EP1433166B1 - Speech extender and method for estimating a broadband speech signal from a narrowband speech signal - Google Patents

Description

The invention relates to a speech extender according to the preamble of claim 1 and to a method for estimating a wideband speech signal using a narrow-band speech signal according to the preamble of claim 6.

For compressing the data transmission rate in speech signals, speech encoders, also referred to as speech codecs, are used. They are mainly used in mobile radio systems. GSM (Global System for Mobile Communication) mobile systems use speech encoders that work according to the Linear Predictive Coding (LPC) method. Here, in contrast to a waveform coding not the speech signal itself, but its development process is reproduced in the human speech tract. At a sampling rate of the speech signal of 8 kHz results in a data rate of 104 kbit / s at a resolution of 13 bits. With GSM, this data rate is reduced to a constant 13 KBit / s (so-called code rate) by means of LPC. For this purpose, full-rate codecs or improved full-rate codecs are used, for example, in GSM. With a half-rate codec (Half Rate Codec), the bit rate can be greatly reduced with correspondingly reduced voice quality, namely to 5.6 kbit / s.

In future communication systems such as UMTS (Universal Mobile Telecommunications System) voice encoders are used, which can encode voice signals with a variable bit rate. One such speech coder is, for example, the Adaptive Multirate (AMR) speech coder, which allows encoding at various bit rates. It was designed for GSM cellular systems, but is also intended for use in UMTS mobile radio systems are used as standard speech coders.

In such speech encoders, the bit rate can be adapted to the bandwidth available for transmitting the coded speech signal. If sufficient bandwidth is available for transmission, the speech signal is coded at a high bit rate. This is also referred to as broadband coding. Otherwise, ie with a low bandwidth, is coded with a low bit rate (narrowband coding).

The adaptation of the bit rate can take place during the transmission of a speech signal. For this purpose, the bandwidth of a transmission channel in the form of the available bit rate is continuously measured. If the available bit rate drops below a predetermined threshold during a transmission of the speech signal, the coding is switched so that the speech signal is narrow-band coded.

For example, two different codes with correspondingly different bit rates may be provided. A broadband coding takes place, for example, at a sampling frequency of about 16 kHz, while a narrow-band coding takes place at a sampling frequency of 8 kHz. In the first case, a voice frequency range up to 8 kHz, in the second case up to 4 kHz is covered. The problem, however, is caused by the switching of the bit rate fluctuation of the signal quality and the associated quality variation of a communication link. Due to the predetermined threshold, the switching takes place relatively abruptly, so that the quality of the connection can suddenly drop during a conversation.

By using a so-called wideband speech extender (Wideband Speech Extender) or short language extender a much better can be achieved without changing a communication protocol Speech quality even with narrow-band coding and the quality-reducing effect of the abrupt switching slightly reduced. Such a broadband speech extender is for example from WO 01/56021 known. For this purpose, the wideband speech extender estimates the broadband speech signal by synthesizing the speech frequency components above 4 kHz on the basis of the narrowband signal. This avoids excessive loss of quality when switching the bit rate for encoding. The method provided by the wideband speech extender for adapting the bit rate of a speech signal works quite well for almost all speakers, in particular it offers quite acceptable speech quality.

US-A-5 581 652 discloses a speech extender which, in a training phase, generates codebooks with which narrowband speech signals are converted into broadband in an application phase.

With the present invention, the speech quality of a speech extender, in particular a wideband speech extender, is to be further improved. Furthermore, a method for estimating a wideband speech signal based on a narrow-band speech signal is to be specified, which enables an improved speech quality.

This improvement is achieved by a Sprachxtender with the features of claim 1. Furthermore, a significant improvement in the estimation of wideband speech signals is achieved by driving with the features of claim 6. Preferred embodiments of the speech extender and of the method for estimating a wideband speech signal on the basis of a narrowband speech signal are the subject of the respective dependent claims.

The core of the invention is to make an adaptation to a communication terminal and / or to a speaker during a voice signal transmission. As a result, the voice quality can be improved again compared to known methods and speech extender.

Specifically, the invention relates to a speech extender which is adapted to estimate a wideband speech signal based on a narrow-band speech signal. Further it is adaptive such that it adapts a codebook to a communication terminal and / or to a speaker. The adaptation takes place during a voice transmission. This allows the Sprachxtender to constantly adapt to the remote party.

The speech extender analyzes and stores at least one speech parameter and uses it for adaptation. The at least one speech parameter is a wideband speech parameter that occurs during a speech transmission. In particular, the at least one speech parameter can be speaker and / or communication terminal-specific. For example, the speech extender can be used in various mobile phones and adapt to their acoustic properties. Furthermore, it can address different users, i. adapt to their acoustic properties such as different speech frequency spectra. As speech parameters therefore preferably characteristic acoustic properties of the communication terminal and / or the speaker are used, such as frequency characteristics, attenuation of certain frequencies or frequency ranges and the frequency spectrum of the voice of the speaker. Such speech parameters can be determined in particular by measurements during a speech transmission.

Preferably, the speech extender makes estimates by evaluating at least one stored speech parameter. As already explained above, different speech parameters can be used for the adaptation. These are stored after their determination and are therefore available for adaptation at any time. It would also be conceivable to constantly update the stored speech parameters in order to always be optimally adapted to the current acoustic conditions.

Finally, the speech extender can be used in a speech coder of a mobile and / or base station designed for a third generation mobile radio system. In particular, the third generation mobile radio system may be UMTS.

If the mobile station is a mobile radio terminal, preferably a mobile telephone, the speech extender is preferably implemented in hardware, in particular in an integrated circuit, and / or in software. An implementation in hardware offers the advantage that the speech extender can be integrated on a chip together with other essential circuit elements of the mobile radio terminal. For example, a chip manufacturer may offer such language extenders to mobile terminal manufacturers. On the other hand, an implementation in software offers the advantage of easier changeability of the speech extender, and above all of the subsequent change, in particular if the software of the speech extender is stored in an erasable and rewritable memory such as an EEPROM.

Furthermore, the invention relates to a method for estimating a wideband speech signal based on a narrow-band speech signal. According to the method, an adaptation of a codebook to a communication terminal and / or to a speaker is carried out during estimation.

At least one language parameter is analyzed, stored and used for adaptation. The at least one speech parameter is a wideband speech parameter that occurs during a speech transmission. In particular, the at least one speech parameter may be speaker and / or communication terminal-specific.

The method can be advantageously used in a speech coder of a mobile and / or base station, the or are designed for a third generation mobile radio system, in particular UMTS. In a preferred embodiment, the mobile station is a mobile radio terminal and the method is implemented in hardware, in particular in an integrated circuit, and / or at least partially in software.

The invention will be explained below with reference to exemplary embodiments.

First, the operation of a broadband speech extender will be exemplified to better understand the use of the invention.

In order to generate speech frequency components above a certain frequency, a broadband excitation signal and broadband filter coefficients are required for the synthesis filter in the speech coder. Since usually only the narrow-band excitation signal and the narrow-band filter coefficients are known, it is necessary to carry out a transformation from "narrowband" to "broadband". This is done by means of a broadband speech extender. The excitation signal can be extended, for example, by a non-linear signal processing. Another possibility is to superimpose the excitation signal with white noise.

The filter coefficients can be estimated by using two codebooks. The entries of the codebooks represent possible sets of filter coefficients. In a training phase, a narrowband and a wideband codebook are trained. Since they are trained simultaneously with the same excitation signal (once narrowband and once wideband), the relationship between the entries of both codebooks is known. For example, entry 1 of the narrow-band codebook corresponds to the entry 2 of the wideband codebook.

1. 1. Vom schmalbandigen Sprachsignal werden die schmalbandigen Filterkoeffizienten berechnet.
2. 2. Diese Filterkoeffizienten werden mit den Einträgen des schmalbandigen Codebuchs verglichen.
3. 3. Der am besten passende Eintrag wird ausgewählt. Da der Zusammenhang zwischen den Codebüchern bekannt ist, werden die optimalen breitbandigen Filterkoeffizienten gleichzeitig gefunden.

In the broadband speech extender deployment phase, both codebooks are used as follows:

1. 1. From the narrow-band speech signal, the narrow-band filter coefficients are calculated.
2. 2. These filter coefficients are compared with the entries of the narrowband codebook.
3. 3. The best matching entry is selected. Since the relationship between the codebooks is known, the optimal wideband filter coefficients are found simultaneously.

On the basis of the found broadband filter coefficients and the extended excitation signal speech signal sections are synthesized.

During language extension, therefore, the narrow-band filter coefficients are calculated from the narrow-band speech. These coefficients are compared with the entries of the narrow-band coefficient codebook and the best matching entry is chosen. Since, as already mentioned above, the relationship between the codebooks is known, the optimal filter coefficients for the speech synthesis filter of the wideband speech extender are estimated in this way.

1. 1. Während der breitbandigen Sprachübertragung werden die "echten" Filterkoeffizienten (FK) berechnet und der am besten passende Eintrag des Codebuches mit breitbandigen Sprachkoeffizienten wird gesucht.
2. 2. Der Codebucheintrag wird neu berechnet. $$\mathrm{CB}-\mathrm{Vector}\left(\mathrm{new}\right)=\mathrm{C}*\mathrm{FK}+\left(1-\mathrm{c}\right)*\mathrm{CB}\sim \mathrm{Vector}\left(\mathrm{old}\right)$$
   
3. 3. Während der schmalbandigen Sprachübertragung werden die breitbandigen Filterkoeffizienten anhand des modifizierten Codebuches geschätzt.

The method can also be implemented adaptively, for example in the following steps:

1. 1. During wideband speech transmission, the "true" filter coefficients (FK) are calculated and the best matching entry of the codebook with broadband speech coefficients is searched.
2. 2. The codebook entry is recalculated. $$\mathrm{CB}-\mathrm{Vector}\left(\mathrm{new}\right)=\mathrm{C}*\mathrm{FK}+\left(1-\mathrm{c}\right)*\mathrm{CB}~\mathrm{Vector}\left(\mathrm{old}\right)$$
   
3. 3. During narrowband speech transmission, the wideband filter coefficients are estimated from the modified codebook.

1. 1. Während der breitbandigen Sprachübertragung werden die "echten" Filterkoeffizienten (FK)berechnet.
2. 2. Vom breitbandigen Sprachsignal wird das schmalbandige Sprachsignal berechnet und die schmalbandigen Filterkoeffizienten werden ermittelt.
3. 3. Der am besten passenden Eintrag des Codebuches mit schmalbandigen Koeffizienten für die schmalbandigen Filterkoeffizienten wird gesucht.
4. 4. Mittels des bekannten Zusammenhanges zwischen den Codebüchern wird der optimale Eintrag des Codebuches mit breitbandigen Koeffizienten gesucht.
5. 5. Dieser Codebucheintrag wird neu berechnet. $$\mathrm{CB}-\mathrm{Vector}\left(\mathrm{new}\right)=\mathrm{C}*\mathrm{FK}+\left(1-\mathrm{c}\right)*\mathrm{CB}-\mathrm{Vector}\left(\mathrm{old}\right)$$
   
6. 6. Während der schmalbandigen Sprachübertragung werden die breitbandigen Filterkoeffizienten anhand des modifizierten Codebuches geschätzt.

An alternative is realized in the following steps:

1. 1. During wideband speech transmission, the "true" filter coefficients (FK) are calculated.
2. 2. The wideband speech signal is used to calculate the narrow-band speech signal and the narrow-band filter coefficients are determined.
3. 3. The best matching entry of the codebook with narrowband coefficients for the narrowband filter coefficients is searched.
4. 4. By means of the known relationship between the codebooks the optimal entry of the codebook is searched with broadband coefficients.
5. 5. This codebook entry is recalculated. $$\mathrm{CB}-\mathrm{Vector}\left(\mathrm{new}\right)=\mathrm{C}*\mathrm{FK}+\left(1-\mathrm{c}\right)*\mathrm{CB}-\mathrm{Vector}\left(\mathrm{old}\right)$$
   
6. 6. During narrowband speech transmission, the wideband filter coefficients are estimated from the modified codebook.

The speech extender according to the invention achieves a further improvement of the speech quality. It can be used particularly advantageously in all communication systems in which variable bit rate speech coders can be used, which can code both narrowband and broadband, for example in the case of UMTS.

Claims (10)

1. Speech extender which is embodied so that it estimates a wideband speech signal on the basis of a narrowband speech signal, characterized in that it matches a codebook adaptively to a communication terminal and/or to a speaker during a wideband speech transmission and features means with which it analyses, stores and uses for adaptation at least one wideband speech parameter which arises during speech transmission.
2. Speech extender according to claim 1, characterized in that the at least one speech parameter is speaker-specific and/or communication terminal-specific.
3. Speech extender according to claim 2, characterized in that it estimates by evaluating at least one stored speech parameter.
4. Speech extender according to one of the previous claims, characterized in that it is used in a speech codec of a mobile station and/or base station which are or is embodied for a mobile radio system of the third generation, especially UMTS.
5. Speech extender according to claim 4, characterized in that the mobile station is a mobile radio terminal and the speech extender is implemented in hardware, especially in an integrated circuit, and/or at least partly in software.
6. Method for estimating a wideband speech signal based on a narrowband speech signal characterized in that, in a wideband speech transmission an adaptive matching of a codebook to a communication terminal and/or to a speaker is performed and at least one wideband speech parameter which occurs during speech transmission is analysed, stored and used for adaptation.
7. Method according to claim 6, characterized in that the at least one speech parameter is speaker- and/or communication terminal-specific.
8. Method according to claim 7, characterized in that estimation is undertaken on the basis of at least one stored speech parameter.
9. Method according to one of the claims 6 to 8, characterized in that it is used in a speech codec of a mobile station and/or base station which are or is embodied for a mobile radio system of the third generation, especially UMTS.
10. Method according to claim 9, characterized in that the mobile station is a mobile radio terminal and the method is implemented in hardware, especially in an integrated circuit, and/or at least partly in software.