KR20040095205A - A transcoding scheme between celp-based speech codes - Google Patents

Abstract

The present invention relates to a system and method for transcoding a compressed speech bitstream based on CELP from a source codec to a destination codec. The method of the present invention processes the source codec input bitstream, unpacks (1) the CELP parameters from the input CELP bitstream, and extracts the unpacked CELP parameters if there is a difference between the destination codec parameters and the source codec parameters. Interpolate (2). In the method of the present invention, in the case of mapping (4) the CELP from the source codec format to the destination codec format, the parameter mapping scheme can be preset or selected (3) alone. The method includes encoding the CELP parameters for the destination codec and processing the destination CELP bitstream by packing 7 the CELP parameters for the destination codec.

Description

A transcoding method between speech codes based on CPL {A TRANSCODING SCHEME BETWEEN CELP-BASED SPEECH CODES}

Coding is a processing technique that converts a raw signal (audio, image, video, etc.) into a format that can be transmitted or stored. Such coding typically takes a lot of compression, but generally involves significant signal processing techniques. The result of the coding is a bitstream (a sequence of frames) of parameters encoded (coded) according to a given compression format. Compression is accomplished by using perceptual and statistical removal of redundant information using various techniques for modeling the signal. Accordingly, the encoded format is referred to as "compression format" or "parameter space". The decoder receives the compressed bitstream and reconstructs the raw signal. Compression of speech coding typically involves a loss of information.

The conversion process between different compression formats and / or the process of reducing the bit rate of a previously encoded signal is known as transcoding. This may be implemented to maintain bandwidth or connect incompatible client and / or server devices. Transform coding differs from direct compression processing in that a transcoder can only access a compressed signal and not a raw signal.

Transform coding can be implemented using brute force techniques such as tandems that undergo decompression and then recompress. Since a large amount of processing is often required and there may be a delay in recompressing the signal after decompressing the signal, transform coding may be considered in the compression space or parameter space. This conversion coding aims at mapping between compression formats, but keeps mapping in the parameter space wherever possible. This position is where the sophisticated algorithm of "smart" transcoding will work. Although the conversion coding technique has advanced, it is desirable to further improve the change coding technique. Further description of the limitations of the prior art is described in more detail below throughout this specification.

See related applications

This application is incorporated by reference in US Provisional Application No. 60 / 347,270, filed Jan. 8, 2002, No. 60 / 364,403, filed Mar. 12, 2002, No. 60 / 421,446, 60 / 421,449, filed October 25, 2002, and 60 / 421,446, filed October 25, 2002, the provisional application of which is hereby incorporated by reference as if fully set forth herein. It is included in the specification and quoted.

STATEMENT AS TO RIGHT TO S MADE UNDER FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT (In the case of invention invented by Federal Development Fund

- Not applicable

REFERENCE TO A "SEQUENCE LISTING," A TABLE, OR A COMPUTER PROGRAM LISTING APPENDIX SUBMITTED ON A COMPACT DISK (Reference to Attaching a "Sequence List", Table, or List of Computer Programs Submitted to CD)

- Not applicable

The present invention relates generally to information processing techniques. More specifically, the present invention relates to a CELP-based standard from one based on one Code Excited Linear Prediction (CELP), and / or to a different mode within a single standard. Provided are a method and an apparatus for converting. Details of the present invention will be described below in detail through the present specification.

The objects, features and advantages of the invention are considered novel and are specifically disclosed in the appended claims. The present invention will be clearly understood from the following detailed description with reference to the accompanying drawings, along with other objects and advantages with respect to the configuration and operation thereof.

1 is a block diagram of a decoder stage of a typical CELP coder.

2 is a block diagram illustrating an encoder stage of a general CELP coder;

3 shows a mathematical model of a codec as a simple block;

4 shows a mathematical model of a tandem transform codec as a simple block.

5 is a block diagram illustrating a mathematical model of a smart transform codec.

FIG. 6 shows an example of one of the typical devices for transcoding based on CELP. FIG.

FIG. 7 shows an example of one of the typical devices for conversion coding based on CELP. FIG.

8 is a block diagram illustrating general conversion coding between CELP codecs.

9 is a simplified diagram illustrating subframe interpolation for GSM-AMR and G.723.1.

10 is a simplified block diagram illustrating a system configured in accordance with an embodiment of the present invention for transcoding an input CELP bitstream from a source CELP codec into an output CELP bitstream of a destination codec.

11 is a simple block diagram detailing a source codec CELP parameter unpack module.

12 is a simple block diagram showing interpolation of subframes and parameters in sample-by-sample for G.723.1 to GSM-AMR.

FIG. 13 is a simple block diagram illustrating excitation adjusted by source codec LPC coefficients and destination codec coded LPC coefficients. FIG.

14 is a simple block diagram detailing a parameter mapping and tuning module for CELP parameter mapping.

Fig. 15 is a simple block diagram showing details of destination CELP parameter tuning module.

FIG. 16 is a simplified illustration of an embodiment of destination CELP code packing in a frame for GSM-AMR. FIG.

17 illustrates an embodiment of a G.723.1 to GSM-AMR conversion coder.

18 illustrates an embodiment of a SGM-AMR to G.723.1 transform coder.

The present invention provides a technique for processing information. More specifically, the present invention provides a method and apparatus for converting a CELP frame from one CELP (Code Excited Linear Prediction) based standard to another based on another CELP, and / or in a different mode within a single standard. To provide. More detailed description of the present invention will be provided below in detail through the present specification.

As a specific embodiment, the present invention provides an apparatus for converting a CELP frame from a standard based on one CELP to a standard based on another CELP, and / or in a different mode within a single standard. The apparatus includes a bitstream unpacking module that extracts one or more CELP parameters from a source codec. The apparatus of the present invention also includes an interpolator module coupled to the bitstream unpacking module. This interpolator module performs interpolation between the frame size, subframe size and / or sampling rate of different source codec and destination codec. The interpolator module is combined with a mapping module. The mapping module is configured to map one or more CELP parameters of the source codec to one or more CELP parameters of the destination codec. The apparatus of the present invention includes a destination bitstream packing module coupled to the mapping module. The destination bitstream packing module is adapted to construct at least one destination output CELP frame based on at least one CELP parameter from the destination codec. The controller is coupled to at least the destination bitstream packing module, the mapping module, the interpolator module and the bitstream unpacking module. Preferably, the controller is arranged to monitor the operation of one or more modules and to receive commands from one or more external applications. The controller is adapted to provide status information to one or more external applications.

In another specific embodiment, the present invention provides a method of transcoding a compressed speech bitstream based on CELP from a source codec to a destination codec. The method comprises processing a source codec input CELP bitstream to unpack at least one CELP parameter from an input CELP bitstream, and a plurality of destinations having a frame size, subframe size, and / or sampling rate of a destination codec format. Multiple unpacked CELPs from the source codec format to the destination codec format, if one or more differences exist in the nation codec parameter and multiple source codec parameters having a frame size, subframe size or sampling rate of the source codec format. Interpolating one or more of the parameters. The method includes encoding one or more CELP parameters for the destination codec and processing the destination CELP bitstream by at least packing one or more CELP parameters for the destination codec.

In another specific embodiment, the present invention provides a method for processing a compressed voice bitstream based on CELP from a source codec to a destination codec format. The method includes transmitting one control signal of the plurality of control signals from an application processor and selecting one CELP mapping scheme from a plurality of different CELP mapping schemes based on at least control signals from the application. The method includes performing a mapping process using a CELP mapping scheme selected to map one or more CELP parameters from a source codec format to one or more CELP parameters of a destination codec format.

The present invention also provides a system for processing a compressed voice bitstream based on CELP from a source codec to a destination codec format. The system includes one or more memories. Such a memory may include one or more codes for receiving a control signal of one of a plurality of control signals from an application processor. One or more codes may also be included that select one CELP mapping scheme from a plurality of different CELP mapping schemes based on at least control signals from the application. The one or more memories may also include one or more codes that perform mapping processing using a CELP mapping scheme selected to map one or more CELP parameters from a source codec format to one or more CELP parameters of a destination codec format. According to this embodiment, there may be other computer code that performs the functions disclosed in the present specification or in the non-parts that may be used in the present invention.

Many advantages can be taken with the present invention. Depending on the embodiment, one or more of the following advantages may be achieved.

The computational complexity of the conversion coding process can be reduced.

ㆍ Delay can be reduced through the conversion coding process.

• The amount of memory required for conversion coding can be reduced.

Dynamic rate control can be introduced.

Silence frames can be supported through the built-in voice activity detector.

A structure that can use various parameter mapping methods can be provided.

ㆍ To provide general conversion coding structure that can apply codec based on various CELP now and in the future.

The conversion coding technique of the present invention can achieve one or more of these advantages. In a specific embodiment, the conversion encoding apparatus,

A source CELP parameter unpacking module for extracting CELP parameters from an input encoded CELP bitstream;

A CELP parameter interpolator that converts the input source CELP parameters into destination CELP parameters corresponding to the subframe size difference between the source and destination codec; The parametric interpolator is used when the subframe sizes of the source and destination codecs are different.

A destination CELP parameter mapping and tuning engine for converting CELP parameters from the interpolator module to destination CELP codec parameters;

A destination CELP code packer to pack the mapped CELP parameters into a destination CELP code frame;

Improved feature manager to manage optional functions and features in CELP to CELP conversion coding;

A controller for monitoring the entire transcoding process;

Status reporting function providing the status of the conversion coding process.

The source CELP parameter unpacking module is a simplified CELP decoder without formant and post-filters.

The CELP parameter interpolator has a set of interpolators associated with one or more CELP parameters.

The destination CELP parameter mapping and tuning module includes a parameter mapping scheme switching module and a following parameter mapping scheme, that is, a module of CELP parameter direct space mapping, an analysis module in excitation space mapping and a filtered excitation space mapping. At least one of the analysis modules.

The present invention performs transform coding in units of subframes. In other words, when a frame (of source compression information) is received by the transcoding system, the transcoder may initiate an operation on the frame and generate an output subframe. Once a sufficient number of subframes have been created, a frame (of compressed information according to destination format) can be generated and sent to the communication channel if the communication is for the purpose. If storage is the purpose, the generated frames can be stored as needed. If the durations of the frames defined by the source and destination format standards are the same, a single input frame will produce a single output frame, and if the durations are not the same, then either buffer one input frame or Generation will be required. If the subframes have different periods, interpolation between subframe parameters will be required. Thus, the transcoding operation has four operations: (1) unpacking the bitstream, (2) interpolation and subframe buffering of the source CELP parameters, (3) mapping and tuning for the destination CELP parameters, and (4) output. It consists of code packing to create a frame.

Thus, upon receiving a frame, the transcoder unpacks the bitstream to generate a CELP parameter for each of the subframes contained within the frame (block 1 in FIG. 10). Relevant parameters are LPC coefficients, excitations (generated from adaptive and fixed codewords) and pitch lag. In the case of low complexity, the solution for producing good quality is that decoding is required for this and no full synthesis of speech waveforms is necessary. If subframe interpolation is required, subframe interpolation is implemented by the smart interpolation engine (block 2 in FIG. 10).

The subframe may be processed by the destination parameter mapping and tuning module (block 5 of FIG. 10). Short-term LPC filter coefficients are mapped independently of excitation CELP parameters. Simple linear mapping in the LSP pseudo frequency space can be used to generate the LSP coefficients for the destination codec. The CELP parameters here can be mapped in many ways at the expense of computational complexity and given a better quality output. These three mapping schemes are disclosed in this document and become part of the parameter mapping and tuning scheme module (block 4 in FIG. 10).

CELP parameter direct spatial mapping (DSM);

Analysis in the excitation space domain;

Analysis in the filtered excitation space domain.

The selection of the mapping and tuning scheme is made by the mapping and tuning scheme switching module (block 3 in FIG. 10).

Since the three methods cancel the quality to reduce the computational load, it can be used to provide a graceful degradation of quality when the device is overloaded with a large number of simultaneous channels. Thus, the performance of the transform coder can adapt the available resources. Alternatively, the transcoding system can be built using only one way to provide the desired quality and performance. In this case, the mapping and tuning scheme switching module (block 3 in FIG. 10) will not be included.

At this point, if applicable to the destination standard, a voice activity detector (operating in parameter space) may be employed to reduce the outbound bandwidth.

The mapped parameter may be packed into a destination bitstream format frame (block 7 in FIG. 10) and is generated for transmission or storage.

The present invention relates to algorithms and methods used to perform smart transcoding between coding methods and standards based on CELP (Code Excited Linear Prediction). The present invention also relates to transcoding within a single standard for performing rate control (by transcoding to lower the mode or introduce silence frames through the built-in voice activity detector (VAD)).

The entire process of the conversion coding is monitored (controlled) by the control module (Fig. 10 (8)) which sends the command based on the status of the conversion coding and the external command.

In order to apply different transcoding requirements, the apparatus of the present invention provides the ability to add optional features and functions (block 6 in FIG. 10).

Other features and advantages of the present invention will become apparent from the following detailed description with reference to the accompanying drawings, in which like reference characters designate the same or similar elements throughout the figures thereof.

According to the present invention, a technique for processing information is provided. Specifically, the present invention provides a method and apparatus for converting a CELP frame from a standard based on one CELP to a standard based on another CELP, and / or in a different mode within a single standard. A more detailed description of the invention is described below in detail throughout this specification.

The present invention relates to algorithms and methods used to perform smart transcoding between coding methods and standards based on code excited linear prediction (CELP). The focus is on CELP coding methods standardized by organizations such as the International Telecommunication Union (ITU) or the European Telecommunications Standards Institute (ETSI). The present invention also relates to transform coding within a single standard for performing rate control (by transform coding to lower the mode or introduce a silence frame through the built-in voice activity detector (VAD)). It is about.

Speech coding techniques generally include waveform coders (e.g., standards G.722, G.726, and G.722 at ITU) and analysis-by-synthesis (ABS) type coders (e.g., at ITU). G.723.1 and G.729 standards, GSM-AMR standards in ETSI, Enhanced Variable-Rate Codec (EVRC), and optional mode vocoder (TIA) in the Telecommunications Industry Association (TIA) SMV: Selectable Mode Vocoder (Standard). The waveform coder operates in the time domain and is based on a sample-by-sample approach that uses correlation between speech samples. The coder of the analysis type according to the synthesis is a frame unit (typically a frame of 10-30 ms size is used), and a human model is generated by a simple model of a source (glottis) and a filter (vocal tract) forming an output speech spectrum. Try to imitate your voice generation system.

Coders of the analysis type by synthesis have been introduced to provide high quality voice at low bit rates, but many computations are required. Compression technology is a meaningful way to conserve resources at the communication interface.

Mathematically, every speech codec starts with a one-dimensional analog speech signal x _α ( t ), and this speech signal is uniformly sampled and quantized so that the digital domain representation x ( n ) = Ｑ ( xα ( nT )) You get The sampling rate f = 1 / T for the audio signal is usually 8 kHz or 16 kHz, and the sampled signal is usually quantized to a maximum value of 16 bits.

A codec based on CELP can be thought of as an algorithm for mapping between sampled speech x ( n ) and some parameter space θ using a model of speech generation. In other words, the digital voice is encoded and decoded. All algorithms based on CELP operate in speech frame units (which can be further separated into several subframes). In some codecs, voice frames overlap each other. The speech frame may be defined as a vector of speech samples starting at any time n . In other words,

Where L is the length (number of samples) of the audio frame. Note that frame index i relates to the first frame sample n by a linear relationship.

n = for iL non-overlapping frames

n = i ( L - K ) for nested frames

Where K is the number of samples superimposed between the frames.

Compression (lossy coding) processing is a parameterθ _i Essex frame onIs a function that maps theθ _i Raw speech frames fromMap to an approximation of. The speech frame generated by the decoder is not the same as the initially encoded speech frame. The codec is designed to produce an output voice that is perceptually similar to a possible input voice. That is, the encoder must generate a parameter that maximizes some perceptual reference measurement between the frame generated by the decoder and the input speech frame when processing the parameter.

In general, the mapping from input to parameter and from parameter to output must know all previous inputs or parameters. This can be achieved by maintaining a state in the codec S , for example in the construction of an adaptive codebook used by a method based on CELP. The encoder state and decoder state must be kept in sync. This is accomplished only by both sides (encoders and decoders) updating small data, i.e. a state based on parameters. 3 shows a general model of an encoder, a channel and a decoder.

The frame parameter θ used in the CELP-based model is used for short-term prediction of speech signals (and physically related to vocal, mouth, nasal and lip) as well as excitation signals of adaptive and fixed codes. It consists of a linear predictive coefficient (LPC). The adaptation code is used to model long term pitch information in speech. Codes (adaptation and fixation) are associated with codebooks preset for a particular CELP codec. Figure 1 shows a typical CELP decoder in which the adaptive and fixed codebook vectors are scaled independently by gain coefficients, combined and filtered to produce synthesized speech. This voice passes through a post-filter to remove artifacts caused by the model.

The CELP encoding (analysis) process includes, as shown in FIG. 2, preprocessing the speech signal to remove undesirable frequency components, applying a windowing function, and extracting short-term LPC parameters. do. This process is typically implemented using the Levinson-Durbin algorithm. LPC parameters are converted into LSPs (Line Spectral Pairs) to facilitate quantization and subframe interpolation. Next, the voice is back filtered by a short term LPC filter to produce a residual excitation signal. This residual signal is perceptually weighted to improve quality and analyzed to find estimates of speech pitch. In order to determine the optimal pitch, a closed-loop synthesis method is used. If the pitch is found, the adaptive codebook component here is subtracted from the residual signal and the best codeword found. The internal memory of the encoder is updated to reflect changes in the codec state (adaptive codebook, etc.).

The simplest method of transform coding is the Brute force method called tandem transform coding, see FIG. This method performs full decoding of the compressed bit input to generate the synthesized speech. Next, the synthesized speech is encoded against the target target standard. This method not only requires a significant amount of computation when recoding a signal, but also the degradation caused by pre- and post-processing filtering of speech waveforms and the potential caused by the look-ahead conditions of the encoder. Has a delay problem.

A method for "smart" conversion coatings similar to that shown in FIG. 5 is disclosed in the literature. However, these methods still need to reconstruct the speech signal and perform an important task of extracting various CELP such as LPC and pitch. In other words, these methods still operate in the speech signal space. In particular, the excitation signal already optimally matched to the original voice by the far-end encoder (an encoder at the end that produces compressed speech according to the compression format) is used only for the generation of synthesized speech. . This synthesized speech is used to calculate a new optimal excitation. Since the impulse response filtering operation in the closed loop search is necessary, this operation is very expensive. 6 illustrates a method disclosed in US Pat. No. 6,260,009 B1. The reconstructed signal used by the searcher as the target signal is generated from the input excitation parameter and the quantized output formant filter coefficients. Since there is a difference between the quantized formant filter coefficients in the source and destination codecs, the target signal for the searcher deteriorates and the output sound quality from the transform coding is greatly degraded. See FIG. 6. Other problems are described in more detail below throughout this specification.

Another " smart " conversion coding method is shown in FIG. US patent application US2002 / 0077812 A1 is disclosed. This method performs transcoding through the mapping of each CELP parameter that directly ignores the interaction between the CELP parameters. This method is only applicable in special cases that require very limited conditions between the source and destination CELP codecs. For example, this method requires the same subframe size as the algebraic CELP (ACELP) in both the source and destination codecs. Most of the CELP-based conversion codings do not produce good sound quality. This method is only suitable for one of the GSM-AMR modes and does not apply to all GSM-AMR modes.

The method and apparatus of the present invention are described in detail below. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present invention. For example use the case of GSM-AMR and G.723.1. The method described here is general and applies to conversion coding between arbitrary CELP codec pairs. Those skilled in the art will appreciate that other steps, structures, and configurations may be used without departing from the scope of the present invention.

The present invention relates to algorithms and methods used to perform smart transform coding between CELP-based speech coding standards. The invention also relates to transcoding within a single standard for performing rate control (by transcoding to lower the mode or introduce silent frames via the built-in voice activity detector (VAD)). The following describes the present invention in detail.

The present invention performs transform coding in units of subframes. That is, when the transcoding system receives the frame, the transcoder may start the operation of the received subframe and the generation of the output subframe. Once a sufficient number of subframes have been created, a frame can be created. If the durations of the frames defined by the source and destination standards are the same, one input frame will produce one output frame, otherwise the buffering of one of the input frames or the generation of multiple output frames will be required. . If the subframes have different periods, interpolation processing will be required between the subframe parameters. Thus, transcoding has four operations: (1) bitstream unpacking, (2) subframe buffering and interpolation of source CELP parameters, (3) mapping and tuning to destination CELP parameters, and (4) Code packing to generate an output frame (see FIG. 8).

10 is a block diagram showing the principle of a codec conversion coding apparatus based on CELP according to the present invention. This block includes a source bitstream unpacking module, a smart interpolation engine, a parameter mapping and tuning module, an optional improved feature module, a control module and a destination bitstream packing module.

The parameter mapping & tuning module includes a mapping & tuning strategy switching module and a parameter mapping & tuning strategies module.

The conversion coding operation is monitored (controlled) by the control module.

Upon receiving the frame, the transform coder decompresses and side unpacks the bitstream to generate a CELP parameter for each subframe included in the frame. Relevant parameters are LPC coefficients, excitations (generated from adaptive and fixed codewords) and pitch lag.

Note that only decoding for excitation is required, not full synthesis of speech waveforms. This greatly reduces the complexity of the source codec bitstream unpacking. Codebook gains and fixed codewords are also associated with CELP parametric Direct Space Mapping (DSM) transcoding schemes. Implemented when subframe interpolation is required.

The subframe may be processed by the destination parameter mapping and tuning module shown in FIG. 14. The short term LPC filter coefficients are mapped independently of the excitation CELP parameters. Simple linear mapping in the LSP pseudo frequency space can be used to generate the LSP coefficients for the destination codec. More sophisticated non-linear interpolation can also be used. The CELP parameters here can be mapped in a variety of ways, with complex calculations but improved output quality. These three mapping schemes have been disclosed and become part of the parameter mapping and tuning scheme module (block (4) of FIG. 10).

CELP parameter direct spatial mapping (DSM);

Analysis in the excitation space domain;

Analysis in the filtered excitation space domain.

The selection of the mapping and tuning scheme is made in a mapping & tuning strategy switching module (block 3 in FIG. 10).

These three methods are described in detail below. These three methods can be used to provide a gradual degradation of quality when the device is overburdened by a large number of simultaneous channels because the quality is offset to reduce the computational burden. Thus, the performance of the transform coder can adapt the available resources. Alternatively, the transcoding system can be built using only one way to provide the desired quality and performance. In this case, the mapping and tuning scheme switching module (block 3 in FIG. 10) will not be included.

At this point, if applicable to the destination standard, a voice activity detector (operating in parameter space) may be employed to reduce the outbound bandwidth.

The output of the parameter mapping and tuning module is the destination CELP codec code. This output is packed into destination bitstream frames according to the codec CELP frame format. This packing process is necessary to configure the output bits in a format that the destination CELP decoder can understand. For doing the storage, the destination CELP parameters may be stored or packed in an application specific format. The packing process is also changeable if the frame is to be transmitted in accordance with the multimedia protocol, for example a bit scramble process will be implemented in the packing process.

In addition, the apparatus of the present invention provides the ability to add a signal processing function or module as an option in the future.

Subframe interpolation

Subframe interpolation is necessary when subframes for different standards represent different time intervals in the signal region, or when different sampling rates are used. For example, G.723.1 uses frames with a 30ms interval (7.5ms per subframe) and GSM-AMR uses frames with a 20ms interval (5ms per subframe). This is illustrated in FIG. 9. Subframe interpolation is performed on two different types of parameters: (1) sample-based parameters (eg, excitation and codeword vectors), and (2) subframe parameters (eg, LSP coefficients and pitch delay estimates). . The parameters in sample units are mapped by considering their discrete time indices and replicating to appropriate locations in the target subframe. Upsampling or downsampling may be required if different sampling rates are used by different CELP standards. Subframe parameters are interpolated by some interpolation functions to produce a smooth estimate of the parameters in the target subframe. Smart interpolation algorithms can improve speech conversion coding not only by computational performance but more importantly by sound quality. Simple interpolation functions are linear interpolators.

As an example, FIG. 9 shows that three GSM-AMR frames are needed to describe the same spacing of voice signals as two G.723.1 frames. Similarly, three GSM-AMR subframes are needed for every two G.723.1 subframes. As mentioned above, there are two types of parameters. That is, there are subframe-wide parameters (eg, LSP coefficients) and sample-level parameters (eg, low and fixed codewords).

The subframe parameter, denoted θ , is converted linearly by computing the weighted sum of the overlapping subframes, and the parameter in the sample unit, denoted by v [·], is formed by duplicating an appropriate sample. For interpolation from a G.723.1 subframe to a GSM-AMR subframe, the analytic formula is expressed as follows.

Here, i = 0 is the first subframe of the first GSM-ARM frame, i = 4 is the first subframe of the second GSM-AMR frame, and so on. 12 shows this process.

The LSP parameter should be interpolated in the pseudo frequency domain, i.e. f = cos ^-1 ( q ), as a full subframe parameter. This results in a more improved output quality. The other subframe parameters do not need to be converted before interpolation.

The equation is derived from a simple linear interpolator. It is important to note that this equation can be replaced with any suitable interpolation structure, such as splines and sinusoids. In addition, each CELP parameter (LSP coefficient, delay, pitch gain, codeword gain, etc.) may use different interpolation schemes to obtain the best perceptual quality.

LSP Parameter Mapping and Excitation Vector Calibration by LSP Coefficients

Although almost all CELP-based audio codecs use the same method to obtain LPC coefficients, there are still some minor differences. These differences are due to different window sizes and shapes, different LPC interpolation for each subframe, different subframe sizes, different LPC quantization schemes, different lookup tables.

In order to further improve the audio transcoding quality processed through the subframe interpolation method described above, by applying LPC data from the source and destination codec, excitation vectors used as target signals in the transcoding are used. .

The following two methods can be adopted to improve the perceptual quality.

First method: linear transformation of LSP coefficients

The general method for converting between LSP coefficients is linear transformation.

q ' = Aq + b

Where q ' is the destination LSP vector (in the pseudo frequency domain), q is the source (first) LSP vector, A is the linear transformation matrix, and b is the bias term. In the simplest case, A decreases to the unit matrix and b decreases to zero. As an embodiment of the GSM-AMR to G.723.1 conversion coder, the DC bias term used in the GSM-AMR codec is different from that used in G.723.1, and the b term in the equation is used to compensate for this difference.

Method 2: Adjust Excitation Vector by LSP Coefficient

The decoded source excitation vector is synthesized by the source LPC coefficients in each subframe to convert to the speech domain, and filtered using the quantized LP parameters of the destination codec to form the target signal in transform coding. This adjustment is optional and can greatly improve the perceptual sound quality with significant differences in LPC parameters. 13 shows an excitation adjustment scheme.

The parameter mapping and the tuning module (Mapping Parameter & Tuning Module)

Hereinafter, three methods for mapping the CELP excitation parameter will be described. These schemes are represented on the basis of continuous computational complexity and output quality. The key to the present invention is that excitation can be mapped directly without the need to reconstruct the speech signal. This means that important calculations are saved during closed loop codebook searches, since the signal does not need to be filtered by the short term impulse response, as required in the prior art. This mapping process works because the input bitstream already contains the optimal excitation according to the source CELP codec for speech generation. The present invention uses this fact to perform a quick search in the excitation region instead of the speech region.

As mentioned above, each of the three methods for excitation mapping has successfully better performance, whereby the transform coder can be applied to available computing resources.

CELP Parameters Direct Space Mapping

This is the simplest conversion coding technique. The mapping is based on the similarity with physical meaning between source and destination parameters, and directly performs transform coding using analytical equations without any repetition or searching. The advantage of this technique is that it does not require much memory and uses nearly zero MIPS, but it can still produce clear sound despite the degradation of sound quality. Importantly, the CELP parameter direct spatial mapping method of the present invention is different from the prior art apparatus shown in FIG. The method of the present invention is general and applies to all kinds of transform coding based on CELP by different CELP codes, different frame or subframe sizes in source and destination.

Analysis at this spatial region (Excitation Analysis in Space Domain)

This technique is an improvement over previous techniques where adaptive and fixed codebooks are searched, and the gain estimated in the general method defined by the destination CELP standard is not in the speech domain except that it is implemented in the excitation region. Pitch contribution is first determined by local search using the pitch from the input CELP subframe as an initial estimate. Once found, the pitch contribution is subtracted from the fixed codebook, which is determined by optimally matching the residual. Compared to the tandem scheme, the open loop pitch estimate does not need to be calculated from the autocorrelation method used by the CELP standard, but instead has the advantage that it can be determined from the pitch delay of the decoded CELP subframe. In addition, since such a search is performed in the excitation region rather than the speech region, impulse response filtering during pitch and codebook search is not necessary. Thereby, a significant amount of calculation can be reduced without canceling output quality.

Analysis of the filtered spatial domain (Space Domain Analysis in Filtered Excitation)

In this case, the LP parameters are still mapped directly from the source codec to the destination codec, and the decoded pitch delay is used as an open loop pitch estimate for the destination codec. Closed loop pitch search is subsequently performed in the excitation region. However, fixed codebook searches are performed in the filtered excitation space region. The choice of filter type will depend on the desired quality and complexity requirements, depending on whether the target vector is transformed into this region for one or two searches.

As various filters, it is possible to use filters to compensate for differences between excitation characteristics in the source and destination codecs, including lowpass filters to smooth out irregularities, and filters to reinforce perceptually important signal characteristics. The advantage is that, unlike the calculation of the target signal in standard encoding, it uses a weighted LP synthesis filter, whose parameters [frequency, frequency emphasis / de-emphasis, phase] are complete. Thus, this technique enables tuning to improve the quality for conversion coding between specific codec pairs and to offset the quality for reducing complexity.

Silent frame transform coding and generating (Silence Frame Transcoding and Generation)

Some CELP-based standards implement Voice Activity Detectors (VAD) that enable discontinuous transmission (DTX) and comfort noise generation (CNG) during periods of no voice. When silent frames are not generated by the source codec, conversion coding between these frames is also required for generation of silent frames for the destination codec. In general, a frame consists of parameters for generating appropriate comfort noise at the decoder. These parameters can be transform coded using a simple algebraic method.

Embodiment of the present invention

Embodiments of the present invention for the G.723.1 and GSM-AMR speech coding standards are described below. The present invention is not limited to these standards. The present invention applies to all coding standards based on CELP. Those skilled in the art will understand how to apply these methods for transform coding between different CELP-based coding standards. Before describing the preferred embodiment, a brief description of the GSM-ARM and G.723.1 codecs is given first.

GSM-AMR codec

The GSM-AMR codec uses eight source codecs with bit rates of 12.2, 10.2, 7.95, 7.40, 6.70, 5.90, 5.15 and 4.75 kilobits per second (kbit / s).

This codec is based on code excited linear prediction (CELP) coding model. Tenth order linear prediction (LP) or short term synthesis filters are used. The long term or pitch synthesis filter is implemented using a so-called adaptive codebook scheme.

In the CELP speech synthesis model, the excitation signal at the input of the short-term LP synthesis filter is constructed by adding two excitation vectors from an adaptive and fixed (innovative) codebook. This speech is synthesized by feeding two appropriately selected vectors from these codebooks through a short-term synthesis filter. The optimal excitation sequence in the codebook is selected using an analytical search procedure by synthesis where the error between the original synthesized speech is minimized in accordance with the perceptually weighted distortion measurement. Perceptual weighted filters used in synthetic analytical search techniques use unquantized LP parameters.

The coder operates on a 20 ms speech frame corresponding to 160 samples at a sampling frequency of 8000 samples / s. For each of the 160 speech samples, the speech signal is analyzed to extract the parameters of the CELP model (LP filter coefficients, indexes and gains of the adaptive and fixed codebooks). These parameters are encoded and transmitted. At the decoder, these parameters are decoded and the speech is synthesized by filtering the reconstructed excitation signal through the LP synthesis filter.

LP analysis is performed twice per frame for 12.2 kilobits per second (12.2 kbit / s) mode and once for the other mode. For 12.2 kbit / s mode, two sets of LP parameters are converted into line spectrum pairs (LSP) and jointly quantized using split matrix quantization (SMQ) with 38 bits. For other modes, a single set of LP parameters is converted into line spectrum pairs (LSP) and vector quantized using split vector quantization (SVQ).

The speech frame is divided into four subframes (40 samples) of 5 ms each. The adaptive and fixed codebook parameters are transmitted per subframe. Depending on the subframe, quantized LP parameters and unquantized LP parameters or interpolated versions thereof are used. The open loop pitch delay is estimated every other subframe based on the perceptually weighted speech signal (except for 5.15 and 4.75 kbit / s mode, which is done once per frame).

The following operation is repeated for each subframe.

The target signal is calculated by filtering the LP residual through a weighted synthesis filter in the initial state of the updated filter by filtering the error between the LP residual and the excitation (this is a weighted synthesis from the weighted speech signal). Equivalent to a common way of subtracting the filter's zero input response).

The impulse response of the weighted synthesis filter is calculated.

Closed loop pitch analysis is performed by searching around the open loop pitch delay using the target and impulse response (to find the pitch delay and gain). A fractional pitch with 1 / 6th or 1 / 3th sample resolution of the sample resolution (depending on the mode) is used.

The target signal is updated by removing the adaptive codebook contribution (filtered adaptive codevector), and this new target is used for fixed algebraic codebook search (to find the optimal innovation codeword).

The gains of the adaptive and fixed codebooks are scalar quantized to 4 bits and 5 bits respectively or vector quantized to 6-7 bits (with a moving average (MA) prediction applied to the fixed codebook gains).

Finally, the filter memory is updated to find the target signal in the next subframe (using the determined excitation signal).

In 20 ms speech frames, bit allocations of 95, 103, 118, 134, 148, 159, 204, or 244 bits correspond to bit rates of 4.75, 5.15, 5.90, 6.70, 7.40, 7.95, 10.2 or 12.2 kbps, respectively. .

G.723.1 codec

The G.723.1 codec has two associated bit rates, 5.3 kbps and 6.3 kbps. These two bit rates are an integral part of the encoder and decoder. It is possible to swap two bit rates for any 30 ms frame boundary.

The coder is based on the principle of analysis by synthesis of linear prediction and minimizes the perceptually weighted error signal. The encoder works for each block (frame) of 240 samples. That is, 30 ms at an 8 kHz sampling rate. Each block is first high-pass filtered to remove the DC component and separated into four subframes with 60 samples each. For each subframe, a tenth order linear prediction coder (LPC) filter is calculated using the unprocessed input signal. The LPC filter for the final subframe is quantized using a Predictive Split Vector Quantizer (PSVQ). Unquantized LPC coefficients are used to construct a short-term perceptual weighted filter, which is used to filter the entire frame and obtain a perceptually weighted speech signal.

For every two subframes (120 samples), the open loop pitch period L _OL is calculated using the weighted speech signal. This pitch estimate is performed on 120 sample blocks. The pitch period is searched in the range of 18 to 142 samples.

From this point, speech is processed for 60 samples per subframe unit.

Using the previously calculated estimated pitch period, a high frequency noise shaping filter is constructed. To produce an impulse response, a combination of LPC synthesis filter, formant perceptual weighting filter and high frequency noise shaping filter are used. The impulse response is used for future calculations.

Using the pitch period estimate L _OL and the impulse response, a closed loop pitch predictor is calculated. A fifth order pitch predictor is used. The pitch period is calculated as a small difference near the open loop pitch estimate. The contribution of the pitch predictor is subtracted from the initial target vector. The pitch period and the difference value are both sent to the decoder.

Finally, an approximation of the periodic component of the excitation is obtained. For high bitrates, multi-pulse maximum likelihood quantization (MP-MLQ) excitation is used, and for low bitrates, algebraic codebook excitation (ACELP) is used.

Embodiment 1-GSM-AMR vs. G.723.1

17 is a block diagram showing a conversion coder from GSM-AMR to G.723.1 according to the first embodiment of the present invention. The GSM-AMR bitstream consists of 20 ms frames with a length from 244 bits (31 bytes) in the fastest mode 12.2 kbps to the lowest mode 4.75 kbps codec. There are eight modes in all. Each of the eight GSM-AMR modes of operation produces a different bitstream. Since a G.723.1 frame with a duration of 30ms consists of 1 and 1/2 GSM-AMR frames, two GSM-AMR frames are required to generate a single G.723.1 frame. When the third GSM-AMR frame arrives, the next G.723.1 frame may be generated. Thus, two G.723.1 frames are generated each time three GSM-AMR frames are processed.

The ten LSP parameters used by the short-term filter in the GSM-AMR speech generation model are encoded using the same technique, but the bitstream formats for the different modes of operation are different. The algorithm for reconstructing LSP parameters is in the GSM-AMR standard document.

Once the short-term filter parameters have been generated for each subframe, an excitation vector needs to be formed by combining the adaptive codeword and the fixed (logarithmic) codeword. The adaptive codeword is constructed using 60-tap interpolation based on the 1 / 6th or 1 / 3th resolution pitch delay parameter. Next, the fixed codewords consist of those defined by excitation and standards formed as follows.

Where x is here, v is interpolated adaptive codeword, c is fixed codevector, Wow Denotes the adaptive and fixed code gains, respectively. Excitation is used to update the memory state of the GSM-AMR unpacker by the G.723.1 bitstream packer for mapping.

An adaptive codeword is determined for each subframe by forming a linear combination of excitation vectors and finding the best match for the target excitation signal x [] constructed by the GSM-AMR unpacker. This combination is the weighted sum of previous excitations at five consecutive delays. This can best be explained by the following equation:

Where v [] is the reconstructed adaptive codeword, u [] is the previous excitation buffer, L is the (integer) pitch delay between 18 and 143 (determined by the GSM-AMR unpacking module), and β _j Is the delay weight that determines the gain and delay phase. The vector table of β _j values is searched to optimize the matching between the adaptive codeword v [] and the excitation vector x [] .

If an adaptive codebook component of excitation is found, this component is subtracted from the excitation value, leaving a residual for encoding by the fixed codebook. The residual signal for each subframe is calculated as follows.

Where x ₂ [] is a target for fixed codebook search, x [] is an excitation value derived from GSM-AMR unpacking, and v [] is an adaptive codeword (interpolated and scaled).

The fixed codebook is different in the high rate mode and the low rate mode of the G.723.1 codec. The fast mode uses an MP-MLQ codebook at any position, allowing six pulses per subframe for even subframes and five pulses per subframe for odd subframes. The low speed mode uses an algebraic codebook (ACELP) that allows four pulses per subframe at restricted locations. Both codebooks use a grid flag to indicate whether a transitioning codeword should be shifted by one position. These codebooks are retrieved by the method defined in the standard, with the exception of impulse response filters. This is because such a search is performed in the excitation region, not in the speech region.

The (nonvolatile) memory for the codec needs to be updated when the processing for each subframe is completed. This update is accomplished by shifting the previous excitation buffer u [] by 60 samples, discarding the oldest sample and duplicating the excitation from the current subframe to the top 60 samples of the buffer.

Where index n is the first sample of the current subframe and other parameters are defined above.

All mapped parameters are coded into the output G.723.1 bitstream, and the system is ready to process the next frame.

Second Embodiment-G.723.1 vs. GSM-AMR

18 is a block diagram showing a G.723.1 conversion coder for GSM-AMR according to a second embodiment of the present invention. The G.723.1 bitstream consists of frames having a length of 192 bits (24 bytes) for the high speed (6.3 kbps) codec or 160 bits (20 bytes) for the low speed (5.3 kbps) codec. Frames have a very similar structure and differ only in the representation of fixed codebook parameters.

The ten LSP parameters used to model the short-term vocal tract filter are coded in the same way for both high speed and low speed, and can extract bits 2 to 25 of the G.723.1 frame. Only the LSP of the fourth subframe is encoded and used to regenerate LSPs for three subframes having different interpolation between frames. The encoding uses three lookup tables and uses the reconstructed LSP vectors by combining three subvectors derived from these tables. Each table has 256 vector entries. That is, the first two tables have three element subvectors and the last table has four element subvectors. Combining these results in a 10-element LSP vector.

The adaptive codeword is constructed for each subframe by combining the previous excitation vector. This combination is the weighted sum of the previous excitations at five consecutive delays. The following equation best explains this.

Where v [] is the reconstructed adaptive codeword, u [] is the previous excitation buffer, L is the (integer) pitch delay between 18 and 143, and β _j is the delay weight determined by the pitch gain parameter.

The delay parameter L is extracted directly from the bitstream. The first and third subframes use the full dynamic range of the delay, while the second and fourth subframes encode the delay as an offset from the previous subframe. The delay weighting parameter β _j is determined by table lookup. As a result of adaptive codeword unpacking, an approximation to the partial pitch delay and its associated gain can be determined by calculating the following equation.

Fixed codebooks are different in the fast and slow modes of the G.723.1 codec. The fast mode uses an MP-MLQ codebook that allows six pulses per subframe for even subframes and five pulses per subframe for odd subframes at any location. The low speed mode uses an algebraic codebook (ACELP) that allows four pulses per subframe at restricted locations. Both codebooks use a grid flag to indicate whether a transitioning codeword should be shifted by one position. Algorithms for generating codewords from coded bitstreams are in the GSM-AMR standard document.

The (nonvolatile) memory for the codec needs to be updated when the processing for each subframe is completed. This update is accomplished by shifting the previous excitation buffer u [] by 60 samples (ie 1 subframe), discarding the oldest sample and duplicating the excitation from the current subframe to the top 60 samples of the buffer.

Here, the index n is set relative to the first sample of the current subframe, and other parameters are defined above.

The GSM-AMR parameter mapping portion of the transform coder takes the interpolated CELP parameters as described above and uses these parameters as a basis for searching the GSM-AMR parameter space. The LSP parameter is simply encoded upon reception, while the other parameters, excitation and pitch delay, are used as estimates for local search in the GSM-AMR space. The following shows the main operations that need to be done for each subframe to complete the transform coding.

The adaptive codeword is formed by searching the previous excitation vector up to a maximum delay of 143 for optimal matching within the target excitation. Target excitation is determined from interpolated subframes. The previous excitation can be interpolated by 1 / 6th or 1 / 3rd intervals depending on the mode. The optimum delay is determined by searching a small range of pitch delays determined from the G.723.1 unpacking module. This range is searched to find the optimal integer delay and separated to determine the fractional part of the delay. The process uses a 24-tap interpolation filter to perform partial search. The first and third subframes are processed differently from the second and fourth subframes. The interpolated adaptive codeword v [] is formed as follows.

Where u [] is the previous excitation buffer, L is the (integer) pitch delay, t is the partial pitch delay at 1 / 6th resolution, and b ₆₀ is a 60 tap interpolation filter.

The pitch gain is calculated and quantized to be encoded and transmitted to the decoder and also for the calculation of the fixed codebook target vector. All modes calculate the pitch gain in the same way for each subframe.

Where g _p is the unquantized pitch gain, x is the target for adaptive codebook search, and v is the (interpolated) adaptive codeword vector. The 12.2 kbps and 7.95 kbps modes independently quantize the adaptive and fixed codebook gains, while the other mode uses joint quantization of the fixed and adaptive gains.

Once the adaptive codebook component is determined, this component is subtracted from the excitation value, leaving a residual for encoding by the fixed codebook. The residual signal for each subframe is calculated as follows.

Here, x ₂ [] is a target for fixed codebook search, x [] is a target for adaptive codebook search, Is the quantized pitch gain and v [] is the (interpolated) adaptive codeword.

The fixed codebook search is made to find the best match for the residual signal after the adaptive codebook component is removed. This is important for unvoiced speech and for priming adaptive codebooks. Codebook searches used for transcoding may be simpler than those used in the codec after a large amount of initial speech analysis has already been made. In addition, since the signal on which the codebook search is performed is an excitation signal reconstructed in place of the synthesized speech, it has a structure capable of fixed book coding.

The gain for the fixed codebook is quantized using moving average prediction based on the energy of the previous four subframes. The correlation coefficient between actual and predicted gain is quantized (via table lookup) and sent to the decoder. The exact details are disclosed in the GSM-AMR standard literature.

The (nonvolatile) memory for the codec needs to be updated when processing for each subframe is completed. This update is accomplished by shifting the previous excitation buffer u [] by 40 samples (ie 1 subframe), discarding the oldest sample and duplicating the excitation from the current subframe to the top 40 samples of the buffer.

Here, the index n is set relative to the first sample of the current subframe, and other parameters are defined above.

While what is considered to be an embodiment of the invention has been illustrated and disclosed, those skilled in the art will recognize that various modifications and equivalents may be made without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation to the teachings of the present invention without departing from the essential inventive concept disclosed herein.

Claims (46)

A device for converting a CELP frame from a standard based on one CELP to a standard based on another CELP, and / or in a different mode within a single standard, A bitstream unpacking module that extracts one or more CELP parameters from the source codec; An interpolator module coupled to the bitstream unpacking module to interpolate between different frame sizes, subframe sizes, and / or sampling rates of the source and destination codecs; A mapping module coupled to the interpolator module for mapping one or more CELP parameters from the source codec to one or more CELP parameters of the destination codec; A destination bitstream packing module coupled to the mapping module to form at least one destination output CELP frame based on at least one CELP parameter from the destination codec; Coupled to at least the destination bitstream packing module, the mapping module, the interpolator module, and the bitstream unpacking module to monitor operation of the one or more modules, receive instructions from one or more external applications, And a controller for providing status information to an external application. The apparatus of claim 1, wherein the controller is a single or multiple controller. The apparatus of claim 1, wherein the mapping module and the destination bitstream packing module are in the same module. The apparatus of claim 1, wherein the mapping module is a single or multiple module. The apparatus of claim 1, wherein the interpolator module is a single or multiple module. The method of claim 1, wherein the bitstream unpacking module, A bitstream processor for extracting information from a source CELP codec input frame into a first format of one or more CELP parameters; An LSP decoding module coupled to the bitstream processor to output one or more LSP coefficients using at least the information from the source CELP codec input frame; A decoding module coupled to the bitstream processor for decoding the information to output a pitch delay parameter and a pitch gain parameter from the source CELP codec input frame; A fixed codebook decoding module coupled to the bitstream processor for decoding the information to output a fixed codebook vector; An adaptive codeword decoding module coupled to the bitstream processor for decoding the information to output an adaptive codebook contribution vector; And an excitation generator coupled to the fixed codebook decoding module and the adaptive codeword decoding module to output an excitation vector using at least the fixed codebook vector and the adaptive codebook vector. The method of claim 1, wherein the interpolator module, An LSP processor for converting one or more LSP coefficients of the source codec into one or more LSP coefficients of the destination codec when the source codec and the destination codec have different subframe sizes; An adaptive codebook processor for converting a pitch delay and a pitch gain of the source codec into a pitch delay and a pitch gain of the destination codec when the source codec and the destination codec have different subframe sizes; And a CELP parameter buffer for holding one or more CELP parameters that need to be buffered for interpolation when the source codec and the destination codec have different subframe sizes. The method of claim 1, wherein the parameter mapping and tuning module, A parameter mapping and tuning scheme switching module for selecting a CELP parameter mapping scheme based on a plurality of schemes; And a parameter matching and tuning method module for outputting the at least one destination CELP parameter. The method of claim 8, wherein the plurality of schemes, A CELP parameter direct spatial mapping module; A filtered excitation space domain analysis module; CELP frame conversion device comprising the analysis in the excitation space domain module. The method of claim 8, wherein the parameter mapping and tuning method module, An LSP coefficient converter for encoding the destination LSP coefficients; And a CELP excitation mapping unit having CELP excitation parameters including pitch delay, gain and excitation vector from interpolation to obtain coded CELP excitation parameters. The method of claim 10, wherein the CELP excitation mapping unit, A CELP parameter direct spatial mapping module for generating coded destination CELP parameters using analysis without any repetition; An analysis module of excitation space domain mapping for generating a destination CELP parameter encoded by the search in the excitation space domain; And an analysis module of the filtered excitation space domain for generating an encoded destination CELP parameter by searching for an adaptive closed loop in the excitation space and a fixed codebook in the filtered excitation space. 2. The destination bitstream packing module of claim 1, wherein the destination bitstream packing module comprises a plurality of frame packing facilities, each facility being applicable to a preselected application from a plurality of applications for a selected destination CELP coder. The CELP coder is one of a plurality of CELP coders including the destination CELP coder. The method of claim 1, wherein the controller, A control unit for receiving an external command and controlling each signal processing module; CELP frame conversion apparatus characterized in that it comprises a state unit for transmitting the conversion coding information, such as frame, count, error delay to the outside on request. 10. The apparatus of claim 1, wherein the interpolator module can be selected from linear or nonlinear interpolation. The method of claim 7, wherein the CELP parameter buffer, An excitation vector buffer that stores a reconstructed excitation vector waiting for mapping in the next subframe or frame; An LSP coefficient buffer for storing the LSP coefficients waiting for matching processing in the next subframe or frame before or after interpolation; And a CELP parameter buffer for storing a pitch delay, a pitch gain, a codebook gain, and an index before or after interpolation waiting for a mapping process in a next subframe or frame. A method of transcoding a compressed speech bitstream based on CELP from a source codec to a destination codec. Processing the source codec input CELP bitstream to unpack at least one CELP parameter from the input CELP bitstream; One or more of a plurality of destination codec parameters having a frame size, subframe size, and / or sampling rate of the destination codec format, and a plurality of source codec parameters having a frame size, subframe size, or sampling rate of the source codec format. Interpolating one or more of a plurality of unpacked CELP parameters from the source codec format to the destination codec format if there is a difference in one or more; Encoding at least one CELP parameter for the destination codec; Processing a destination CELP bitstream by at least packing one or more CELP parameters for the destination codec. The method of claim 16, wherein processing the source codec input comprises: Converting an input bitstream frame into information associated with one or more of the CELP parameters; Decoding the information into one or more CELP parameters; Reconstructing the excitation vector based on at least one of the CELP parameters; And converting the CELP parameter to an interpolator. The method of claim 16, wherein the interpolation process, Interpolating one or more of the LSP coefficients from the source codec to one or more LSP coefficients for the destination codec; Interpolating other CELP parameters other than the LSP coefficients from the source codec to other CELP parameters for the destination codec; And if the excitation vector does not require adjustment, transmitting the excitation vector of the source to an encoding processor. 19. The method of claim 18, further comprising transforming one or more LSP coefficients using a linear transform process. 19. The method of claim 18, further comprising: converting the source codec excitation vector into a synthesized speech vector using at least one decoded source LPC coefficient; Quantizing the destination LPC coefficients; Converting the synthesized speech vector back into a adjusted excitation vector using at least the quantized destination LPC coefficients; And transmitting the adjusted excitation vector to another processor. The method of claim 16, wherein the encoding step, Quantizing the destination LPC coefficients; Parameter mapping and tuning method comprising the step of selecting one of the CELP mapping methods, ie, CELP parameter direct spatial mapping, analysis in the excitation space domain and analysis in the filtered excitation space domain according to a control signal from the switching module. A conversion coding method characterized by the above-mentioned. The method of claim 21, wherein the CELP parameter direct spatial mapping operation, Encoding a pitch delay from the interpolated pitch delay parameter; Encoding a pitch gain from the interpolated pitch gain parameter; Encoding the index of the fixed codebook from the analysis format; And encoding the gain of the fixed codebook gain parameter. The method of claim 21, wherein the analyzing operation in the excitation space domain mapping comprises: Selecting a pitch delay as an initial value from the interpolated pitch delay parameter; Searching for a pitch delay in the closed loop of the excitation space; Searching for a pitch gain in the excitation space; Constructing a target signal for fixed codebook search; Retrieving a fixed codebook index in the space here; Retrieving a fixed codebook gain in the excitation space; And updating the previous excitation vector. The method of claim 21, wherein the analyzing operation in the filtered excitation space domain mapping comprises: Selecting a pitch delay as an initial value from the interpolated pitch delay parameter; Searching for a pitch delay in the closed loop of the excitation space; Searching for a pitch gain in the excitation space; Constructing a target signal for fixed codebook search; Retrieving a fixed codebook index in the filtered excitation space; Searching for a fixed codebook gain in the filtered excitation space; And updating the previous excitation vector. 22. The method of claim 21, wherein the selection is not limited to the three schemes, but a combination of the three schemes can be selected as a new mapping scheme. 2. The CELP frame of claim 1, further comprising a silent frame transform coding unit capable of performing a quick conversion of the silent frame from one speech coding standard to another, and mapping a comfort noise parameter. Converter. The method of claim 1, wherein the parameter mapping and tuning module includes a voice activity detector for generating a silence frame, wherein the voice activity detector determines voice / silence based on parameters in a CELP space. CELP frame converter. 10. The apparatus of claim 1, further comprising a system for modifying the excitation mapping scheme used, applying it to available computing resources and allowing for gradual degradation of the burden. Wherein the excitation mapping is performed without returning back to the speech signal region. A method of processing a compressed voice bitstream based on CELP from a source codec to a destination codec format. Sending one control signal of the plurality of control signals from the application processor; Selecting one CELP mapping scheme from a plurality of different CELP mapping schemes based on at least a control signal from the application; And performing a mapping process using the selected CELP mapping scheme to map one or more CELP parameters from a source codec format to one or more CELP parameters of a destination codec format. 31. The method of claim 30, wherein the plurality of CELP mapping schemes include CELP parameter direct spatial mapping, analysis in an excitation space domain, or analysis in a filtered excitation space domain. 31. The method of claim 30, wherein the step of selecting one CELP mapping scheme is for a predetermined application during setup or configuration processing. 31. The method of claim 30, further comprising receiving the control signal at a switching module coupled to each of the plurality of mapping schemes. 31. The method of claim 30, wherein the control signal is provided based on a computing resource characteristic of the selected CELP mapping scheme. 31. The method of claim 30, wherein at least one of the plurality of mapping schemes is provided to a library of memory. 32. The method of claim 31, further comprising: encoding one or more CELP parameters for the destination codec; Processing a destination CELP bitstream by at least packing one or more CELP parameters for the destination codec. 37. The method of claim 36, further comprising transmitting a packed destination CELP bitstream to the destination codec. A system for processing a compressed voice bitstream based on CELP from a source codec to a destination codec format. One or more codes for receiving a control signal of one of the plurality of control signals from an application processor; One or more codes for selecting one CELP mapping scheme from a plurality of different CELP mapping schemes based on at least control signals from the application; And one or more codes that perform mapping processing using the selected CELP mapping scheme to map one or more CELP parameters from a source codec format to one or more CELP parameters of a destination codec format. The method of claim 38, wherein the plurality of CELP mapping schemes comprise one or more codes for CELP parameter direct spatial mapping, one or more codes for analysis in an excitation space domain, or one or more codes for analysis in a filtered excitation space domain. System comprising a. 39. The system of claim 38, wherein the selected CELP mapping scheme is for a predetermined application. 39. The system of claim 38, wherein the one or more codes for receiving the control signal are provided to a method switching module, wherein the method switching module is coupled to each of a plurality of mapping methods. 39. The system of claim 38, wherein the control signal is provided based on a computing resource characteristic of the selected CELP mapping scheme. 39. The system of claim 38, wherein one or more codes for the plurality of mapping schemes are provided to a library of memory. 44. The apparatus of claim 43, further comprising: one or more codes for encoding one or more CELP parameters for the destination codec; And at least one code for processing a destination CELP bitstream by at least packing one or more CELP parameters for the destination codec. 45. The system of claim 44, further comprising one or more codes for transmitting the destination CELP bitstream to the destination codec. 45. The system of claim 44, further comprising one or more codes for transmitting the destination CELP bitstream to a storage location.