Concept for Switching of Sampling Rates at Audio Processing Devices Description The present invention is concerned with speech and audio coding, and more particularly to an audio encoder device and an audio decoder device for processing an audio signal, for which the input and output sampling rate is changing from a preceding frame to a current frame. The present invention is further related to methods of operating such devices as well as to computer programs executing such methods. Speech and audio coding can get the benefit of having a multi-cadence input and output, and of being able to switch instantaneously and seamlessly for one to another sampling rate. Conventional speech and audio coders use a single sampling rate for a determine output bit-rate and are not able to change it without resetting completely the system. It creates then a discontinuity in the communication and in the decoded signal. On the other hand, adaptive sampling rate and bit-rate allow a higher quality by selecting the optimal parameters depending usually on both the source and the channel condition. It is then important to achieve a seamless transition, when changing the sampling rate of the input/output signal. Moreover, it is important to limit the complexity increase for such a transition. Modern speech and audio codecs, like the upcoming 3GPP EVS over LTE network, will need to be able to exploit such a functionality. Efficient speech and audio coders need to be able to change their sampling rate from a time region to another one to better suit to the source and to the channel condition. The change of sampling rate is particularly problematic for continuous linear filters, which can only be applied if their past states show the same sampling rate as the current time section to filter. More particularly predictive coding maintains at the encoder and decoder over time and frame different memory states. In code-excited linear prediction (CELP) these memories are usually the linear prediction coding (LPC) synthesis filter memory, the de-emphasis filter memory and the adaptive codebook. A straightforward approach is to reset all memories when a sampling rate change occurs. It creates a very annoying discontinuity in the decoded signal. The recovery can be very long and very noticeable. Fig. 1 shows a first audio decoder device according to prior art. With such an audio decoder device it is possible to switch to a predictive coding seamlessly when coming from a non-predictive coding scheme. This may be done by an inverse filtering of the decoded output of non-predictive coder for maintaining the filter states needed by predictive coder. It is done for example in AMR-WB+ and USAC for switching from a transform-based coder, TCX, to a speech coder, ACELP. However, in both coders, the sampling rate is the same. The inverse filtering can be applied directly on the decoded audio signal of TCX. Moreover, TCX in USAC and AMR-WB+ transmits and exploits LPC coefficient also needed for the inverse filtering. The LPC decoded coefficients are simply re-used in the inverse filtering computation. It is worth to note that the inverse filtering is not needed if switching between two predictive coders using the same filters and the same sampling-rate. Fig. 2 shows a second audio decoder device according to prior art In case the two coders have a different sampling rate, or in case when switching within the same predictive coder but with different sampling rates, the inverse filtering of the preceding audio frame as illustrated in Fig. 1 is no more sufficient. A straightforward solution is to resample the past decoded output to the new sampling rate and then compute the memory states by inverse filtering. If some of the filter coefficients are sampling rate dependent as it is the case for the LPC synthesis filter, one need to do an extra analysis of the resampled past signal. For getting the LPC coefficients at the new sampling rate fs_2 the autocorrelation function is recomputed and the Levinson-Durbin algorithm applied on the resampled past decoded samples. This approach is computationally very demanding and can hardly be applied in real implementations. The problem to be solved is to provide an improved concept for switching of sampling rates at audio processing devices. In a first aspect the problem is solved by an audio decoder device for decoding a bitstream, wherein the audio decoder device comprises: a predictive decoder for producing a decoded audio frame from the bitstream, wherein the predictive decoder comprises a parameter decoder for producing one or more audio parameters for the decoded audio frame from the bit-stream and wherein the predictive decoder comprises a synthesis filter device for producing the decoded audio frame by synthesizing the one or more audio parameters for the decoded audio frame; a memory device comprising one or more memories, wherein each of the memories is configured to store a memory state for the decoded audio frame, wherein the memory state for the decoded audio frame of the one or more memories is used by the synthesis filter device for synthesizing the one or more audio parameters for the decoded audio frame; and a memory state resampling device configured to determine the memory state for synthesizing the one or more audio parameters for the decoded audio frame, which has a sampling rate, for one or more of said memories by resampling a preceding memory state for synthesizing one or more audio parameters for a preceding decoded audio frame, which has a preceding sampling rate being different from the sampling rate of the decoded audio frame, for one or more of said memories and to store the memory state for synthesizing of the one or more audio parameters for the decoded audio frame for one or more of said memories into the respective memory. The term "decoded audio frame" relates to an audio frame currently under processing whereas the term "preceding decoded audio frame" relates to an audio frame, which was processed before the audio frame currently under processing. The present invention allows a predictive coding scheme to switch its intern sampling rate without the need to resample the whole buffers for recomputing the states of its filters. By resampling directly and only the necessary memory states, a low complexity is maintained while a seamless transition is still possible. According to a preferred embodiment of the invention the one or more memories comprise an adaptive codebook memory configured to store an adaptive codebook memory state for determining one or more excitation parameters for the decoded audio frame, wherein the memory state resampling device is configured to determine the adaptive codebook state for determining the one or more excitation parameters for the decoded audio frame by resampling a preceding adaptive codebook state for determining of one or more excitation parameters for the preceding decoded audio frame and to store the adaptive codebook state for determining of the one or more excitation parameters for the decoded audio frame into the adaptive codebook memory. The adaptive codebook memory state is, for example, used in CELP devices. For being able to resample the memories, the memory sizes at different sampling rates must be equal in terms of time duration they cover. In other words, if a filter has an order of M at the sampling rate fs_2, the memory updated at the preceding sampling rate fs_1 should cover at least M*(fs_1)/(fs_2) samples. As the memory is usually proportional to the sampling rate in the case for the adaptive codebook, which covers about the last 20ms of the decoded residual signal whatever the sampling rate may be, there is no extra memory management to do. According to a preferred embodiment of the invention the one or more memories comprise a synthesis filter memory configured to store a synthesis filter memory state for determining one or more synthesis filter parameters for the decoded audio frame, wherein the memory state resampling device is configured to determine the synthesis memory state for determining the one or more synthesis filter parameters for the decoded audio frame by resampling a preceding synthesis memory state for determining of one or more synthesis filter parameters for the preceding decoded audio frame and to store the synthesis memory state for determining of the one or more synthesis filter parameters for the decoded audio frame into the synthesis filter memory. The synthesis filter memory state may be a LPC synthesis filter state, which is used, for example, in CELP devices. If the order of the memory is not proportional to the sampling rate, or even constant whatever the sampling rate may be, an extra memory management has to done for being able to cover the largest duration possible. For example, the LPC synthesis state order of AMR-WB+ is always 16. At 2.8 kHz, the smallest sampling rate it covers 1.25ms although it represents only 0.33ms at 48kHz. For being able to resample the buffer at any of the sampling rate between 12.8 and 48kHz, the memory of the LPC synthesis filter state has to be extended from 16 to 60 samples, which represents 1.25 ms at 48kHz. The memory resampling can be then described by the following pseudocode: mem_syn_r_size_old = (int)(1.25*fs_1/1000); mem_syn_r_size_new = (int)(1.25*fs_2 /1000); mem_syn_r+L_SYN_MEM-mem_syn_r_size_new= resamp(mem_syn_r+L_SYN_MEM-mem_syn_r_size_old, mem_syn_r_size_old, mem_syn_r_size_new ); where resamp(x,l,L) outputs the input buffer x resampled from I to L samples. L_SYN_MEM is the largest size in samples that the memory can cover. In our case it is equal to 60 samples for fs_2<=48kHz. At any sampling rate, mem_syn_r has to be updated with the last L_SYN_MEM output samples. For(i=0 ;i