling frequency for conversion of the mixed signal to a digital signal in the sampling unit (22).
5. An arrangement for deterrnining a pitch frequency in an analog acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies, comprising a sampling unit (1; 11; 21), a filtering element (3a, 3b; 13; 24) adapted for filtering of the said mixed signal in the speech frequency range and element for determination of the pitch frequency, characterised in that - the sampling unit (1; 11; 21) and the filtering element (3a, 3b; 13; 24) is adapted to interact so that adjustment to different pitch frequency ranges is obtained; and
- half-wave rectifiers (4al, 4a2, 4bl, 4b2; 14al, 14a2; 26) are adapted for connection to the filtering element (3a, 3b; 13; 24) and bandpass filters (5al, 5a2, 5bl, 5b2; 15al, 15a2; 27), which are adapted to filter out fundamental frequency components for each respective pitch frequency range and to connect the half-wave rectifiers (4al, 4a2, 4bl, 4b2; 14al, 14a2; 26) to the said element for determination ofthe pitch frequency.
6. An arrangement according to Claim 5, characterised in that
- the filtering element comprises a plurality of filter banks (3a, 3b), each consisting of a predeterrnined number of overlapping filters with rising centre frequency and different filter bandwidth for filters with the same centre frequency in different filter banks.
7. An arrangement according to Claim 5, characterised in that
- the filtering element comprises a filter bank (13) adjusted to a highest pitch frequency range of interest, which filter bank consists of a predetermined number of overlapping filters with rising centre frequency and filter bandwidth which increases with the centre frequency; and
- the sampling unit (12) is adapted for a varying sampling frequency, by means of which the arrangement can be adjusted to lower pitch frequency ranges.
8. An arrangement according to Claim 5, characterised in that
- the filtering element comprises a filter (24), the filter coefficients of which are adapted to be updated in order to simulate different fllters in a filter bank adjusted to a highest pitch frequency range of interest; and
- the sampling unit (22) is adapted for a varying sampling frequency, by means of which the arrangement can be adjusted to lower pitch frequency ranges.
9. A method for the separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies, comprising the following stages:
- establishment of a sampling frequency as a function of a pitch frequency range of interest;
- conversion of the mixed analog signal to a digital signal, sampling being performed in a sampling unit (12) with the said sampling frequency;
- filtering of the digital signal in a filter bank (13), which consists of a predetermined number of overlapping filters, each with a different centre frequency and different filter bandwidth which increases with the centre frequency and which is adjusted to a highest pitch frequency range of interest;
- half-wave rectification ofthe signal obtained from each filter in a half-wave rectifier (14al, 14a2); - bandpass filtering of all half-wave rectified signals in bandpass filters (15al, 15a2), which are adapted to filter out the fundamental frequency component of a signal in the highest pitch frequency range of interest and which all have the same centre frequency and bandwidth; - establishment of the pitch frequency through analysis of the signals obtained from the bandpass filters (15al, 15a2);
- half-wave rectification and lowpass filtering ofthe bandpass filtered signals, according to which the spectral envelope is established;
- reconstruction ofthe above-mentioned pitch-modulated acoustic signal from the established pitch frequency and spectral envelope.
10. A method for separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals with different pitch frequencies, comprising the following stages:
- establishment of a sampling frequency as a function of a pitch frequency range of interest;
- conversion of the mixed analog signal to a digital signal, sampling being performed at the said sampling frequency in a sampling unit (22); - storage of a predetermined number of samples of the digitalised signal in a buffer unit (23);
- filtering ofthe digital signal in a filter (24), the coefficients of which are updated after each read-out from the buffer unit, thereby simulating a filter bank which comprises a predeterrnined number of filters, each with a different centre frequency and different filter bandwidth, which increases with the centre frequency and which is adjusted to a highest pitch frequency range of interest;
- half-wave rectification of the signal obtained at the output of the filter (24) in a half-wave rectifier (26); - bandpass filtering ofthe half-wave rectified signal in a bandpass filter (27), which is adapted to filter out a fundamental frequency component for a signal in the highest pitch frequency range of interest;
- establishing of the pitch frequency through analysis of a number of signals obtained from the bandpass filter;
- half-wave rectification and lowpass filtering off the bandpass-filtered signal, following which the resulting lowpass-filtered signal is stored in a second buffer unit (33); - averaging of the values from two successive blocks in an averaging unit
(34), by which means the spectral envelope is established;
- reconstruction ofthe above-mentioned pitch modulated acoustic signal from the established pitch frequency and spectral envelope.
11. An arrangement for separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals of different pitch frequencies, comprising:
- an element (12) for digitalisation ofthe mixed signal with varying sampling frequency;
- a filter bank (13) comprising a predetermined number of filters with overlapping filter band in the speech frequency range, all filters having different centre frequencies and a filter bandwidth for each respective filter which increases with the centre frequency and which is adjusted to a highest pitch frequency range of interest;
- half-wave rectifiers (Hal, 14a2) which are connected to respective filters in the filter bank (13);
- bandpass filters (15al, 15a2) equal to the number of filters in the filter bank, which filters are adjusted to the said highest pitch frequency range of interest and which are adapted to filter each respective half-wave rectified signal;
- element for performing pitch frequency analysis of the bandpass-filtered signals;
- element for establishing the spectral envelope for the filtered signals; and
- element for reconstructing a desired signal in the time plane from a combination ofthe result ofthe pitch frequency analysis and establishing of the spectral envelope.
12. An arrangement for the separation and reconstruction of a pitch-modulated acoustic signal, which forms part of a mixed analog signal lying in a speech frequency range and composed of a plurality of acoustic signals of different pitch frequencies, comprising: - element (22) for digitalisation of the mixed signal with varying sampling frequency:
- a first buffer unit (23) for storing a predetermined number of samples ofthe digitalised signal;
- a filter (24) which is connected to the buffer unit (23) and the filter coefficients of which can be updated after each completed read-out of the contents ofthe buffer unit, the filter coefficients being updated so as to simulate filters in a filter bank;
- a first half-wave rectifier (26) which is connected to the said filter;
- a bandpass filter (31) which is adapted to filter out a fundamental frequency component within the pitch frequency range in question;
- element for analysis ofthe pitch frequency for the signals obtained after filtering;
- a second half-wave rectifier (30) which is connected to the said bandpass filter; - a lowpass filter (31) which is connected to the output of the second half- wave rectifier;
- a third half-wave rectifier (32) which connects the lowpass filter to a second buffer unit;
- a circuit (34) for averaging of the amplitude values from two successive blocks of samples; and
- element (35) for reconstructing the desired signal in the time plane, which element is adapted to combine the result of the pitch frequency value analysis with the result from the averaging circuit.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| SE9502217A SE9502217L (en) | 1995-06-19 | 1995-06-19 | Method and apparatus for determining a pitch frequency in a speech signal |
| SE9502217-4 | 1995-06-19 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO1997000515A1 true WO1997000515A1 (en) | 1997-01-03 |
Family
ID=20398661
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/SE1996/000670 WO1997000515A1 (en) | 1995-06-19 | 1996-05-23 | Method and arrangement for determining a pitch frequency in an acoustic signal |
Country Status (2)
| Country | Link |
|---|---|
| SE (1) | SE9502217L (en) |
| WO (1) | WO1997000515A1 (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4044204A (en) * | 1976-02-02 | 1977-08-23 | Lockheed Missiles & Space Company, Inc. | Device for separating the voiced and unvoiced portions of speech |
| DE3305045A1 (en) * | 1983-02-14 | 1984-08-16 | Siemens AG, 1000 Berlin und 8000 München | Arrangement for identifying the basic speech frequency |
| US4833714A (en) * | 1983-09-30 | 1989-05-23 | Mitsubishi Denki Kabushiki Kaisha | Speech recognition apparatus |
| EP0334023A2 (en) * | 1988-03-25 | 1989-09-27 | Telenorma Gmbh | Method for speech signals detection |
| US4937868A (en) * | 1986-06-09 | 1990-06-26 | Nec Corporation | Speech analysis-synthesis system using sinusoidal waves |
| DE4315677A1 (en) * | 1993-05-06 | 1994-11-17 | Ekkehard Dr Ing Stuerzebecher | Circuit arrangement for determining the fundamental frequency from a signal which is not bandwidth-limited and contains harmonics and interference signals, in particular for determining the fundamental voice frequency from a voice and speech signal |
-
1995
- 1995-06-19 SE SE9502217A patent/SE9502217L/en not_active IP Right Cessation
-
1996
- 1996-05-23 WO PCT/SE1996/000670 patent/WO1997000515A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4044204A (en) * | 1976-02-02 | 1977-08-23 | Lockheed Missiles & Space Company, Inc. | Device for separating the voiced and unvoiced portions of speech |
| DE3305045A1 (en) * | 1983-02-14 | 1984-08-16 | Siemens AG, 1000 Berlin und 8000 München | Arrangement for identifying the basic speech frequency |
| US4833714A (en) * | 1983-09-30 | 1989-05-23 | Mitsubishi Denki Kabushiki Kaisha | Speech recognition apparatus |
| US4937868A (en) * | 1986-06-09 | 1990-06-26 | Nec Corporation | Speech analysis-synthesis system using sinusoidal waves |
| EP0334023A2 (en) * | 1988-03-25 | 1989-09-27 | Telenorma Gmbh | Method for speech signals detection |
| DE4315677A1 (en) * | 1993-05-06 | 1994-11-17 | Ekkehard Dr Ing Stuerzebecher | Circuit arrangement for determining the fundamental frequency from a signal which is not bandwidth-limited and contains harmonics and interference signals, in particular for determining the fundamental voice frequency from a voice and speech signal |
Also Published As
| Publication number | Publication date |
|---|---|
| SE503875C2 (en) | 1996-09-23 |
| SE9502217L (en) | 1996-09-23 |
| SE9502217D0 (en) | 1995-06-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kataoka et al. | The relationship between spectral characteristics and perceived hypernasality in children | |
| EP0737351B1 (en) | Method and system for detecting and generating transient conditions in auditory signals | |
| US4051331A (en) | Speech coding hearing aid system utilizing formant frequency transformation | |
| US4905285A (en) | Analysis arrangement based on a model of human neural responses | |
| US4536844A (en) | Method and apparatus for simulating aural response information | |
| WO1987003128A1 (en) | Analog signal encoding and decoding apparatus and methods | |
| WO1990011593A1 (en) | Method and apparatus for speech analysis | |
| JP2002082685A (en) | Device and method for expanding audio bandwidth | |
| Howard | Peak‐picking fundamental period estimation for hearing prostheses | |
| JPH0251303B2 (en) | ||
| WO1997000515A1 (en) | Method and arrangement for determining a pitch frequency in an acoustic signal | |
| Knorr | Reliable voiced/unvoiced decision | |
| Traunmüller | The sentiphone: a tactual speech communication aid | |
| EP0114814B1 (en) | Apparatus and method for articulatory speech recognition | |
| Chang | Two schemes of speech compression system | |
| Darwin et al. | Speech perception seen through the ear | |
| JPH02253298A (en) | Voice pass filter | |
| Zwicker | Peripheral preprocessing in hearing and psychoacoustics as guidelines for speech recognition | |
| JP2740866B2 (en) | Electronics | |
| WO1993009531A1 (en) | Processing of electrical and audio signals | |
| Sandler et al. | A system for recognition and translation of the speech of handicapped individuals | |
| Takefuta et al. | Intelligibility of speech signals spectrally compressed by a sampling-synthesizing technique | |
| Rosen et al. | Sinusoidal plucks and bows are not categorically perceived, either | |
| EP1652171A1 (en) | Method for analysing signals containing pulses | |
| Jovičić | Dichotic integration of speech-envelope information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AK | Designated states |
Kind code of ref document: A1 Designated state(s): US |
|
| AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): AT BE CH DE DK ES FI FR GB GR IE IT LU MC NL PT SE |
|
| DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
| 122 | Ep: pct application non-entry in european phase |