EP1825461A1

EP1825461A1 - Method and apparatus for artificially expanding the bandwidth of voice signals

Info

Publication number: EP1825461A1
Application number: EP06840370A
Authority: EP
Inventors: Bernd Geiser; Peter Jax; Stefan Schandl; Herve Taddei; Aulis Telle; Peter Vary
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 2005-07-13
Filing date: 2006-06-30
Publication date: 2007-08-29
Anticipated expiration: 2026-06-30
Also published as: KR100915733B1; CN100568345C; CN101676993B; CA2580622C; ES2309969T3; CN101676993A; ATE407424T1; EP1825461B1; JP4740260B2; US8265940B2; CN101061535A; DE102005032724B4; US20080126081A1; PL1825461T3; CA2580622A1; DE102005032724A1; KR20070090143A; DK1825461T3; DE502006001491D1; WO2007073949A1

Abstract

The method involves providing a broadband input speech signals, and determining signals components of the signals from an increased band of the signals. The temporal and spectral envelopes of the components are determined. The information of the envelopes is coded by a coder (1), and the coded information is provided to execute the increment of the bandwidth of the signals. The coded information is decoded and the temporal and spectral envelopes are generated from the coded information to create a bandwidth increased output speech signals. An independent claim is also included for a device for artificially increasing bandwidth of a speech signal.

Description

description

Method and device for artificially expanding the bandwidth of speech signals

The invention relates to a method and a device for artificially expanding the bandwidth of speech signals.

Speech signals cover a wide frequency range, which ranges from the basic speech frequency, which is dependent on the speaker in the range between 80 to 160 Hz, to the frequencies beyond 10 kHz. However, for voice communication over certain transmission media, such as telephones, only a limited portion is transmitted for bandwidth efficiency, ensuring a sentence intelligibility of about 98%.

Corresponding to the minimum bandwidth specified for the telephone system from 300 Hz to 3.4 kHz, a speech signal can essentially be subdivided into three frequency ranges. Each of these frequency ranges characterizes specific speech characteristics as well as subjective sensations. This results in lower frequencies below about 300 Hz, essentially during voiced speech segments, such as vowels. This frequency range in this case contains tonal components, i. H. in particular the basic voice frequency and, depending on the pitch, possibly some harmonics.

For the subjective perception of volume and dynamics of a speech signal, these depth frequencies are essential. In contrast, the speech base frequency can be perceived by a human listener due to the psychoacoustic property of the virtual pitch sensation even in the absence of the depth frequencies from the harmonic structure in higher frequency ranges. Thus, medium frequencies in the range of about 300 Hz to about 3.4 kHz in voice activities are fundamental. additionally available in the voice signal. Their time-variant spectral coloring by several formants as well as the temporal and spectral fine structure characterize the respective spoken sound or phoneme. In such a way, the middle frequencies carry the bulk of the information relevant to the intelligibility of the language.

On the other hand, during unvoiced sounds, as is especially the case with sharp sounds such as "s" or "f", high frequency components are generated above about 3.4 kHz. Even so-called plosives such as "k" or "t" have a broad spectrum with strong high-frequency components. Therefore, the signal in this upper frequency range has more of a noise-like than a tonal character. The structure of the formants also present in this area is relatively time-invariant, but differs for different speakers. The high frequency components are essential for the clarity, the presence and the naturalness of a speech signal, since without these high frequency components, the speech seems dull. In addition, such high frequency components enable a better differentiation of fricatives and consonants, whereby these high frequency components thereby also ensure an increased intelligibility of the speech.

In a transmission of a voice signal via a voice communication system, which has a transmission channel with limited bandwidth, it is always desirable and always the goal, the speech signal to be transmitted with the best possible quality from a transmitter to a

To be able to transmit receivers. However, the speech quality is a subjective quantity with a plurality of components, of which the intelligibility of the speech signal is the most important for such a speech communication system.

In modern digital transmission systems, a relatively high speech intelligibility can already be achieved. there It is known that by extending the telephone bandwidth to high frequencies (greater than 3.4 kHz) as well as to low frequencies (less than 300 Hz), it is possible to improve the subjective assessment of the speech signal. In the sense of a subjective quality improvement, a bandwidth which is increased compared to the customary telephone bandwidth is therefore to be striven for in systems for voice communication. A possible approach here is to modify the transmission and to effect a wider transmitted bandwidth by means of coding techniques, or alternatively to perform an artificial bandwidth expansion. Such an expansion of the bandwidth widens the frequency bandwidth to the range of 50 Hz to 7 kHz at the receiving end. Using suitable signal processing algorithms, parameters of the broadband model are determined from short segments of a narrowband speech signal using pattern recognition methods, which are then used to estimate the missing signal components of the speech. In the method, the narrow-band speech signal becomes a broadband equivalent with frequency components in the range 50

Hz to 7 kHz generated and an improvement in subjectively perceived speech quality causes.

Current speech signal and audio signal coding algorithms are increasingly using artificial bandwidth expansion techniques. For example, in the wideband range (acoustic bandwidth 50 Hz to 7 kHz), speech coding standards such as the AMR-WB (Adaptive Multirate Wideband) coding / decoding algorithm are used. In this AMR-WB standard, upper frequency subbands (frequency range approximately 6.4 to 7 kHz) are extrapolated from low frequency components. In such encoding-decoding methods, the bandwidth extension is generally generated by a comparatively small amount of side information. These side information may be, for example, filter coefficients or gain factors, wherein the filter coefficients may be generated, for example, by an LPC (Linear Prediction Filter) method. This secondary information tions are transmitted in a coded bit stream to a receiver. Other standards based on the extension of the bandwidth technique are currently seen in the AMR-WB + and extended aacPlus speech / audio coding / decoding techniques. Methods designed to encode and decode information are referred to as codecs and include both an encoder and a decoder. Any digital telephone, whether built for a fixed or mobile network, includes such a codec that converts analog to digital signals and digital to analog. Such a codec can be implemented in hardware or in software.

In present implementations of speech / audio signal encoding algorithms using the bandwidth expansion technique, components of an extension band, for example in the frequency range of 6.4 to 7 kHz, are encoded and decoded using the aforementioned LPC encoding technique. In this case, an LPC analysis of the extension band of the input signal is performed in an encoder and the LPC coefficients and the amplification factors of subframes of a residual signal are encoded. In a decoder, the remainder of the expansion band is generated and the transmitted gain factors and the LPC synthesis filters are used to generate an output signal. The procedure described above can be applied either directly to the wideband input signal or else to a subband signal of the extension band that is downsampled in the limit range or in the critical range.

The extended aacPlus encoding standard uses SBR (Spectral Band Replication) technology. The broadband audio signal is split into frequency subbands by means of a 64-channel QMF filter bank. For the high-frequency filter bank channels, a sophisticated and technically advanced parametric coding is applied to the sub-bands of the signal components, requiring and using a large number of detectors and estimators to control the bitstream contents. Although it is already possible to achieve an improvement in the speech quality of speech signals in particular with the known standards and coding-decoding methods, a further improvement of this speech quality is nevertheless to be aimed for. Moreover, the standards and encoding-decoding methods discussed above are very expensive and have a very complex structure.

It is therefore the object of the present invention to provide a method and a device for the artificial extension of the bandwidth of speech signals with which an improved speech quality and improved speech intelligibility can be achieved. This should also be able to be realized in a relatively simple and low-effort manner.

This object is achieved by a method having the features of claim 1, and a device having the features of claim 23, solved.

In a method according to the invention for artificially expanding the bandwidth of speech signals, the following steps are carried out: a) provision of a broadband input speech signal; b) determining the signal components of the wideband input speech signal required for bandwidth extension from an extension band of the wideband input speech signal; c) determining the temporal envelopes of the bandwidth expansion signal components; d) determining the spectral envelope of the bandwidth expansion signal components; e) encoding the information of the temporal envelope and the spectral envelope and providing the encoded information for performing the extension of the bandwidth; and f) decoding the encoded information and generating the temporal envelope and the spectral envelope from the encoded information to produce a bandwidth-expanded output speech signal.

By means of the method according to the invention, an improvement in speech intelligibility and speech quality in the transmission of speech signals can be achieved, whereby speech signals are also understood as audio signals. In addition, the inventive method is also very robust against interference during transmission.

Advantageously, the signal components required for bandwidth expansion are determined by filtering, in particular bandpass filtering, from the wideband input speech signal, whereby a simple and low-cost selection of the required signal components can be performed.

Determining the temporal envelopes in step c) is preferably carried out independently of the determination of the spectral envelopes in step d). As a result, the determination of the envelope takes place in a precise manner, whereby a mutual influence can be avoided.

Preferably, prior to encoding the temporal envelope and the spectral envelope in step e), quantization of the temporal envelope and the spectral envelope is performed. In an advantageous manner, the signal powers of spectral subbands of the signal components intended for bandwidth expansion are determined in step d) for determining the spectral envelopes. The determination of the characterization of the temporal and the spectral envelope can thereby be carried out very accurately.

For determining the signal powers of the spectral subbands, signal segments of the bandwidth tenerweiterung certain signal components generated, these signal segments in particular transformed, in particular FF (Fast Fourier) transformed, are. Furthermore, the signal powers of temporal signal segments of the signal components intended for bandwidth expansion are advantageously determined in step c) for determining the time-dependent envelopes. In a labor-saving manner, the determination of the required parameters can thereby be carried out.

Advantageously, in step f) the encoded information for reconstructing the temporal envelope and the spectral envelope are decoded.

An excitation signal is advantageously generated in a decoder from a signal transmitted to the decoder, the transmitted signal having such a signal power in the frequency range which corresponds to that of the extension signal of the wideband input speech signal, which enables generation of an excitation signal. A modulated narrowband signal having a band range with frequencies below the frequencies of the band range of the extension band of the wideband input speech signal for generating the excitation signal is preferably transmitted to the decoder. The excitation signal preferably has harmonics of the fundamental frequency of the signal transmitted to the decoder.

From the decoded information of the temporal envelope and the excitation signal, a first correction factor is advantageously determined. Furthermore, from the first correction factor and the excitation signal, a reconstructive shaping of the temporal envelope, in particular by a multiplication of the first correction factor with the excitation signal, is performed. In addition, the reconstructed shaping of the temporal envelopes is filtered in an advantageous manner and impulse responses are generated during filtering. From the impulse responses and the reconstructed research tion of the temporal envelope, a reconstructive shaping of the spectral envelope is performed. Furthermore, the signal components of the expansion band of the wideband input speech signal are reconstructed from the reconstructed shaping of the spectral envelope. The reconstruction of the temporal and the spectral envelopes can be carried out very reliably and very accurately.

In an advantageous embodiment, a narrowband signal having a band range with frequencies below the frequencies of the extension band of the broadband input signal is transmitted to the decoder.

The bandwidth-expanded output speech signal is advantageously determined from the narrow-band signal transmitted to the decoder and the reconstructed shaping of the spectral envelope, in particular from a summation of these two signals, and is provided as an output signal of the decoder. As a result, an output signal can be generated and provided which ensures high speech intelligibility and speech quality.

The steps a) to e) are preferably carried out in an encoder, which is preferably arranged in a transmitter. The encoded information generated in step e) is advantageously transmitted as a digital signal to the decoder. At least step f) is preferably performed in a receiver with the decoder located in the receiver. However, it can also be provided that all steps a) to f) of the method according to the invention are carried out in a receiver. In this case, steps a) to e) in the receiver are replaced by an estimation method (to be implemented differently). The steps a) to e) can also be carried out separately in a transmitter. The wideband input speech signal advantageously comprises a bandwidth between about 50 Hz and about 7 kHz. The extension band of the wideband input speech signal preferably comprises the frequency range of about 3.4 kHz to about 7 kHz. Furthermore, the narrowband signal comprises a

Signal range of the wideband input speech signal from about 50 Hz to about 3.4 kHz.

A device according to the invention for artificially expanding the bandwidth of speech signals to which a broadband input speech signal can be applied comprises at least the following components: a) means for determining the signal components of the wideband input speech signal required for bandwidth expansion from an extension band of the wideband input speech signal; b) means for determining the temporal envelope of the signal components intended for bandwidth extension; c) means for determining the spectral envelope of the signal components intended for bandwidth extension; d) an encoder for encoding the temporal envelope and the spectral envelope and providing the encoded information for performing the extension of the bandwidth; and e) a decoder for decoding the encoded information and generating the temporal envelope and the spectral envelope from the encoded information to produce a bandwidth-expanded output speech signal.

The inventive device enables improved speech quality and improved speech intelligibility of speech signals during transmission in communication devices, such as mobile devices or ISDN devices.

The means in a) to d) are advantageously designed as encoders. The encoder can be in a transmitter or in a receiver, wherein the decoder is arranged in a receiver.

Advantageous embodiments of the method according to the invention can, insofar as it is transferable, also be regarded as advantageous embodiments of the device according to the invention.

An embodiment of the invention will be explained in more detail with reference to schematic drawings. Show it:

1 shows an encoder of a device according to the invention; and

2 shows a decoder of a device according to the invention.

In the invention explained in more detail below, the term speech signals also includes audio signals. In FIG 1 and FIG 2 the same or functionally identical elements are provided with the same reference numerals.

1 shows a schematic block diagram representation of an encoder 1 of a device according to the invention for the artificial extension of the bandwidth of speech signals. The coder 1 can be implemented as an algorithm both in hardware and in software. In the exemplary embodiment, the encoder 1 comprises a block 11, which is designed for bandpass filtering of a broadband input speech signal s _w ' _b (k). Furthermore, the encoder 1 comprises a block 12 and a block 13, which are connected to the block 11. Block 12 is designed to determine the temporal envelope of the signal components intended for bandwidth expansion, which are determined from an extension band of the wideband input speech signal. Similarly, the block 13 is configured to determine the spectral envelope of the bandwidth expansion signal components determined from the extension band of the wideband input speech signal. Moreover, it can be seen from the representation in FIG. 1 that the block 12 and the block 13 are connected to a block 14, the block 14 for quantizing the temporal envelope and the spectral envelope generated by the blocks 12 and 13, respectively be, is trained.

FIG. 1 further shows a block 2, which is designed as a bandpass filter and to which the broadband input speech signal s _w ' _b (k) is applied. The block 2 is further connected to a further block 3, wherein the block 3 is formed as a further encoder.

In the exemplary embodiment, the encoder 1 and the blocks 2 and 3 are arranged in a first telephone set. In the exemplary embodiment, the broadband input speech signal has a bandwidth of approximately 50 Hz to approximately 7 kHz. According to the invention, as can be seen from the illustration in FIG. 1, this wideband input speech signal s _w ' _b (k) is applied to the bandpass filter or block 11 of the coder 1.

By means of this block 11, the signal components required for bandwidth expansion from the expansion band, which in the exemplary embodiment comprises a bandwidth of about 3.4 kHz to about 7 kHz, are determined. The signal components required for the bandwidth expansion are characterized by the signal s _eb (k) and are transmitted as an output signal of the block 11 to the two blocks 12 and 13. In block 12, from this signal s _eb (k) the temporal

Envelope determined. In a corresponding manner, the spectral envelope of the signal components which are characterized by the signal s _eb (k) is determined in block 13.

This determination of the temporal envelope and the spectral envelope will be explained in more detail below. First, the required for bandwidth expansion

Segmented signal components s _eb (k) and transformed these fenestrated signal segments. The segmentation of the signal s _eb (k) takes place in frames with a length of each of k samples. All subsequent steps and subalgorithms are performed frame by frame. Each speech frame (eg with 10 ms or 20 ms or 30 ms duration) can advantageously be subdivided into several subframes (duration eg 2.5 or 5 ms).

The windowed signal segments are then transformed. In the exemplary embodiment, a transformation into the frequency domain is carried out by means of an FFT (Fast Fourier Transform). The FFT-transformed signal segments are determined according to the following formula 1):

In this formula 1) Nf denotes the FFT length or the frame size, μ denotes the frame index and Mf denotes the overlap of the frames of the windowed signal segments. Furthermore, W _y (κ) denotes the window function. Subsequently, in the frequency domain, the signal power is calculated in subbands of the frequency range of the extension band. This calculation of the signal strength or the signal power takes place in accordance with the following formula 2):

In this formula 2) λ denotes the index of the corresponding subband, wherein EB _λ characterizes that set which contains all FFT interval ranges i with non-zero coefficients in the λ th frequency space window w _λ (i). The signal powers i _y (μ, λ) of the subbands according to formula 2) characterize the information of the spectral envelopes which are transmitted to a decoder.

The determination of the time envelopes in the time period is performed in a manner similar to the determination of the spectral envelopes and is based on short-term windowed ones Segments of the band-pass filtered wideband input speech signal s _w ' _b (k). Thus, signal segments of the signal s _eb (k) are also taken into account in the determination of the time envelopes. For each fenestrated segment the signal power is calculated according to formula 3) below:

In this formula 3), N _t denotes the frame length, v denotes the frame index and M _t again denotes the overlap of the frames of the signal segments. It should be noted that, in general, the frame length N _t and the overlap of the frames M _{t used} to extract the temporal envelopes are smaller and much smaller than the corresponding magnitudes Nf and Mf, respectively Spectral envelopes are used.

An alternative for extracting the temporal envelope parameters from the signal s _eb (k) is to _perform a Hilbert transform (90 ° phase shift filter) of the signal s _eb (k). A summation of

Short segment signal powers of the filtered parts and the original parts of the signal s _eb (k) gives the short time envelope, which is downsampled to determine the signal powers P _t (y). The signal powers

P _t (y) of the signal segments then characterize the temporal envelope information.

The signals s _{p (y} ^ or ^ _(μλ) characterizing the temporal envelope and the spectral envelope, which characterize the extracted parameters of the signal _{powers according to formulas} 2) and 3), are quantized and coded in block 14. The output signal of the block 14 is a digital signal BWE, which characterizes a bit stream which contains in coded form information of the temporal envelope and the spectral envelope. This digital signal BWE is transmitted to a decoder, which will be explained in more detail below. It should be noted that in the case of a redundancy between the extracted parameters of the signal strengths according to formulas 2) and 3), a common coding, such as may be made possible, for example, by vector quantization, can be carried out.

As can also be seen from the illustration in FIG. 1, the wideband input speech signal s _w ' _b (k) is also transmitted to the block 2. By means of this block 2 designed as a bandpass filter, the signal components of a narrowband range of the wideband input speech signal s _w ' _b (k) are filtered. The narrowband range in the exemplary embodiment is between 50 Hz and 3.4 kHz. The output signal of the block 2 is a narrowband signal s _nb (k) and is transmitted to the block 3, which is formed in the embodiment as a further encoder. In this block 3, the narrowband signal s _nb (k) is encoded and transmitted as a digital signal BWN as a bit stream to the decoder explained below.

FIG. 2 shows a schematic block diagram illustration of such a decoder 5 of a device according to the invention for artificially expanding the bandwidth of speech signals. As can be seen in FIG. 2, the digital signal BWN is first transmitted to a further decoder 4 which decodes the information contained in the digital signal BWN and in turn generates the narrowband signal s _nb (k) from it. Furthermore, the decoder 4 generates a further signal s _sι (k), which contains side information. These side information may be, for example, gain factors or filter coefficients. This signal s _sι (k) of the decoder 5 to block wear exceeds 51st The block 51 is formed in the embodiment for generating an excitation signal in the frequency range of the extension band, to which the information of the signal s _sι (k) are taken into account. In addition, the decoder 5, which is arranged in the embodiment in a receiver, a block 52, which is designed for decoding the transmitted over a transmission distance between the encoder 1 and the decoder 2 signal BWE. It should be noted that also the digital signal BWN is transmitted via this transmission path between the encoder 1 and the decoder 5. As can be seen from the illustration in FIG. 2, both the block 51 and the block 52 are connected to decoder areas 53 to 55. The functional principle of the decoder 5 or the sub-steps of the method according to the invention carried out in the decoder 5 are explained in more detail below.

As already mentioned above, the information contained in the encoded digital signal BWE is decoded in block 52 and the signal powers, which are calculated according to formulas 2) and 3) and which characterize the temporal envelope and the spectral envelope, are reconstructed. As can be seen from the illustration in FIG. 2, the excitation signal S _{^ x} (Jc) generated in block 51 is the

Input signal for the reconstructive shaping of the temporal envelope and the spectral envelope. This excitation signal can essentially be any

Signal must be, as an essential condition for this signal must be that it has a sufficient signal power in the frequency range of the extension band of the wideband input spectral signal s _w ' _b (k). For example, the narrowband signal s _nb (k) or any noise are used as excitation signal s _exc (k) is a modulated version. As already mentioned, this excitation signal is responsible for the fine structuring of the spectral envelope and the temporal envelope in the signal components of the extension band of a broadband output _speech signal s _wb (k). For this reason, it is advantageous for this excitation signal s (k) to be present in one of these is generated such that it has the harmonics of the fundamental frequency of the narrowband signal s _nb (k).

In the case of hierarchical speech coding, one way to achieve this is to use parameters of the further decoder 4. For example, if A _{k is} a fractional or real shift of the fundamental frequency and b is the adaptive codebook LTB gain in a CELP narrowband decoder, then harmonic frequency excitation is at an integer multiple of the current fundamental frequency by LTP synthesis filtering a bandpass filter (frequency range of the extension band) from an arbitrary signal n _eb (k) possible.

The excitation signal results according to the following formula 4):

* «(*) =» * (*) + / (*) ^• * «(* - **)

In this case, the LTP amplification factor can be reduced or limited by the function f (b) in order to be able to prevent overstimulation of the generated signal components of the expansion band. It should be noted that a plurality of further alternatives can be carried out in order to be able to carry out synthetic broadband excitation by means of parameters of a narrowband codec.

Another way to generate an excitation signal is to modulate the narrowband signal s _nb (k) with a sine function at a fixed frequency or by directly using an arbitrary signal n _eb (k), as already defined above was, is performed. It should be emphasized that the method used for the generation of the excitation signal is completely independent of the generation of the digital signal BWE and the format of this digital signal BWE and the decoding of this digital signal BWE. Therefore, can In this regard, an independent setting be performed.

In the following, the reconstructive shaping of the temporal envelope will be explained in more detail. As already mentioned, the digital signal BWE is decoded in the block 52 and the parameters of the signal power characterizing the temporal envelope and the spectral envelope, which are calculated according to the formulas 2) and 3), corresponding to the signals j _{(v )} and s _{p (μ> λ)} . As can be seen from the illustration in FIG. 2, a reconstructive shaping of the temporal envelopes is first carried out in the exemplary embodiment. This is done in the decoder area 53. For this purpose, the excitation signal S _exc ik) and the signal j _{(v) are} transmitted to this decoder area 53. As shown in FIG. 2, the excitation _signal s _exc [k _{] is transmitted} both to a block 531 and to a multiplier 532. The signal -J _{(v) is also} transmitted to the block 531. From these signals transmitted to block 531, a scalar correction factor gi (k) is generated.

This scalar correction factor gi (k) is transferred from the block 531 to the multiplier 532. In the multiplier 532, the excitation _signal s _exc [k _] then becomes scalar

Correction factor gi (k) multiplied and generates an output signal S _0x [Ic], which characterizes the reconstructed shaping of the temporal envelope. This output signal s _exc [k] has the approximately correct temporal envelope, but is still inaccurate or imprecise with respect to the correct frequency, which in a subsequent step, the performing a reconstructed shaping of the spectral envelope is required to this imprecise frequency to be able to adjust the required frequency.

As can be seen in FIG. 2, the output signal S _0x [Ic) is transmitted to a second decoder area 54 of the decoder 5, to which the signal ^ _{(μ λ) is also} transmitted. The second decoder area 54 has a block 541 and a Block 542, wherein the block 541 is designed to filter the output signal S _0x (Ic). From the output signal ^s _exc (k) ^and the signal ^ _{(μ λ)} , an impulse response h (k) is generated, which is transmitted from block 541 to block 542. In this block 542, the reconstructive shaping of the spectral envelope is then carried out from the output signal ^s _exc (k) ^and d of the impulse response h (k). This reconstructed spectral envelope is then characterized by the output ^s _exc (k) of block 542.

In the exemplary embodiment shown according to FIG. 2, the following is based on the generation of the output signal of the second decoder area 54, a reconstructing shaping of the temporal envelope in a third decoder area 55 of the decoder 5 is again carried out. This reconstructing shaping of the temporal envelope takes place analogously as it is carried out in the first decoder region 53. In this case, in this third decoder region 55, a second scalar correction factor g ₂ (k) is generated by the block 551 from the output signal ^s _exc (k) ^and the signal J _(v) , which is transmitted to a multiplier 552. As the output signal of the third decoder region 55 of the decoder 5, the signal s _eb (k) characterizing the signal components required for the bandwidth extension is then provided. This signal s _eb (k) is transmitted to a summer 56, to which also the narrowband signal s _nb (k) is transmitted. By summing the narrowband signal s _nb (k) and the signal s _eb (k), the bandwidth-extended output signal s _w ^° _b (k) is generated and provided as the output signal of the decoder 5.

It should be noted that the embodiment shown in FIG. 2 is merely exemplary and that the invention already has a single reconstructive shaping of the temporal envelopes, as is done in the first decoder region 53, and a single reconstructive shaping of the spectral envelopes, as in the second decoder region 54 carried out is sufficient. It should also be noted that it can also be provided that the reconstructive shaping of the spectral envelope in the second decoder area 54 is performed before the reconstruction of the temporal envelope in the first decoder area 53. This means that the second decoder region 54 is arranged before the first decoder region 53 in such an embodiment. Likewise, however, it can also be provided that the alternate execution of a reconstructing shaping of the temporal envelope and a reconstructive shaping of the spectral envelope is continued and, for example, in the embodiment shown in FIG. 2, a further decoder region is arranged adjacent to the third decoder region 55 in turn, a reconstructive shaping of the spectral envelope is carried out.

As already stated above, the invention is advantageously used in the exemplary embodiment for a wideband input speech signal having a frequency range of about 50 Hz to 7 kHz. Likewise, the invention is provided in the exemplary embodiment for the artificial extension of the bandwidth of speech signals, wherein the extension band is predetermined by the frequency range of about 3.4 kHz to about 7 kHz. However, it can also be provided that the invention is used for an extension band, which is located in a low-frequency frequency range. For example, the extension band may comprise a frequency range of about 50 Hz or even lower frequencies, up to a frequency range of about 3.4 kHz. It should be explicitly emphasized that the method according to the invention for the artificial extension of the bandwidth of speech signals can also be used such that the extension band comprises a frequency range which is at least partially above a frequency of about 7 kHz and for example up to 8 kHz, in particular 10 kHz , or even higher. As already explained, a reconstructed formation of the temporal envelope in the first decoder area 53 is riert according to FIG 2 by a multiplication of the scalar first correction factor gi (k) and the excitation signal S _{^ x} (Jc) generation. It should be noted that a multiplication in

Period corresponding to a convolution operation in the frequency domain, resulting in the following formula 5):

*! (*) = «(*) ^• *« (*);

As long as the spectral envelope is in principle not changed by the first decoder region 53, the first scalar correction factor or gain gi (k) should have strict low-pass frequency characteristics.

For calculating these amplification factors or this first correction factor gi (k), the excitation signal S _{^ 0} (It) is segmented and analyzed in a manner already described above for the segmentation and the analysis of the extraction of the temporal envelope or the generation of the Signal s _pfy) from the signal s _eb (k) is performed in the encoder 1 by means of the block 12. The ratio between the decoded signal power as calculated by formula 3) and the analyzed result of the signal strength P ^ fy) results in a desired gain γ (v) for the vth signal segment. This amplification factor of the vth signal segment is calculated in accordance with the following formula 6):

From this amplification factor γ (v), the amplification factor or first correction factor gi (k) is calculated by interpolation and low-pass filtering. Low-pass filtering is crucial to This gain factor or this first correction factor gi (k) to limit the spectral envelope.

The reconstructive shaping of the spectral envelope of the required signal components of the extension band is determined by filtering the output signal S _0x (Ic), which characterizes the reconstructed shaping of the temporal envelope. The filter operation can be implemented in the period or in the frequency domain. In order to avoid a large time dispersion or time expansion of the impulse response h (k), the corresponding frequency characteristic H (z) can be smoothed. In order to be able to determine the desired frequency characteristics, the output signal ^s _exc (k) of the first decoder region 53 is analyzed in order to be able to find the signal powers of the Pf ° (\ i, X). The desired amplification factor Φ (μ, λ) of a corresponding subband of the frequency range of the expansion band is calculated according to the following formula 7):

The frequency characteristic H (μ, i) of the shape filters of the spectral envelope can be calculated by interpolation of the amplification factor Φ (μ, λ) and with a smoothing taking into account the frequency. If the shaping filter of the spectral envelope is to be used in the period, for example by a linear phase FIR filter, the filter coefficients can be calculated by an inverse FF transformation of the frequency characteristic H (μ, i) and a subsequent windowing.

As explained and shown by the above explanations, the reconstructive shaping of the temporal envelope influences the reconstructive shaping of the spectral envelopes and vice versa. It is therefore advantageous that, as explained in the exemplary embodiment and shown in FIG. provides an alternate performance of reconstructing a temporal envelope and a spectral envelope in an iterative process. Thereby, a substantially improved match of the temporal and spectral envelopes of the signal components of the enhancement band, which are reconstructed in the decoder and the corresponding temporal and spectral envelopes generated in the coder, can be achieved.

In the described embodiment according to FIG. 2, one and a half times the iteration (reconstruction of the temporal envelopes, reconstruction of the spectral envelopes and renewed reconstruction of the temporal envelopes) is carried out. Bandwidth expansion, as enabled by the invention, facilitates the generation of an excitation signal having harmonics at the correct frequency, for example at an integer multiple of the fundamental frequency of the current sound. It should be noted that the invention can also be applied to downsampled subband signal components of the broadband input signal. This is advantageous when a low computational effort is required.

Advantageously, the encoder 1 and the blocks 2 and 3 are arranged in a transmitter, wherein logically, the process steps carried out in the blocks 2 and 3 and the encoder 1 are then also carried out in the transmitter. The block 4 as well as the decoder 5 can advantageously be arranged in a receiver, whereby it is also clear that the preliminary steps carried out in the decoder 5 and in the block 4 are executed in the receiver. It should be noted that the invention can also be implemented in such a way that the method steps carried out in the coder 1 are carried out in the decoder 5 and are thus carried out exclusively in the receiver. It can be provided that the signal powers, which are calculated according to the formulas 2) and 3), in the deco 5 can be estimated. In particular, the block 52 is designed to estimate these parameters of the signal powers. This embodiment allows the concealment of potential transmission errors of the side information transmitted in the digital signal BWE. By a temporary

Estimation of lost parameters of an envelope, for example by a loss of data, can be a troublesome switching of the signal bandwidth can be prevented.

In contrast to the known methods for artificially widening the bandwidth of speech signals, in the invention no transfer of already used amplification factors and filter coefficients is carried out as secondary information, but only the desired temporal and spectral envelopes are transmitted as side information to a decoder. Gain factors and filter coefficients are only then calculated in the decoder, which is arranged in a receiver. It can thereby be achieved that the artificial extension of the bandwidth in the receiver can be analyzed and, if necessary, corrected in a low-effort manner. In addition, the method according to the invention and the device according to the invention are very robust against disturbances of the excitation signal, whereby, for example, such a disturbance of a received narrowband signal can be caused by transmission errors.

By carrying out the analysis, the transmission and the reconstructing shaping of the temporal and spectral envelopes separately, a very good resolution or splitting in the time domain and in the frequency domain can be achieved both in the time domain and in the frequency domain. This leads to a very good reproducibility of both stationary sounds and sounds as well as transient or short-term signals. For speech signals, in particular, the reproduction of stop consonants and plosives benefits from the significantly improved time resolution. Unlike conventional bandwidth extensions, the invention allows the frequency shaping to be performed by linear phase FIR filters rather than LPC synthesis filters. It can also be achieved that typical artifacts ("filter ringing") can be reduced, In addition, the invention allows a very flexible and modular design, which also allows the individual blocks in the receiver or in the Decoder 5 can advantageously be exchanged or set in an advantageous manner For such a change or adjustment, no change of the transmitter or the coder 1 or the format of the transmission signal with which the coded information to the decoder 5 or the receiver ü In addition, different decoders can be operated with the method according to the invention, as a result of which a restoration of the broadband input signal can be carried out with different precision as a function of the available computing power.

It should also be noted that the received parameters which characterize the spectral and temporal envelopes can be used not only for an extension of the bandwidth, but also for the support of subsequent signal processing blocks, such as post-filtering, or additional coding steps such as Transformer encoder, can be used.

The resulting narrowband speech signal s _nb (k), as available to the bandwidth _expansion algorithm, may be present, for example, after a reduction of the sampling frequency by a factor of 2 at a sampling rate of 8 kHz.

With the invention and the underlying principle of bandwidth expansion, it is possible to generate a broadband excitation of information of the G.729AH standard. The data rate of the secondary signals transmitted in the digital signal BWE Information can be about 2 kbit / s. Moreover, in the invention, a relatively low-complexity calculation system or a relatively low complex computational effort is required, which is less than 3 WMOPS. In addition, the inventive method and the device according to the invention is very robust against baseband disturbances of the G.729AH standard. The invention may also be used advantageously for use in voice-over-IP. Moreover, the method according to the invention and the device according to the invention are compatible with TDAC envelopes. Last but not least, the invention also has a very modular and flexible structure and a modular and flexible conception.

Claims

claims

A method of artificially extending the bandwidth of speech signals characterized by the steps of: a) providing a wideband input speech signal

b) determining the signal components (s _eb (k)) required for the bandwidth extension of the wideband input speech signal (s _w ' _b (k)) from an extension band of the wideband input speech signal (s _w ' _b (k)); c) determining the temporal envelope of the signal components intended for bandwidth extension (s _eb (k)); d) determining the spectral envelopes of the signal components intended for bandwidth broadening (s _eb (k)); e) encoding the information of the temporal envelope and the spectral envelope and providing the encoded information for performing the extension of the bandwidth; f) decoding the coded information and generating the temporal envelope and the spectral envelope from the coded information to produce a bandwidth-expanded output _speech signal (s _wb (k)).

2. The method according to claim 1, characterized in that the signal components required for bandwidth expansion (s _eb (k)) by filtering, in particular a bandpass filter

Filtering, are determined from the wideband input speech signal (s _w ' _b (k).

3. The method according to any one of the preceding claims, characterized in that the determination of the temporal envelope in step c) is performed independently of the determination of the spectral envelope in step d).

4. The method according to any one of the preceding claims, characterized in that prior to the coding of the temporal envelope and the spectral envelope in step e) a quantization of the temporal envelope and the spectral envelope is performed.

5. The method according to any one of the preceding claims, characterized in that in step d) for determining the spectral envelope, the signal powers (i _y (μ, λ)) of spectral subbands of the band width extension specific signal components (s _eb (k)) are determined ,

6. The method according to claim 5, characterized in that for determining the signal powers (i _y (μ, λ)) of the spectral

Subbands signal segments of the band width extension specific signal components (s _eb (k)) are generated, these signal segments in particular transformed, in particular FF-transformed, are.

7. The method according to any one of the preceding claims, characterized in that in step c) for determining the temporal envelope, the signal strengths (P _t (y)) of temporal signal segments of the band width extension determined signal components (s _eb (k)) are determined.

8. The method according to any one of the preceding claims, characterized in that in step f) the encoded information for reconstructing forms of the temporal envelope and the spectral envelope are decoded.

9. The method according to any one of the preceding claims, characterized in that an excitation signal (s _^ ik)) in a decoder (5) from a to the decoder (5) transmitted signal (s _sι (k)) generated is, wherein the transmitted signal (s _sι (k)) such a signal strength in the frequency range, which corresponds to that of the extension band of the wideband input speech signal (s _w ' _b (k)), which comprises generating a excitation signal allows.

10. The method according to claim 9, characterized in that to the decoder (5) a modulated narrow-band signal having a band range below the band range of the extension band of the wideband input speech signal (s _w ' _b (k)) for generating the excitation signal (s _^ ik) ) is transmitted.

11. The method according to claim 9 or 10, characterized in that the excitation signal (s ^ k)) harmonics of the fundamental frequency of the decoder (5) transmitted signal (s _sι (k)).

12. The method according to claim 8 and 11, characterized in that from the decoded information of the temporal envelope and the excitation signal (s ^ k)) a first correction factor (g _λ (k)) is determined.

13. The method according to claim 12, characterized in that from the first correction factor (gι (k)) and the excitation _signal (s _∞c ik)) a reconstructive shaping of the temporal envelope, in particular by a multiplication of the first correction factor (g \ (& )) with the excitation signal (s _^ ik)).

14. The method according to claim 13, characterized in that the reconstructed shaping of the temporal envelope is filtered and in the filtering impulse responses (h (k)) are generated.

15. The method according to claim 14, characterized in that from the impulse responses (h (k)) and the reconstructed shaping of the temporal envelope, a reconstructive shaping of the spectral envelope is performed.

16. The method according to claim 15, characterized in that from the reconstructed shaping of the spectral envelope, the signal components (s _eb (k)) of the extension band of the wideband input speech signal (s _w ' _b (k)) are reconstructed.

17. The method according to any one of the preceding claims, characterized in that to a decoder (5) a narrow-band signal (s _nb (k)) with a band range below the extension band of the wideband input signal (s _w ' _b (k)) is transmitted.

18. The method according to claim 16 and 17, characterized in that the bandwidth-expanded output speech signal (s _w ^° _b (k)) from the decoder (5) transmitted narrow-band signal (s _nb (k)) and the reconstructed shaping of the spectral Ein - enveloping, in particular from a summation of these two

Signals, is determined and provided as an output signal of the decoder (5).

19. The method according to any one of the preceding claims, characterized in that the steps a) to e) are performed in an encoder (1) and the coded information generated in step d) are transmitted as a digital signal (BWE) for decoding.

20. The method according to any one of the preceding claims, characterized in that the wideband input speech signal (s _w ' _b (k)) comprises a bandwidth between about 50 Hz and about 7 kHz.

21. The method according to any one of the preceding claims, characterized in that the extension band of the wideband input speech signal (s _w ' _b (k)) comprises the frequency range from about 3.4 kHz to about 7 kHz.

22. The method according to claim 17, characterized in that the narrow-band signal (s _nb (k)) comprises a signal range of the wideband input speech signal (s _w ' _b (k)) of about 50Hz to about 3.4 kHz.

23. Device for the artificial extension of the bandwidth of speech signals to which a broadband input speech signal (s _w ' _b (k)) can be applied, characterized by a) means for determining the signal components (s _eb (k)) required for the bandwidth extension of the broadband input speech signal (s _w ' _b (k)) from an extension band of the wideband input speech signal (s _w ' _b (k)); b) means for determining the temporal envelope of the signal components intended for bandwidth extension

(s _eb (k)); c) means for determining the spectral envelope of the signal components intended for bandwidth extension

(s _eb (k)); d) an encoder (1) for encoding the temporal envelope and the spectral envelope and providing the encoded information for performing the extension of the bandwidth; and e) a decoder (5) for decoding the encoded information and generating the temporal envelope and the spectral envelope from the encoded information for generating a bandwidth-extended output speech signal (s _w ^° _b (k)).

24. The device according to claim 23, characterized in that the means in a) to d) as encoder (1) are formed.