EP0685836B1 - Method and apparatus for preprocessing an acoustic signal before speech coding - Google Patents

Description

The present invention relates to a method and a acoustic signal preprocessing device supplied to a speech coder. It applies in particular, but not exclusively, to improve the performance of encoders low-speed speech.

Low bit rate speech coders (typically 5 kbit / s for a sampling frequency of 8 kHz) give their best performance on signals presenting a "telephone" spectrum, that is to say in the band 300-3400 Hz and with a pre-emphasis in the high frequencies. These spectral characteristics correspond to the IRS (Intermediate Reference System) template defined by the CCITT in Recommendation P48. This template has been defined for telephone handsets, both in input (microphone) than output (headphones).

However, it is more and more common that the speech encoder input signal has a spectrum more "flat", for example when a hands-on installation free is used, using a response microphone in linear frequency. The usual vocoders are designed to be independent of the input with which they function, and they are not informed of the characteristics of this entry. If microphones different characteristics are likely to be connected to the vocoder, or more generally if the vocoder is likely to receive acoustic signals with different spectral characteristics, it there are then cases where the vocoder is used sub-optimally.

In this context, a main purpose of this invention is to improve the performance of a vocoder by making them less dependent on features spectral signal intended for it.

The method according to the invention consists in subjecting the acoustic input signal to high-pass filtering, to compare the energy of the high pass filtered signal to that of the unfiltered signal to determine a signal state among a first state for which the energy of the filtered high pass signal is greater than a predetermined fraction of the energy of the unfiltered signal, and a second state for which the energy of the filtered high pass signal is lower at the predetermined fraction of the signal energy not filtered, and to send the filtered signal to the encoder input high pass subject to high emphasis frequencies when the signal is in its second state.

The high pass filter used is typically a 400 Hz abrupt cutoff filter, and the fraction predetermined energy is typically 85 to 95%. The first signal state corresponds to IRS characteristics, and the second state corresponds to a flatter spectrum of the input acoustic signal containing proportionally more energy at low frequencies. With the process according to the invention, such a flat spectrum signal is preprocessed (high pass filtering and pre-emphasis) to make spectral characteristics closer to those of the IRS template. The use of high pass filtering for determining the signal state has the advantage over low-pass filtering, to allow the signal to be used filtered to address it (after pre-emphasis) to the input of the vocoder.

Preferably, the determined state of the signal cannot only be changed when the input acoustic signal, or the high pass filtered signal, has an energy greater than one predetermined threshold. In fact, otherwise (by example in a zone of silence or low ambient noise), signal energy is too low to be able to reliably assess its spectral characteristics.

When the acoustic signal is digitized in frames successive, it is detected whether the signal included in each frame is in a first condition corresponding to first state or in a second condition corresponding to second state, and the state of the signal is determined on the basis frame by frame conditions, by not modifying the state determined that after several successive frames show a signal condition different from that corresponding to the previously determined state. This introduces a kind hysteresis which allows variations to be taken into account of the speech signal spectral envelope, due to ambient noise or speech itself (the timbre of the voice is not constant). This reduces the risk of false determination of the signal state, which leads to a better quality of the coded signal and avoids introducing stamp discontinuities which could be due to untimely modifications of the determined state.

The pretreatment device according to the invention includes a high-pass filter receiving the acoustic signal input, means to calculate the energies contained respectively in said acoustic signal and in the high pass filter output signal, means for comparison of calculated energies, and a filter of high frequency pre-emphasis, which input receives the output signal of the high pass filter, and the output of which delivers the signal sent to the encoder input when the comparison means reveal that the output signal from the high pass filter contains less than a fraction predetermined energy of said acoustic signal.

• la figure 1 est un diagramme illustrant les caractéristiques d'un signal acoustique de type IRS et d'un signal de type linéaire ;
• la figure 2 est un schéma synoptique d'un dispositif de prétraitement selon l'invention ;
• la figure 3 est un schéma plus détaillé des moyens de comparaison du dispositif de la figure 2 ; et
• la figure 4 montre des chronogrammes illustrant le mode de détermination de l'état du signal par les moyens de la figure 3.

Other features and advantages of the present invention will appear in the description below of a preferred but nonlimiting embodiment, with reference to the accompanying drawings, in which:

• FIG. 1 is a diagram illustrating the characteristics of an acoustic signal of IRS type and of a linear type signal;
• Figure 2 is a block diagram of a pretreatment device according to the invention;
• Figure 3 is a more detailed diagram of the comparison means of the device of Figure 2; and
• FIG. 4 shows timing diagrams illustrating the mode of determining the state of the signal by the means of FIG. 3.

In Figure 1, the two solid lines correspond to the framing of the IRS template defined for microphones in CCITT Recommendation P48. We see that an IRS type microphone signal has a strong attenuation in the lower part of the spectrum (between 0 and 300 Hz) and a relative emphasis in the high frequencies. In comparison, a linear type signal, provided by example through the microphone of a hands-free system, presents a flatter spectrum, notably not having the strong attenuation at low frequencies (a typical example of such a linear type signal is illustrated by a line in dashes on the diagram in Figure 1).

We take advantage of these spectral properties in the pretreatment device 10 according to the invention, shown diagrammatically in Figure 2. This device processes the input signal supplied by an acoustic signal source to address it to a speech encoder 12. Encoder 12 is a low encoder flow optimized for an IRS type input signal. he can be, among other things, a linear prediction coder with excitation by regular pulse vectors (RP-CELP), as described in document EP-A-0 347 307. The coder 12 has no prior knowledge of the source of the acoustic signal addressed to him.

In the diagram of FIG. 2, the acoustic input signal S _I is the output signal from a microphone 13 which has been amplified and digitized by an analog-digital converter 14. The signal is typically digitized at a rate of 8 kHz sampling, and put into successive 30 ms frames each containing 240 16-bit samples.

The pretreatment device 10 comprises a high-pass filter 16 receiving the input acoustic signal S _I and delivering a filtered signal S _I '. The filter 16 is typically a digital filter of the bi-quad type having an abrupt cutoff at 400 Hz. The energies E1 and E2 contained in each frame of the acoustic input signal S _I and of the filtered signal S _I 'are calculated by two units 17, 18 each carrying out the sum of the squares of the samples of each frame which it receives. The calculated energies E1 and E2 are supplied to a comparison unit 20 which determines the state of the signal in the form of a bit Y which is equal to 0 when it is determined that the signal is of IRS type (state Y _A ), and 1 when it is determined that the signal is rather of the linear type (state Y _B ).

The output of the pretreatment device 10 connected to the input of the encoder 12 is constituted by a terminal of a switch 21, the other terminal of which is connected either to the input of the high-pass filter 16, or to the output of a pre-emphasis filter 22, according to the value of the bit Y delivered by the comparison unit 20. When Y = 0 (state Y _A ), the switch 21 is in the position shown in FIG. 2, and the acoustic signal of input S _I is addressed to the input of the encoder 12. In the other position (Y = 1, state Y _B ), it is the output of the pre-emphasis filter 22 which is addressed to the input of the encoder 12. The pre-emphasis filter 22 receives the filtered high-pass signal S _I 'and applies to it a transfer function of the form H (z) = 1-β / z, in which β denotes a pre-emphasis coefficient which is typically of the order 0.4. Thus, when the acoustic signal is of the linear type, it is transformed by high-pass filtering (filter 16) and pre-emphasis (filter 22) to be addressed to the input of the encoder 12 with spectral characteristics closer to those of the IRS mask. .

Since the high-pass filter 16 has little effect on the input signal when the latter has IRS characteristics, it is also possible to supply the encoder 12 with the filtered high-pass signal S _I 'when determined that the signal is in state Y _A corresponding to the IRS characteristics. A variant of the diagram in FIG. 2 then consists in dispensing with the switch 21 by directly connecting the output of the pre-emphasis filter 22 to the input of the encoder 12, and in controlling the value of the coefficient β in the filter 22 as a function of the value of the status bit Y (for example β = 0 when Y = 0 and β = 0.4 when Y = 1).

The comparison unit 20 is for example in accordance with the diagram illustrated in FIG. 3. The energy E1 of each frame of the input signal S _I is sent to the input of a threshold comparator 25 which delivers a bit Z of value 0 when the energy E1 is less than a predetermined energy threshold, and of value 1 when the energy E1 is greater than the threshold. The energy threshold is typically of the order of -38 dB relative to the signal saturation energy. The comparator 25 serves to inhibit the determination of the state of the signal when the latter contains too little energy to be representative of the characteristics of the source. In this case, the determined state of the signal remains unchanged.

The energies E1 and E2 are sent to a digital divider 26 which calculates the ratio E2 / E1 for each frame. This E2 / E1 ratio is sent to another threshold comparator 27 which delivers a bit X of value 0 when the E2 / E1 ratio is greater than a predetermined threshold, and of value 1 when the E2 / E1 ratio is less than the threshold. This threshold on the E2 / E1 ratio is typically of the order of 0.93. Bit X is representative of a signal condition on each frame. The condition X = 0 corresponds to the IRS characteristics of the input signal (state Y _A ), and the condition X = 1 corresponds to the linear characteristics (state Y _B ). To avoid repeated and untimely changes of state during short-term variations in the voice excitation, the state bit Y is not taken directly equal to the condition bit X, but it results from a processing of successive condition bits X by a state determination circuit 29.

The operation of the state determination circuit 29 is illustrated in FIG. 4, where the upper timing diagram illustrates an example of evolution of the bit X provided by the comparator 27. The status bit Y (lower timing diagram) is initialized to 0 , because IRS characteristics are most frequently encountered. A counting variable V initially set to 0 is calculated frame after frame. The variable V is incremented by one each time the condition X of the signal on a frame differs from that corresponding to the determined state Y (X = 1 and Y = 0, or X = 0 and Y = 1). Otherwise (X = Y = 0 or 1) the variable V is decremented by two units if it is different from 0 and 1, decremented by one unit if it is equal to 1, and kept unchanged if it is equal to 0. As soon as the variable V reaches a predetermined threshold (8 in the example considered), it is reset to 0 and the value of the bit Y is changed, so that it is determined that the signal has changed state. Thus, in the example shown in FIG. 4, the signal is in state Y _A up to frame M, in state Y _B between frames M and N (change of signal source), then again in state Y _A from frame N. Of course, other modes of incrementation and decrementation and other threshold values would be usable.

The above counting mode can for example be obtained by circuit 29 shown in Figure 3. This circuit includes a counter 32 on four bits, of which the bit most significant corresponds to the status bit Y, of which the three Least significant bits represent the counting variable V. X and Y bits are supplied at the input of an OR gate EXCLUSIVE 33 whose output is addressed to the input incrementation of the counter 32 via a AND gate 34 whose other input receives the Z bit supplied by the threshold comparator 25. Thus, the variable V is incremented when X ≠ Y and Z = 1. The inverted output of the gate 33 is supplied to a decrementing input of the counter 32 via another AND gate 35 whose the other two inputs receive the Z bit respectively provided by comparator 25, and the output of an OR gate to three inputs 36 receiving the three least significant bits of the counter 32. Counter 32 is arranged to split the pulses received on its decrement input when its least significant bit is 0 or when one at less than the next two bits is 1, as shown by the OR gate 37 in FIG. 3. Thus, the counter 32 is decremented (by one unit if V = 1 and by two units if V> 1) when X = Y and Z = 1 and V ≠ 0. When the signal energy input is insufficient, we have Z = 0 and the circuit of determination 29 is not activated because AND gates 34, 35 prevent the value of counter 32 from being changed.

Claims (8)

1. Method of preprocessing an acoustic signal upstream of a speech coder (12), characterized in that the acoustic signal (S_I) is subjected to high-pass filtering, the energy (E2) of the high pass filtered signal (S_I') is compared with that (E1) of the unfiltered signal in order to determine a state (Y) of the signal from among a first state (Y_A) for which the energy of the high-pass filtered signal is above a predetermined fraction of the energy of the unfiltered signal and a second state (Y_B) for which the energy of the high-pass filtered signal is below the predetermined fraction of the energy of the unfiltered signal, and the high-pass filtered signal subjected to pre-emphasis of the high frequencies is addressed to the input of the coder (12) when the signal is in its second state.
2. Method according to Claim 1, characterized in that the determined state of the signal is not modified when said acoustic signal or the high-pass filtered signal has energy below a predetermined threshold.
3. Method according to Claim 1 or 2, characterized in that the acoustic signal (S_I) being digitized as successive frames, there is frame-by-frame detection of whether the signal is in a first condition, corresponding to the first state (Y_A), for which the calculated energy (E2) of the frame of the high-pass filtered signal (S_I') is above the predetermined fraction of the calculated energy (E1) of the frame of the unfiltered signal (S_I) or in a second condition, corresponding to the second state (Y_B), for which the calculated energy of the frame of the high-pass filtered signal is below the predetermined fraction of the calculated energy of the frame of the unfiltered signal, and the state (Y) of the signal is determined on the basis of the frame-by-frame conditions (X), by modifying the determined state only after several successive frames show a signal condition different from that corresponding to the previously determined state.
4. Method according to Claim 3, characterized in that a counting variable (V) is incremented when the condition (x) of the signal in a frame differs from that corresponding to the determined state (Y) of the signal, in that said counting variable (V) is decremented when the condition of the signal in a frame is that corresponding to the determined state of the signal unless this variable equals zero and in that, when the counting variable (V) reaches a predetermined threshold, it is reset to zero and the signal is determined to have changed state.
5. Device (10) for preprocessing an acoustic signal upstream of a speech coder (12), characterized in that it comprises a high-pass filter (16) receiving said acoustic signal (S_I), means (17, 18) for calculating the energies (E1, E2) contained respectively in said acoustic signal (S_I) and in the output signal (S_I') of the high-pass filter, means (20) for comparing the calculated energies, and a filter (22) for pre-emphasis of the high frequencies, the input of which receives the output signal from the high-pass filter, and the output of which delivers the signal addressed to the input of the coder (12) when the means of comparison (20) reveal that the output signal from the high-pass filter contains less than a predetermined fraction of the energy of said acoustic signal.
6. Device according to Claim 5, characterized in that, with the acoustic signal being digitized as successive frames, the energies (E1, E2) are calculated for each frame by the means of calculation (17, 18), and the means of comparison (20) comprise a comparator (27) which detects frame by frame whether the signal is in a first or a second condition according to whether the ratio (E2/E1) between the calculated energy of the output signal from the high-pass filter (16) and the calculated energy of said acoustic signal (S_I) is above or, respectively, below a predetermined value, and means (29) for determining a state (Y) of the signal from among first and second states (Y_A, Y_B) corresponding respectively to the first and second conditions of the signal per frame, these means (29) modifying the determined state of the signal only after the comparator (27) indicates for several successive frames a signal condition different from that corresponding to the previously determined state, and the pre-emphasis filter (22) being used to filter the signal addressed to the input of the coder (12) only when the means (29) have determined that the signal is in its second state.
7. Device according to Claim 6, characterized in that the means (29) for determining the state of the signal comprise a counter (32) calculating after each frame a counting variable (V), incrementing it when the comparator (27) indicates a signal condition different from that corresponding to the determined state of the signal, decrementing it, unless it equals zero, when the comparator (27) indicates a signal condition identical to that corresponding to the determined state of the signal, and resetting it to zero when it reaches a predetermined threshold, the determined state (Y) of the signal being modified on each reset to zero of the counting variable (V).
8. Device according to Claim 6 or 7, characterized in that it comprises another comparator (25) which compares the calculated energy of said acoustic signal or of the high-pass filtered signal with a predetermined threshold, so as to activate the means (29) of determining state of the signal only when said threshold is exceeded.

Images