AI voice software data preprocessing system

Abstract

The invention belongs to the field of voice processing, and discloses an AI voice software data preprocessing system, which comprises a pre-emphasis module used for performing pre-emphasis on a voice signal to be emphasis to obtain a voice signal subjected to pre-emphasis; the pre-emphasis module comprises a calculation unit and a pre-emphasis unit; the calculation unit is used for calculating a transient identification value transicoeft of a sampling point s (t) of which the sampling moment is t in the voice signal to be emphasized; the pre-emphasis unit is used for carrying out pre-emphasis processing on s (t) based on the transient identification value, and the pre-emphasis processing comprises the following steps: judging whether the transient identification value is greater than a set comparison value of the transient identification value or not; if not, performing pre-emphasis processing on the s (t) by adopting a pre-emphasis formula with a fixed pre-emphasis coefficient; and if yes, pre-emphasis processing is carried out on the s (t) by adopting a self-adaptive pre-emphasis coefficient. According to the invention, the pre-emphasis effect and the pre-emphasis efficiency are well balanced. When further recognition is carried out based on a voice signal after pre-emphasis, the accuracy of recognized characters can be improved.

Description

Technical Field

[0001] The present invention relates to the field of speech processing, and in particular to an AI speech software data preprocessing system. Background Art

[0002] With the development of artificial intelligence technology, there are more and more AI voice software like chat GPT. These software can provide more diverse and humanized feedback based on the interactive information input by users. When users input interactive information, there are generally two ways: text input and voice input. Text input refers to inputting interactive information through a physical keyboard or a virtual keyboard, while voice input is to obtain the user's speaking voice and then convert the voice into text.

[0003] The process of voice input generally includes steps such as preprocessing of voice signals, feature extraction, and model recognition. In the prior art, in the preprocessing stage, it is usually necessary to perform pre-emphasis, framing, windowing, and other steps on the voice signal. However, the pre-emphasis coefficients generally used are constant values, which easily leads to the interference of amplified transient noise on the sound signal after pre-emphasis, resulting in the accuracy of the text recognized based on the voice signal after this pre-emphasis method is not high enough. Transient noise refers to noise that appears briefly in time. Summary of the invention

[0004] The purpose of the present invention is to disclose an AI voice software data preprocessing system to solve the technical problem of how to improve the accuracy of recognized text when the AI ​​voice software obtains interactive information input by the user through voice recognition.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] The present invention provides an AI voice software data preprocessing system, comprising a pre-emphasis module, the pre-emphasis module is used to pre-emphasize a voice signal to be emphasized, and obtain a pre-emphasized voice signal;

[0007] The pre-emphasis module includes a calculation unit and a pre-emphasis unit;

[0008] The calculation unit is used to calculate the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized. t ;

[0009] The pre-emphasis unit is used to perform pre-emphasis processing on s(t) based on the transient recognition value, including:

[0010] Determine whether the transient recognition value is greater than the set transient recognition value comparison value;

[0011] If not, then s(t) is pre-emphasized using a pre-emphasis formula with a fixed pre-emphasis coefficient;

[0012] If yes, then s(t) is pre-emphasized using the following method, including:

[0013] p(t)=sq(t)-δ(t)×sq(t-1)

[0014] p(t) represents the result of pre-emphasis on s(t), sq(t) and sq(t-1) represent the amplitudes of the sampling points at sampling times t and t-1 respectively, and δ(t) represents the adaptive pre-emphasis coefficient; ut represents the set of sampling moments in the time interval [tT, t], T represents the set duration, and ampl i and amplifier t Respectively represent the normalized values ​​of the amplitude of the sampling point at sampling time i and t, max The value of ampl is 1, max It represents the comparison value of the normalized value of the set amplitude, φ(i) represents the pre-emphasis coefficient sampled when pre-emphasis is performed on the sampling point at sampling time i, and β represents the proportional value.

[0015] Preferably, the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized is calculated. t ,include:

[0016]

[0017] α 1 represents the amplitude weight, α 2 represents the persistence weight, ampl t Represents the normalized value of the amplitude of s(t), ampl pre Indicates the maximum value of the normalized amplitude of the sampling points in the time interval [tT, t+T] at the sampling time. N indicates the total number of sampling points in the time interval [tT, t+T] at the sampling time. Nh t It indicates the total number of sampling points whose normalized amplitude value is greater than z times the absolute value of the average amplitude of the speech signal to be emphasized among all sampling points whose sampling moments are within the time interval [tT, t+T]; T indicates the set duration.

[0018] Preferably, the set duration is 10 ms.

[0019] Preferably, the pre-emphasis formula with a fixed pre-emphasis coefficient includes:

[0020]

[0021] Indicates the set pre-emphasis coefficient.

[0022] Preferably, the pre-emphasis coefficient is set to 0.91.

[0023] Preferably, the amplitude weight is 0.6 and the persistence weight is 0.4;

[0024] The contrast value of the transient recognition value is set to 0.7.

[0025] Preferably, the normalized value of the amplitude is calculated using the following formula:

[0026]

[0027] ampl t represents the normalized value of the amplitude of the sampling point at sampling time t, sq(t) represents the amplitude of the sampling point at sampling time t, sq min and sq max They respectively represent the minimum and maximum amplitudes of all sampling points in the speech signal to be emphasized.

[0028] Preferably, it also includes a microphone module;

[0029] The microphone module is used to convert the user's voice into a voice signal to be emphasized when the user needs to input interactive information by voice input.

[0030] Preferably, it also includes a framing module;

[0031] The framing module is used to frame the pre-emphasized speech signal to obtain multiple speech frames.

[0032] Preferably, it also includes a windowing module;

[0033] The framing module is used to perform windowing processing on the speech frames to obtain multiple speech frames that have undergone windowing processing.

[0034] Beneficial effects:

[0035] Different from the existing speech pre-emphasis method, the present invention does not use a fixed pre-emphasis coefficient when pre-emphasis is performed on the amplitude of each sampling point, because the fixed pre-emphasis coefficient cannot cope with the influence of transient noise well, because pre-emphasis is to subtract the amplitude of the previous sampling point from the amplitude of the current sampling point, which easily leads to the ratio between the amplitude of the transient noise and the amplitude between the sampling points of the speech signal without noise after pre-emphasis, compared with before pre-emphasis, the ratio is enlarged, thereby obtaining a noise with a relatively larger ratio between the amplitudes. Therefore, the present invention calculates the transient recognition value, so that when the probability that the sampling point is located at the transient noise is greater, the more inclined to use the formula with a non-fixed pre-emphasis coefficient to pre-emphasize the amplitude of the sampling point, when the probability that the sampling point is located at the transient noise is smaller, the more inclined to use the formula with a fixed pre-emphasis coefficient to pre-emphasize the amplitude of the sampling point, avoiding the use of a fixed pre-emphasis coefficient for all sampling points for pre-emphasis, and achieving a better balance between the effect of pre-emphasis and the efficiency of pre-emphasis. It is possible to improve the accuracy of recognized text when further recognition is performed based on the pre-emphasized speech signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0037] Figure 1 A schematic diagram of an AI voice software data preprocessing system of the present invention.

[0038] Figure 2 Another schematic diagram of an AI voice software data preprocessing system of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The present invention provides an AI voice software data preprocessing system, comprising a pre-emphasis module, the pre-emphasis module is used to pre-emphasize a voice signal to be emphasized, and obtain a pre-emphasized voice signal;

[0041] The pre-emphasis module includes a calculation unit and a pre-emphasis unit;

[0042] The calculation unit is used to calculate the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized. t ;

[0043] The pre-emphasis unit is used to perform pre-emphasis processing on s(t) based on the transient recognition value, including:

[0044] Determine whether the transient recognition value is greater than the set transient recognition value comparison value;

[0045] If not, then s(t) is pre-emphasized using a pre-emphasis formula with a fixed pre-emphasis coefficient;

[0046] If yes, then s(t) is pre-emphasized using the following method, including:

[0047] p(t)=sq(t)-δ(t)×sq(t-1)

[0048] p(t) represents the result of pre-emphasis on s(t), sq(t) and sq(t-1) represent the amplitudes of the sampling points at sampling times t and t-1 respectively, and δ(t) represents the adaptive pre-emphasis coefficient; ut represents the set of sampling moments in the time interval [tT, t], T represents the set duration, and ampl i and amplifier t Respectively represent the normalized values ​​of the amplitude of the sampling point at sampling time i and t, max The value of ampl is 1, max It represents the comparison value of the normalized value of the set amplitude, φ(i) represents the pre-emphasis coefficient sampled when pre-emphasis is performed on the sampling point at sampling time i, and β represents the proportional value.

[0049] The above scheme calculates the transient recognition value, so that when the probability that the sampling point is located in the transient noise is greater, it is more inclined to use the formula with a non-fixed pre-emphasis coefficient to pre-emphasize the amplitude of the sampling point, and when the probability that the sampling point is located in the transient noise is smaller, it is more inclined to use the formula with a fixed pre-emphasis coefficient to pre-emphasize the amplitude of the sampling point, avoiding the use of a fixed pre-emphasis coefficient for pre-emphasis on all sampling points, and achieving a better balance between the effect of pre-emphasis and the efficiency of pre-emphasis. It can improve the accuracy of the recognized text when further recognition is performed based on the speech signal after pre-emphasis.

[0050] When calculating the adaptive pre-emphasis coefficient, the present invention obtains the pre-emphasis coefficient by weighted values ​​of the pre-emphasis coefficients of the sampling points within a certain time length before the sampling time t, so that the greater the time difference between the sampling points and the greater the difference in normalized amplitude, the greater the influence of the pre-emphasis coefficient of the referenced sampling point on the adaptive pre-emphasis coefficient finally calculated; the smaller the time difference between the sampling points and the smaller the difference in normalized amplitude, the smaller the influence of the pre-emphasis coefficient of the referenced sampling point on the adaptive pre-emphasis coefficient finally calculated, thereby achieving better suppression of transient noise in the pre-emphasis stage, which is beneficial to improving the accuracy of the results obtained in the subsequent recognition process.

[0051] Specifically, the time length between sampling times t and t-1 is the length of the sampling time interval.

[0052] Preferably, the ratio value is 0.5.

[0053] Preferably, the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized is calculated. t ,include:

[0054]

[0055] α 1 represents the amplitude weight, α 2 represents the persistence weight, ampl t Represents the normalized value of the amplitude of s(t), ampl pre Indicates the maximum value of the normalized amplitude of the sampling points in the time interval [tT, t+T] at the sampling time. N indicates the total number of sampling points in the time interval [tT, t+T] at the sampling time. Nh t It indicates the total number of sampling points whose normalized amplitude value is greater than z times the absolute value of the average amplitude of the speech signal to be emphasized among all sampling points whose sampling moments are within the time interval [tT, t+T]; T indicates the set duration.

[0056] In the present invention, the transient recognition value is calculated based on two types of parameters: the relationship between the normalized amplitude of the sampling point in a given time interval and the normalized amplitude of other sampling points, and the number of sampling points whose normalized amplitude meets the requirements in the given time interval. The smaller the difference between the normalized amplitude of the sampling point at time t and the maximum value of the normalized amplitude of the sampling point in the given time interval, and the greater the number of sampling points whose normalized amplitude meets the requirements in the given time interval, the greater the transient recognition value, so that the probability of the sampling point belonging to the transient noise can be more accurately represented from different angles.

[0057] Preferably, the value of z is 10.

[0058] Specifically, the amplitude of transient noise is generally relatively high, and there will be a peak significantly larger than the normal speech signal in the time domain graph. Therefore, by setting the value of z, the sampling points with a high probability of belonging to transient noise in a given time interval can be identified.

[0059] Preferably, the average amplitude of the speech signal to be emphasized is an average value of the normalized value of the amplitude of each sampling point of the speech signal to be emphasized.

[0060] Preferably, the set duration is 10 ms.

[0061] Specifically, transient noise has the characteristic of short duration. Therefore, by setting a smaller duration to obtain a smaller time interval, the transient recognition value can represent the probability of the sampling point belonging to transient noise with higher accuracy.

[0062] Preferably, the pre-emphasis formula with a fixed pre-emphasis coefficient includes:

[0063]

[0064] Indicates the set pre-emphasis coefficient.

[0065] Specifically, a fixed value pre-emphasis coefficient uses the same pre-emphasis processing method for both transient noise and non-transient noise, and therefore cannot effectively suppress transient noise.

[0066] Preferably, the time interval between t and the first sampling moment is greater than or equal to T.

[0067] Preferably, when the time interval between t and the first sampling moment is less than T, pre-emphasis is performed on the sampling point using a pre-emphasis formula with a fixed pre-emphasis coefficient.

[0068] Specifically, when t is relatively small, since the number of sampling points used for reference is insufficient, pre-emphasis processing can be performed using a pre-emphasis formula with a fixed pre-emphasis coefficient. When the value of t is larger, pre-emphasis processing can be performed using a pre-emphasis formula with an adaptive pre-emphasis coefficient.

[0069] Preferably, the pre-emphasis coefficient is set to 0.91.

[0070] Preferably, the amplitude weight is 0.6 and the persistence weight is 0.4;

[0071] The contrast value of the transient recognition value is set to 0.7.

[0072] Preferably, the normalized value of the amplitude is calculated using the following formula:

[0073]

[0074] ampl t represents the normalized value of the amplitude of the sampling point at sampling time t, sq(t) represents the amplitude of the sampling point at sampling time t, sq min and sq max They respectively represent the minimum and maximum amplitudes of all sampling points in the speech signal to be emphasized.

[0075] Preferably, the normalized value of the amplitude can also be obtained by logarithmic normalization, square root normalization, etc. Accordingly, the set comparison value of the normalized value of the amplitude should also be changed.

[0076] Preferably, if Figure 2 As shown, a microphone module is also included;

[0077] The microphone module is used to convert the user's voice into a voice signal to be emphasized when the user needs to input interactive information by voice input.

[0078] Specifically, the microphone module may be disposed in a smart phone, or in a notebook, or may be an independent module outside the computing device.

[0079] Preferably, the user refers to the user of the AI ​​voice software.

[0080] The microphone module collects the user's voice and converts the sound waves into electrical signals to obtain analog signals. After that, the analog signals are amplified, filtered, and processed, and then the processed analog signals are sampled, quantized, and encoded to obtain the voice signal to be emphasized.

[0081] Preferably, it also includes a framing module;

[0082] The framing module is used to frame the pre-emphasized speech signal to obtain multiple speech frames.

[0083] Speech signal framing (or segmentation) is a common signal processing technique, which is mainly used to divide continuous speech signals into smaller time windows of a certain length. This technique has many advantages in speech signal analysis, feature extraction, encoding and other processing tasks:

[0084] Handling time invariance

[0085] The characteristics of speech signals vary in time. By dividing the signal into frames, the signal in each frame can be regarded as relatively stable in time, which makes the processing methods (such as Fourier transform) in each frame more accurate and reliable.

[0086] Improve computational efficiency

[0087] Framing processing converts large signals into small blocks of signals, which can reduce the amount of data for each processing and improve computational efficiency. This makes it easy to apply algorithms such as frequency domain analysis (such as short-time Fourier transform).

[0088] Reduce non-stationarity

[0089] Speech signals are non-stationary (i.e., their statistical characteristics change over time). By framing, each frame can be considered as a stationary signal segment, thus simplifying the analysis and processing of the signal. For example, the short-time Fourier transform (STFT) assumes that the signal is stationary within each frame, which makes spectrum analysis more accurate.

[0090] Preferably, the pre-emphasized speech signal is framed to obtain a plurality of speech frames, including:

[0091] For the kth speech frame, the calculation formula of the corresponding inter-frame distance is:

[0092]

[0093] tist k,k-1 represents the inter-frame distance between the kth and k-1th speech frames, tstd represents the duration of the speech frame, sq max Represents the set hlfu k-1 The maximum value of the amplitude of the sampling points in j Indicates hlfu k-1 The amplitude of sampling point j in , avesq represents the set hlfu k-1 The average amplitude of the sampling points in , nhlfu represents the set hlfu k-1 The total number of sampling points in Nhk represents the set hlfu k-1 The number of sampling points in which the transient recognition value is greater than the set transient recognition value comparison value when pre-emphasis is performed, hlfu k-1 Indicates that the sampling time is in the time interval The set of sampling points, midt k-1 represents the sampling time of the sampling point at the center of the k-th speech frame; 1 represents the volatility weight, η 2 represents the transient weight;

[0094] For the kth speech frame, the time interval of the sampling points contained in it is

[0095] In the present invention, the inter-frame distance does not adopt a fixed value, but is obtained by comprehensive calculation based on the amplitude fluctuation degree of the sampling points of the right half of the previous speech frame and the number of sampling points whose transient recognition values ​​are greater than the set transient recognition value contrast value in the pre-emphasis stage. Therefore, when the amplitude fluctuation is greater and the number of sampling points whose transient recognition values ​​are greater than the set transient recognition value contrast value is greater, a smaller inter-frame distance is adopted. By utilizing the calculation results of the pre-emphasis stage, the amount of calculation can be effectively reduced, and the linkage between the two pre-processing processes is realized. Therefore, when the amplitude fluctuation of the sampling points is greater and the probability of being interfered by transient noise is greater, a smaller inter-frame distance is calculated, so that the area with greater recognition difficulty can be repeatedly included by two adjacent speech frames to a greater extent, which can provide more effective information for the subsequent speech recognition process and improve the accuracy of the text results obtained by the subsequent speech recognition.

[0096] Specifically, the value of k is greater than or equal to 2. For the first speech frame, the sampling time of the sampling points contained therein is in a time interval of [0, tstd].

[0097] Specifically, the sampling point at the center of the k-th speech frame refers to a sampling point whose sampling time is at the median of the sampling times corresponding to all the sampling points in the k-th speech frame.

[0098] If the number of all sampling points in the k speech frames is an odd number, the median is the number in the middle at the sampling time; if it is an even number, the median is the average of the two numbers in the middle at the sampling time.

[0099] Preferably, the duration of the speech frame is 40 ms.

[0100] Preferably, the fluctuation weight is 0.7 and the transient weight is 0.3.

[0101] Preferably, it also includes a windowing module;

[0102] The framing module is used to perform windowing processing on the speech frames to obtain multiple speech frames that have undergone windowing processing.

[0103] Windowing the framed speech signal is a common technique, the main purpose of which is to improve the quality of the results when performing signal analysis and processing. The following are some of the main advantages of windowing the framed speech signal:

[0104] Reducing spectral leakage

[0105] When performing Fourier transform on a signal, if the beginning and end of the signal are not smooth, spectrum leakage will occur, making the spectrum of the signal inaccurate. Windowing technology reduces the impact of spectrum leakage by introducing a smooth transition at the boundary of each frame signal, thereby obtaining a clearer spectrum representation.

[0106] Improved frequency domain resolution

[0107] Different window functions (such as Hamming window, Hanning window, Blackman window, etc.) have different frequency domain characteristics. By selecting a suitable window function, better resolution and lower main lobe width can be obtained in the frequency domain, thereby improving the accuracy of the spectrum.

[0108] Improve signal stability

[0109] Windowing can make the start and end of each frame smoother, thereby improving the signal stability of each frame. This helps to reduce the analysis error caused by signal non-stationarity, making subsequent analysis and processing more stable and reliable.

[0110] Preferably, the window function used by the windowing module includes:

[0111] Rectangular Window: A simple window function.

[0112] Hamming window: has good spectrum leakage characteristics.

[0113] Hanning window: can effectively reduce spectrum leakage.

[0114] Blackman Window: Provides lower spectral leakage and higher mainlobe to sidelobe ratio.

[0115] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI voice software data preprocessing system, characterized in that: It includes a pre-emphasis module, which is used to pre-emphasize the voice signal to be emphasized to obtain a pre-emphasized voice signal; The pre-emphasis module includes a calculation unit and a pre-emphasis unit; The calculation unit is used to calculate the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized. t ; The pre-emphasis unit is used to perform pre-emphasis processing on s(t) based on the transient recognition value, including: Determine whether the transient recognition value is greater than the set transient recognition value comparison value; If not, then s(t) is pre-emphasized using a pre-emphasis formula with a fixed pre-emphasis coefficient; If yes, then s(t) is pre-emphasized using the following method, including: p(t)=sq(t)-δ(t)×sq(t-1) p(t) represents the result of pre-emphasis on s(t), sq(t) and sq(t-1) represent the amplitudes of the sampling points at sampling times t and t-1 respectively, and δ(t) represents the adaptive pre-emphasis coefficient; ut represents the set of sampling moments in the time interval [tT, t], T represents the set duration, and ampl i and amplifier t Respectively represent the normalized values ​​of the amplitude of the sampling point at sampling time i and t, max The value of ampl is 1, max It represents the comparison value of the normalized value of the set amplitude, φ(i) represents the pre-emphasis coefficient sampled when pre-emphasis is performed on the sampling point at sampling time i, and β represents the proportional value.

2. The AI ​​voice software data preprocessing system according to claim 1, characterized in that: Calculate the transient recognition value tranoicoef of the sampling point s(t) at the sampling time t in the speech signal to be emphasized t ,include: α1 represents the amplitude weight, α2 represents the persistence weight, and t Represents the normalized value of the amplitude of s(t), ampl pre Indicates the maximum value of the normalized amplitude of the sampling points in the time interval [tT, t+T] at the sampling time. N indicates the total number of sampling points in the time interval [tT, t+T] at the sampling time. Nh t It indicates the total number of sampling points whose normalized amplitude value is greater than z times the absolute value of the average amplitude of the speech signal to be emphasized among all sampling points whose sampling moments are within the time interval [tT, t+T]; T indicates the set duration.

3. An AI voice software data preprocessing system according to claim 1 or 2, characterized in that: The set duration is 10ms.

4. The AI ​​voice software data preprocessing system according to claim 1, characterized in that: The pre-emphasis formula with a fixed pre-emphasis coefficient includes: Indicates the set pre-emphasis coefficient.

5. The AI ​​voice software data preprocessing system according to claim 1, characterized in that: The pre-emphasis factor is set to 0.

91.

6. The AI ​​voice software data preprocessing system according to claim 2, characterized in that: The amplitude weight is 0.6, and the persistence weight is 0.4; The contrast value of the transient recognition value is set to 0.

7.

7. An AI voice software data preprocessing system according to claim 1 or 2, characterized in that: The normalized value of the amplitude is calculated using the following formula: ampl t represents the normalized value of the amplitude of the sampling point at sampling time t, sq(t) represents the amplitude of the sampling point at sampling time t, sq min and sq max They respectively represent the minimum and maximum amplitudes of all sampling points in the speech signal to be emphasized.

8. The AI ​​voice software data preprocessing system according to claim 1, characterized in that: Also included is a microphone module; The microphone module is used to convert the user's voice into a voice signal to be emphasized when the user needs to input interactive information by voice input.

9. The AI ​​voice software data preprocessing system according to claim 1, characterized in that: Also includes a framing module; The framing module is used to frame the pre-emphasized speech signal to obtain multiple speech frames.

10. The AI ​​voice software data preprocessing system according to claim 9, characterized in that: Also includes a windowing module; The framing module is used to perform windowing processing on the speech frames to obtain multiple speech frames that have undergone windowing processing.

Images