Signal-to-Noise Ratio Estimation Method, Device, Medium, and Computing Device

Abstract

Embodiments of the present disclosure provide a signal-to-noise ratio estimation method, apparatus, medium, and computing device. The method includes: dividing a target audio into an audio frame sequence, where the target audio is obtained by superimposing a reference speech and a reference transient noise in the time domain; determining speech frames and superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; respectively determining the speech energy and the noise energy in the superimposed frames according to the audio energy of each of the speech frames and the superimposed frames; and calculating the signal-to-noise ratio of the target audio based on the speech energy and the noise energy in the superimposed frames. Applying this method to perform active signal-to-noise ratio estimation on an audio containing a transient noise signal, the estimation result has a high accuracy.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of signal processing technology. More specifically, embodiments of the present disclosure relate to a signal-to-noise ratio estimation method, apparatus, medium, and computing device. Background Art

[0002] This section aims to provide background or context for the embodiments of the present disclosure stated in the claims. The descriptions herein are not admitted to be prior art merely because they are included in this section.

[0003] In the field of audio processing, the signal-to-noise ratio (SNR) is often used to measure the relative magnitude of the useful signal and noise in the audio to be measured. The process of determining the signal-to-noise ratio of the audio to be measured is usually referred to as signal-to-noise ratio estimation. The methods of signal-to-noise ratio estimation can be divided into two categories: active estimation and passive estimation. Among them, active estimation is usually used for signal-to-noise ratio estimation of synthesized audio.

[0004] In the related art, active estimation is usually implemented by statistically averaging the energy. For example, energy statistics are respectively performed within the speech segment and the noise segment of the audio to be detected, and the signal-to-noise ratio of the audio to be measured is calculated based on the statistically averaged energy. Summary of the Invention

[0005] There is a need for an improved signal-to-noise ratio estimation method to improve the accuracy of active signal-to-noise ratio estimation for audio containing transient noise.

[0006] In this context, embodiments of the present disclosure are expected to provide a signal-to-noise ratio estimation method, apparatus, medium, and computing device.

[0007] In a first aspect of the embodiments of the present disclosure, a signal-to-noise ratio estimation method is provided. The method includes:

[0008] Dividing a target audio into an audio frame sequence, where the target audio is obtained by superimposing a reference speech and a reference transient noise in the time domain;

[0009] Determining speech frames and superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; the speech frames contain speech signals, the superimposed frames contain speech signals and transient noise signals, and the duration of the transient noise signals is less than a duration threshold;

[0010] Respectively determining the speech energy and noise energy in the superimposed frames according to the audio energy of each of the speech frames and the superimposed frames; the audio energy of the superimposed frames is the sum of the speech energy and the noise energy in the superimposed frames;

[0011] Calculating the signal-to-noise ratio of the target audio based on the speech energy and the noise energy in the superimposed frames.

[0012] Optionally, determining the speech energy in the superimposed frame according to the audio energy of each of the speech frame and the superimposed frame includes:

[0013] Determining the speech energy of the speech frames in the audio frame sequence, and determining the average speech energy of the speech frames according to the number of speech frames;

[0014] Determining the speech energy in the superimposed frame according to the average speech energy and the number of superimposed frames of the superimposed frame.

[0015] Optionally, determining the noise energy in the superimposed frame according to the audio energy of each of the speech frame and the superimposed frame includes:

[0016] Determining the audio energy of the superimposed frame;

[0017] Taking the difference between the audio energy of the superimposed frame and the speech energy in the superimposed frame as the noise energy in the superimposed frame.

[0018] Optionally, determining the audio energy of any audio frame includes:

[0019] Collecting the signal energy of the corresponding audio signal at multiple sample points in the any audio frame;

[0020] Taking the accumulated value of the collected multiple signal energies as the audio energy of the any audio frame.

[0021] Optionally, dividing the target audio into an audio frame sequence includes:

[0022] Dividing the time axis of the target audio into multiple time intervals according to a preset duration, and the audio frames corresponding to the respective time intervals form the audio frame sequence in the order of precedence in the time axis.

[0023] Optionally, determining the speech frames and the superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise includes:

[0024] Determining a speech time period, a noise time period, and an overlapping time period between the speech time period and the noise time period in the time axis of the target audio according to the timing information of the reference speech and the reference transient noise;

[0025] Determining the audio frames in the audio frame sequence that belong to the speech time period and do not belong to the overlapping time period as speech frames; and determining the audio frames in the audio frame sequence that belong to the overlapping time period as superimposed frames.

[0026] In a second aspect of the embodiments of the present disclosure, a signal-to-noise ratio estimation device is provided, and the device includes:

[0027] A sequence division module, configured to divide a target audio into an audio frame sequence, where the target audio is obtained by superimposing a reference speech and a reference transient noise in the time domain;

[0028] An audio frame determination module, configured to determine a speech frame and a superimposed frame in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; the speech frame includes a speech signal, and the superimposed frame includes a speech signal and a transient noise signal, and the duration of the transient noise signal is less than a duration threshold;

[0029] An energy determination module, configured to respectively determine the speech energy and the noise energy in the superimposed frame according to the audio energy of each of the speech frame and the superimposed frame; the audio energy of the superimposed frame is the sum of the speech energy and the noise energy in the superimposed frame;

[0030] A signal-to-noise ratio calculation module, configured to calculate the signal-to-noise ratio of the target audio based on the speech energy and the noise energy in the superimposed frame.

[0031] Optionally, the energy determination module is further configured to:

[0032] Determine the speech energy of the speech frames in the audio frame sequence, and determine the average speech frame energy according to the number of speech frames;

[0033] Determine the speech energy in the superimposed frame according to the average speech frame energy and the number of superimposed frames of the superimposed frame.

[0034] Optionally, the energy determination module is further configured to:

[0035] Determine the audio energy of the superimposed frame;

[0036] Take the difference between the audio energy of the superimposed frame and the speech energy in the superimposed frame as the noise energy in the superimposed frame.

[0037] Optionally, the energy determination module is configured to determine the audio energy of any audio frame in the following manner:

[0038] Collect the signal energy of the corresponding audio signal at multiple sample points in the any audio frame;

[0039] Take the cumulative value of the collected multiple signal energies as the audio energy of the any audio frame.

[0040] Optionally, the sequence division module is further configured to:

[0041] The time axis of the target audio is segmented into multiple time intervals according to a preset duration, and the audio frames corresponding to each time interval form the audio frame sequence in the order of precedence in the time axis.

[0042] Optionally, the audio frame determination module is further configured to:

[0043] Determine a speech time period, a noise time period, and an overlapping time period between the speech time period and the noise time period in the time axis of the target audio according to the timing information of the reference speech and the reference transient noise;

[0044] Determine the audio frames belonging to the speech time period and not belonging to the overlapping time period in the audio frame sequence as speech frames; and determine the audio frames belonging to the overlapping time period in the audio frame sequence as superimposed frames.

[0045] In a third aspect of the embodiments of the present disclosure, a medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the signal-to-noise ratio estimation method described in any one of the embodiments in the first aspect above is implemented.

[0046] In a fourth aspect of the embodiments of the present disclosure, a computing device is provided, including:

[0047] A processor;

[0048] A memory for storing instructions executable by the processor;

[0049] Wherein, the processor implements the signal-to-noise ratio estimation method described in any one of the embodiments in the first aspect above by running the executable instructions.

[0050] According to the signal-to-noise ratio estimation method of the embodiments of the present disclosure, after dividing the target audio obtained by superimposing the reference speech and the reference transient noise in the time domain into an audio frame sequence, the speech energy and the noise energy in the superimposed frames are respectively determined according to the audio energy of the speech frames and the superimposed frames, and then the signal-to-noise ratio of the target audio is calculated based on the two.

[0051] In this way, the superimposed segment with a certain duration in the target audio can be divided into multiple superimposed frames. It can be understood that the duration of any superimposed frame is usually less than the duration of a single transient noise. Therefore, the transient noise signal corresponding to any superimposed frame can be approximately considered to be stable, so that for the superimposed segment with a short duration, the noise energy therein can also be estimated more accurately. It can be seen that by improving the estimation accuracy of the noise energy and the speech energy in the superimposed segment, this solution can improve the accuracy of active signal-to-noise ratio estimation for audio containing transient noise. Description of the Drawings

[0052] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the drawings, several embodiments of the present disclosure are shown by way of illustration and not limitation, wherein:

[0053] Figure 1 Schematically shows a schematic diagram of an audio synthesis process according to an embodiment of the present disclosure;

[0054] Figure 2 Schematically shows a flowchart of a signal-to-noise ratio estimation method according to an embodiment of the present disclosure;

[0055] Figure 3 Schematically shows a schematic diagram of an audio frame sequence according to an embodiment of the present disclosure;

[0056] Figure 4 Schematically shows a flowchart of another signal-to-noise ratio estimation method according to an embodiment of the present disclosure;

[0057] Figure 5 Schematically shows a schematic diagram of a medium according to an embodiment of the present disclosure;

[0058] Figure 6 Schematically shows a block diagram of a signal-to-noise ratio estimation device according to an embodiment of the present disclosure;

[0059] Figure 7 Schematically shows a schematic diagram of a computing device according to an embodiment of the present disclosure.

[0060] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. Detailed Embodiments

[0061] The principles and spirit of the present disclosure will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present disclosure, and not to limit the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to be able to fully convey the scope of the present disclosure to those skilled in the art.

[0062] Those skilled in the art know that the embodiments of the present disclosure can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present disclosure can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0063] According to the embodiments of the present disclosure, a signal-to-noise ratio estimation method, apparatus, medium, and computing device are provided.

[0064] In this document, the number of any element in the drawings is for illustration rather than limitation, and any naming is only for distinction without any limiting meaning.

[0065] The principles and spirit of the present disclosure will be elaborated in detail below with reference to several representative embodiments of the present disclosure. Summary of the Invention

[0067] In related technologies, active estimation is usually implemented in the way of statistical energy average. For example, energy statistics are respectively carried out in the speech segment and the noise segment of the audio to be detected, and the signal-to-noise ratio of the audio to be measured is calculated based on the statistical energy average.

[0068] The inventors of the present disclosure have found that the active estimation method in related technologies statistically calculates the average energy of the noise segment and the speech segment, and this method is more applicable to steady-state noise or persistent non-steady-state noise. However, for transient noise such as keyboard tapping sounds that have strong suddenness in time, the duration of a single transient noise is usually short. Therefore, using the above method will disperse the noise energy within the duration of the transient noise to a longer time interval, resulting in a lower estimated noise energy result, and thus leading to a lower accuracy of the signal-to-noise ratio estimation result.

[0069] To solve the above problems, the present disclosure provides a signal-to-noise ratio estimation method, device, medium, and computing device. After dividing the target audio obtained by superimposing the reference speech and the reference transient noise in the time domain into an audio frame sequence, the speech energy and the noise energy in the superimposed frame are respectively determined according to the audio energy of the speech frame and the superimposed frame, and then the signal-to-noise ratio of the target audio is calculated based on the two.

[0070] In this way, the superimposed segment with a certain duration in the target audio can be divided into multiple superimposed frames. It can be understood that the duration of any superimposed frame is usually less than the duration of a single transient noise. Therefore, the transient noise signal corresponding to any superimposed frame can be approximately considered to be stable. Thus, for the superimposed segment with a short duration, the noise energy therein can also be estimated more accurately. It can be seen that by improving the estimation accuracy of the noise energy and the speech energy in the superimposed segment, this solution can improve the accuracy of active signal-to-noise ratio estimation for audio containing transient noise.

[0071] After introducing the basic principles of the present disclosure, the various non-limiting embodiments of the present disclosure will be specifically introduced below.

[0072] Overview of Application Scenarios

[0073] It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in this regard. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0074] According to the signal-to-noise ratio estimation method of the embodiments of the present disclosure, after dividing the target audio obtained by superimposing the reference speech and the reference transient noise in the time domain into an audio frame sequence including a plurality of audio frames, the speech frames and the superimposed frames in the audio frame sequence can be determined according to the timing information of the reference speech and the reference transient noise; then, according to the audio energies of the speech frames and the superimposed frames respectively, the speech energy and the noise energy in the superimposed frames are determined respectively, and finally the signal-to-noise ratio of the target audio is calculated according to the total speech energy and the total noise energy.

[0075] In practical applications, the signal-to-noise ratio estimation method described in the present disclosure can be applied to an audio processing device. The audio processing device can be a server of an audio processing platform. For example, the server can apply this method to estimate the signal-to-noise ratio of the audio to be consumed maintained in the audio consumption platform; wherein, the server can be a physical server including an independent host, a virtual server hosted by a host cluster, a cloud server, etc. Or, the audio processing device can also be a terminal device used by a user. For example, the terminal device can apply this method to estimate the signal-to-noise ratio of local audio or online audio; wherein, the terminal device can be an electronic device such as a tablet device, a laptop computer, a personal digital assistant (PDA), a wearable device (such as smart glasses, smart watches, etc.). In addition, the signal-to-noise ratio estimation method described in the present disclosure can be implemented as an independent application program or as a functional plug-in in audio processing software, and the embodiments of the present disclosure do not limit this.

[0076] The signal-to-noise ratio estimation method described in the present disclosure essentially belongs to an active estimation method. The application object of this method is the target audio obtained by superimposing the reference speech and the reference transient noise on each other in the time domain. When applying this method to estimate the signal-to-noise ratio of the target audio to be estimated, the audio processing device needs to know the reference speech and the reference transient noise.

[0077] As Figure 1 shown is the process of obtaining the target audio by superimposing the real reference speech and the reference transient noise. From Figure 1 it can be seen that the reference speech and the reference transient noise have the same time axis (that is, the audio files of the reference speech and the reference transient noise have the same duration), so the target audio obtained after merging the two also has the same time axis as the two: that is Figure 1The abscissa lengths of the reference speech, the reference transient noise, and the target audio are the same. Those skilled in the art can understand that during the superposition process of the above-mentioned reference speech and reference transient noise, the speech signal and the transient noise signal will affect each other, resulting in changes in the speech signal and the transient noise signal in the target audio in terms of frequency, wavelength, energy, etc. In other words, before and after the superposition of the reference speech and the reference transient noise, its time-domain information does not change, while the frequency-domain information will change.

[0078] Exemplary Method

[0079] Next, a method for signal-to-noise ratio estimation according to an exemplary embodiment of the present disclosure will be described with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this regard. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0080] Reference Figure 2 , Figure 2 FIG. schematically shows a flowchart of a signal-to-noise ratio estimation method according to an embodiment of the present disclosure. The signal-to-noise ratio estimation method may include the following steps:

[0081] Step S201, dividing the target audio into an audio frame sequence, where the target audio is obtained by superposing a reference speech and a reference transient noise in the time domain.

[0082] In the solution of the embodiment of the present disclosure, the target audio is obtained by superposing a reference speech and a reference transient noise in the time domain. For example, the reference speech and the reference transient noise can be superposed by means of recording. For example, two sets of playback devices can be used to play the reference speech and the reference transient noise respectively, and a set of recording devices can be used to simultaneously collect the speech signal corresponding to the reference speech and the noise signal corresponding to the reference transient noise to obtain the target audio. Since the played speech signal and transient noise signal will automatically superpose in the surrounding space (essentially the superposition of audio waveforms), the target audio obtained by continuously recording for a period of time is the superposition result of the reference speech and the reference transient noise during this period of time (i.e., the time domain). For another example, the superposition (or merging) function of audio processing software can also be used to superpose the reference speech and the reference transient noise to obtain the target audio, and the embodiments of the present disclosure do not limit this.

[0083] Among them, the target audio can be music, and the corresponding reference speech can be the sound emitted by a singer, and the reference transient noise can be the sound of harmony or musical instruments. Or, the target audio can also be non-music audio, where the reference speech can be the sound emitted by a sound source, and the reference transient noise can be the background sound of the environment where the sound source is located, etc.

[0084] Among them, the above-mentioned reference speech and reference transient noise are essentially a speech signal and a transient noise signal respectively. For example, the above-mentioned musical instrument can be a percussion instrument, and the corresponding reference transient noise is the hitting sound of this type of instrument; for another example, the background sound of the environment where the above-mentioned sound source is located can be the sound of keyboard tapping, the sound of raindrops hitting the ground, the sound of a ball being slapped, etc., which will not be elaborated here. Among them, the duration of the above-mentioned transient noise signal can be less than a duration threshold, and the duration threshold can be preset to 0.1s, 1s, 0.5s, etc., and the embodiments of the present disclosure do not limit this.

[0085] For the target audio to be detected, the audio processing device can first divide it into an audio frame sequence, which contains multiple audio frames, and each audio frame is arranged in sequence according to the time order to form the audio frame sequence. For example, the audio processing device can divide the time axis of the target audio into multiple time intervals according to a preset duration, and the audio frames corresponding to each divided time interval form the audio frame sequence in the order in which they appear in the time axis.

[0086] As Figure 3 shown, the reference speech is an intermittent steady-state speech signal, and the reference transient noise is a spike-shaped transient noise signal. The reference speech and the reference transient noise are superimposed to obtain the target audio. The abscissa of any one of the reference speech, the reference transient noise, and the target audio is time, that is, the time axis representing the audio; the ordinate is the amplitude of the audio signal, which is used to represent the energy magnitude of the corresponding audio signal. In comparison, Figure 3 any audio signal shown can be regarded as Figure 1 the positive half-axis of the signal above the time axis in the corresponding audio signal shown, for special note.

[0087] From Figure 3It can be seen that the reference speech only contains a speech signal. The time period corresponding to the speech signal in the time axis of the reference speech is recorded as a speech segment, and the part other than the speech signal is a blank segment. The reference transient noise only contains a transient noise signal. The time period corresponding to the transient noise signal in the time axis of the reference transient noise is recorded as a noise segment, and the part other than the transient noise signal is a blank segment. Among them, the speech segment in the reference speech and the noise segment in the reference transient noise can be obtained by detecting the reference speech and the reference transient noise respectively through the Voice Activity Detection (VAD) technology in related technologies. For example, for the reference speech, the detection device can determine the presence, absence, and magnitude of the audio energy (i.e., speech energy) at each time point through sampling: the continuous time period in the time axis where there is audio energy or the audio energy value is higher than the detection threshold is the speech segment, and the remaining time periods are blank segments. For the reference transient noise, the detection device can determine the presence, absence, and magnitude of the audio energy (i.e., noise energy) at each time point through sampling: the continuous time period in the time axis where there is audio energy or the audio energy value is higher than the detection threshold is the noise segment, and the remaining time periods are blank segments. Of course, the specific implementation details of the VAD technology can be referred to the records in related technologies and will not be elaborated here.

[0088] Since the above reference speech and reference transient noise are superimposed in the time domain to obtain the target audio, based on the detection results of the above speech segments and noise segments, the audio processing device can determine the time period types in the target audio. For example, the blank segments in the reference speech and the blank segments in the reference transient noise are superimposed to obtain the blank segments in the target audio, such as Figure 3 the blank segment 1, blank segment 2, and blank segment 3 shown in the target audio; the speech segments in the reference speech and the blank segments in the reference transient noise are superimposed to obtain the speech segments in the target audio, such as Figure 3 the speech segment 1, speech segment 2, and speech segment 3 shown in the target audio; the blank segments in the reference speech and the noise segments in the reference transient noise are superimposed to obtain the noise segments in the target audio, such as Figure 3 the noise segment 1 and noise segment 2 shown in the target audio; the speech segments in the reference speech and the noise segments in the reference transient noise are superimposed to obtain the superimposed segments in the target audio, such as Figure 3 the superimposed segment 1 shown in the target audio.

[0089] It should be noted that Figure 1The reference speech and the reference transient noise shown have the same and overlapping time axes, that is, the time lengths of the reference speech and the reference transient noise, as well as the starting and ending points of their respective time axes, are the same. However, in the specific implementation of this solution, the time axes of the reference speech and the reference transient noise may not overlap. For example, the starting point of the time axis of the reference transient noise may be located in the middle of the time axis of the reference speech, such as at the midpoint (corresponding to starting to play the reference transient noise when the reference speech has been played for half of its duration) during the recording of the target audio. In this case, the time axis length of the target audio will be 1.5 times the sum of the time axis lengths of the reference speech and the reference transient noise. Or, the time axes of the reference speech and the reference transient noise may also be different. For example, when the time length of the reference speech is 1 minute and the time length of the reference transient noise is 0.5 minute, if the starting moment of the reference transient noise is at 20 seconds of the reference speech (corresponding to starting to play the reference transient noise when the reference speech has been played for 20 seconds) during the recording of the target audio, the time length of the target audio obtained by superimposing the two is 1 minute, which will not be elaborated here. The following embodiments of this specification are all based on Figure 3 the scenario where the time axes of the reference speech and the reference transient noise shown are the same and overlapping for illustration.

[0090] For the time axis of the target audio, the audio processing device can divide it into multiple time intervals according to a preset duration. For example, Figure 3 the interval between any two adjacent minimum scales shown in the time axis shown is the above-mentioned preset duration. It can be understood that the smaller the above-mentioned preset duration, the more audio frames are divided, and the corresponding energy calculation accuracy is also higher. Therefore, to ensure that the estimated result of the final signal-to-noise ratio has a high accuracy, the preset duration can be set as small as possible. However, the smaller the preset duration, the greater the energy calculation amount in the subsequent estimation process. Therefore, the above-mentioned preset duration can be comprehensively set according to various factors such as the noise characteristics, the processing performance of the audio processing device, and the real-time requirement of the signal-to-noise ratio estimation. For example, it can be set to 10 ms, 20 ms, 5 ms, etc. The embodiments of the present disclosure do not limit this.

[0091] Step S202: Determine the speech frames and the superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise. The speech frames contain speech signals; the superimposed frames contain speech signals and transient noise signals, and the duration of the transient noise signals is less than the duration threshold.

[0092] After the above-mentioned division is completed, multiple audio frames can be obtained. At this time, the speech frames and the superimposed frames in the audio frame sequence can be further determined according to the timing information of the reference speech and the reference transient noise, that is, the types of each audio frame are determined.

[0093] In one embodiment, the audio processing device may determine a speech time period, a noise time period, and an overlapping time period between the speech time period and the noise time period in the time axis of the target audio according to the timing information of the reference speech and the reference transient noise. On this basis, the audio frames in the audio frame sequence that belong to the above speech time period and do not belong to the overlapping time period may be determined as speech frames; and the audio frames in the audio frame sequence that belong to the overlapping time period may be determined as superimposed frames. Of course, the audio frames in the audio frame sequence that belong to the noise time period and do not belong to the overlapping time period may also be determined as noise frames, and the audio frames in the audio frame sequence that do not belong to the speech time period, the noise time period, and the overlapping time period may also be determined as blank frames.

[0094] As can be seen from the above determination method, the types of audio signals included in different types of audio frames are also different. Specifically: only speech signals are included in speech frames, and both speech signals and transient noise signals are included in superimposed frames. As Figure 3 shown, there are a total of seven speech frames in speech segment 1. Any speech frame is obtained by superimposing the speech segment of the reference speech and the blank segment of the reference transient noise, and only speech signals are included therein; there are a total of three superimposed frames in superimposed segment 1. Any superimposed frame is obtained by superimposing the speech segment of the reference speech and the noise segment of the reference transient noise, and both speech signals and transient noise signals are included therein.

[0095] Of course, not all time points of any of the above audio frames correspond to the corresponding signals. For example, in the first speech frame of speech segment 1, the start time of the speech signal is before the start point of this speech frame; in the first superimposed frame of superimposed segment 2, the start time of the transient noise signal is after the start point of this superimposed frame; in the first noise frame of noise segment 2, the start time of the transient noise signal is after the start point of this noise frame.

[0096] In addition, the preset durations corresponding to each audio frame in the above embodiment are the same, that is, the lengths of all audio frames divided by the audio processing device are the same. In fact, the audio frames may also be divided according to different preset durations. For example, for a blank segment, a blank segment may be divided into a complete blank frame with a longer length. For example, the three blank frames included in blank segment 1 may be used as one blank frame to simplify the calculation steps in the subsequent estimation process. In fact, since any blank frame in the blank segment does not contain any signal or the signal energy of the contained signal is less than the detection threshold (such as less than the minimum value corresponding to the detection accuracy of the detection device), each blank frame may not participate in the subsequent estimation process. Therefore, the blank segment may not even be divided to further improve the signal-to-noise ratio estimation efficiency.

[0097] Step S203: Determine the speech energy and noise energy in the superimposed frame according to the audio energy of the speech frame and the superimposed frame respectively; the audio energy of the superimposed frame is the sum of the speech energy and the noise energy in the superimposed frame.

[0098] It should be noted that the signal-to-noise ratio of the target audio finally estimated by this solution is actually the average signal-to-noise ratio of each superimposed segment in the target audio, that is, the average signal-to-noise ratio of each superimposed frame after division. Therefore, to estimate the signal-to-noise ratio of the target audio, it is necessary to know the speech energy and noise energy in the superimposed frame. Before determining the speech energy and noise energy in the superimposed frame, the audio processing device can first determine the energy of each speech frame and superimposed frame. Among them, the "energy of the speech frame" described in the embodiments of the present disclosure is the sum of the audio energies (i.e., speech energies) of all speech frames in the audio frame sequence, and the "energy of the superimposed frame" is the sum of the audio energies (i.e., the speech energy and noise energy in the superimposed frame) of all superimposed frames in the audio frame sequence. This is hereby explained.

[0099] For any audio frame, the audio processing device can collect the signal energies of the corresponding audio signals at multiple sample points in the audio frame, then calculate the cumulative value of the collected multiple signal energies, and use it as the audio energy of the audio frame. Among them, the above cumulative value can be calculated by the method of root mean square (RMS) accumulation. For example, the energy RMSi calculation formula of any audio frame i is as follows in formula (1):

[0100]

[0101] where framesize is the number of sampling points in audio frame i, and S k (n) is the energy value of the kth sampling point. The specific sampling and calculation method of this energy value can refer to the audio short-time energy calculation method in the related art, which will not be elaborated here. The above number of sampling points can be a preset fixed value. For example, when framesize is 10, if the frame length of the audio frame is 10 ms, the audio processing device can collect the energy data of one sampling point every 1 ms in the audio frame, and calculate the sum of the energy data of 10 sampling points in this frame through the above formula (1) as the audio energy of this frame. In addition, in the case of digital sampling, the above S k (n) can be a floating-point number in the range of [0, 1]. The larger this number is, the greater the corresponding energy value is. As Figure 3 shown, for the fifth speech frame in speech segment 1, the audio energy of this speech frame is the sum of the energy data of 10 sampling points in this frame.

[0102] After obtaining the energies of each audio frame through the above method, the audio energy of each speech segment or superimposed segment can be further determined. For example, for Figure 3 the speech segment 1 shown, the audio energy of this speech segment is the sum of the audio energies of the seven speech frames within this segment; for the superimposed segment 1, the audio energy of this superimposed segment is the sum of the audio energies of the three superimposed frames within this segment. Of course, the audio energy of each noise segment can also be calculated. For example, the audio energy of the noise segment 2 is the sum of the audio energies of the three noise frames within this segment, which will not be elaborated here.

[0103] Since the blank frame does not contain any audio signal, the energy of the blank frame is zero. Based on this fact, when calculating the audio energy of a speech segment or a superimposed segment, in order to avoid data splitting and speed up the calculation, adjacent segments of the same type can be merged for calculation. For example, for the speech segment 1 and the speech segment 2, there is a blank segment 1 in between. When calculating the sum of the audio energies of the speech segment 1 and the speech segment 2, the formula (1) can be directly used to calculate the sum of the audio energies of the seven speech frames in the speech segment 1, the three blank frames in the blank segment 1, and the three speech frames in the speech segment 2 at one time. Of course, in this calculation process, the framesize in the formula (1) can be the sum of the number of sampling points of the speech segment 1, the blank segment 1, and the speech segment 2.

[0104] Of course, since the energy of the blank frame is zero, the blank frame can not participate in the signal-to-noise ratio estimation process. Furthermore, the audio processing device can only determine the audio energy of the speech frames and the superimposed frames respectively. For example, the framesize in the formula (1) can also be the sum of the number of sampling points of the speech segment 1 and the speech segment 2 to further reduce the number of sampling points to be calculated and improve the calculation efficiency. Similar processing can also be carried out for the noise segment 1 and the noise segment 2, which will not be elaborated here.

[0105] It can be understood that since only speech signals are contained in the speech frames, the energy of any speech frame is the energy of the corresponding speech signal; since both speech signals and transient noise signals are contained in the superimposed frames, the energy of the superimposed frame is the sum of the energy of the speech signal and the transient noise signal, that is, the energy of the superimposed frame is the sum of the speech energy in the superimposed frame and the noise energy in the superimposed frame. Based on this, it can be determined that the speech energy in the superimposed frames of the target audio is the sum of the speech energies in all the superimposed frames in the audio frame sequence, and the noise energy in the superimposed frames of the target audio is the sum of the noise energies in all the superimposed frames in the audio frame sequence. For any superimposed frame, since the actually collected audio energy of this frame is the sum of the speech energy and the noise energy in it, to calculate the speech energy and the noise energy in the superimposed frame, the speech energy and the noise energy in the superimposed frame need to be split.

[0106] The inventors found that due to the strong impact of transient noise, its amplitude fluctuations within the duration (even within the same superimposed frame) are often large (i.e., large energy changes), so it is difficult to accurately directly estimate the noise energy in the superimposed frame. However, for speech signals, their amplitude fluctuations in the speech segment and the superimposed segment are usually small (i.e., small energy changes). For example, the speaking volume of the same speaker with and without noise often does not change much. Based on this feature, the average energy of the speech frames can be used to approximately replace the average energy of the speech energy in the superimposed frame to estimate the speech energy in the superimposed frame.

[0107] For example, the audio processing device can first determine the audio energy of the speech frames in the audio frame sequence and the number of speech frames, and then determine the average speech energy of the speech frames based on the audio energy and the number of speech frames. Among them, the average speech energy of the speech frames is the average energy of the speech emitted by the sound source in each speech frame. Based on this, the audio processing device can determine the speech energy in the superimposed frame according to the average speech energy of the speech frames and the number of superimposed frames of the superimposed frame. For example, the product of the average speech energy of the speech frames and the number of superimposed frames of the superimposed frame can be calculated and used as the speech energy in the superimposed frame.

[0108] Obviously, this operation actually uses the average energy of the frames in the speech segment to approximately replace the average speech energy of the frames in the superimposed frame, and this approximation is exactly the specific manifestation of the "estimation" of the signal-to-noise ratio. It can be understood that when the preset duration corresponding to any audio frame (i.e., the frame length of the audio frame) is small enough, the above approximation operation can ensure that the estimation result of the signal-to-noise ratio estimation has a high accuracy. For example, when the duration of the transient noise signal is less than the duration threshold, the frame length of the audio frame can be much smaller than the duration threshold. For example, the duration threshold can be set as a preset multiple (20 times, 50 times, 100 times, etc.) of the preset duration. In this way, it can be ensured that the frame length of the audio frame is much smaller than the duration of the transient noise, so that the energy estimation result of the transient noise has a high accuracy, thereby ensuring that the final signal-to-noise ratio estimation result also has a high accuracy.

[0109] Take Figure 3 [[ID=ll]]as an example. Speech segment 1, speech segment 2, and speech segment 3 together contain fourteen speech frames. The audio processing device can calculate the average speech energy of these fourteen speech frames (i.e., the sum of the audio energy of the fourteen speech frames divided by 14); Superimposed segment 1 and superimposed segment 2 together contain five superimposed frames. Therefore, the audio processing device can use the product of the above average speech energy and 5 as the speech energy in the superimposed frame. In other words, the speech energy in the superimposed frame = (the number of superimposed frames / the number of speech frames) * the sum of the audio energy of all speech frames.

[0110] Further, the audio processing device may determine the audio energy of the superimposed frame, and then use the difference between the audio energy of the superimposed frame and the speech energy in the superimposed frame as the noise energy in the superimposed frame. That is, the noise energy in the superimposed frame = the audio energy of the superimposed frame - the speech energy in the superimposed frame. As Figure 3 shown, the sum of the noise energies of the five superimposed frames in superimposed segment 1 and superimposed segment 2 = the sum of the audio energies of these five superimposed frames - the sum of the speech energies in these five superimposed frames.

[0111] Step S204, calculate the signal-to-noise ratio of the target audio based on the speech energy and noise energy in the superimposed frame.

[0112] According to the speech energy in the superimposed frame and the noise energy in the superimposed frame determined in the foregoing manner, the audio processing device may calculate the signal-to-noise ratio SNRdB of the target audio. The calculation formula is as follows in Equation (2):

[0113]

[0114] where Es is the speech energy in the superimposed frame, En is the noise energy in the superimposed frame, and the unit of the signal-to-noise ratio calculated by the above Equation (2) is decibel (dB).

[0115] It can be understood that the speech energy in the superimposed frame estimated by the above method is the sum of the speech energies in all the superimposed frames in the audio sequence, and the calculated noise energy in the superimposed frame is the sum of the noise energies in all the superimposed frames in the audio sequence. Therefore, the signal-to-noise ratio of the target audio calculated based on the speech energy in the superimposed frame and the noise energy in the superimposed frame is the average signal-to-noise ratio of each superimposed segment in the target audio.

[0116] According to the signal-to-noise ratio estimation method described in the embodiments of the present disclosure, after dividing the target audio obtained by superimposing the reference speech and the reference transient noise in the time domain into an audio frame sequence including a plurality of audio frames, the audio processing device may determine the speech frames and superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; then, according to the audio energies of the speech frames and the superimposed frames respectively, determine the speech energy and noise energy in the superimposed frame, and calculate the signal-to-noise ratio of the target audio according to the speech energy and noise energy in the superimposed frame, and this signal-to-noise ratio is the estimation result of the active signal-to-noise ratio estimation of the target audio.

[0117] In this way, the superimposed segments with a certain duration in the target audio can be divided into multiple superimposed frames. It can be understood that the duration of any superimposed frame is usually less than the duration of a single transient noise. Therefore, the transient noise signal corresponding to any superimposed frame can be approximately considered stable. Thus, for the superimposed segments with a short duration, the noise energy therein can also be estimated more accurately. It can be seen that by improving the estimation accuracy of the noise energy and speech energy in the superimposed segments, this solution can improve the accuracy of the active signal-to-noise ratio estimation for the audio containing transient noise.

[0118] As described above, to estimate the signal-to-noise ratio of the target audio, it is necessary to estimate the speech energy in the superimposed frames and calculate the noise energy in the superimposed frames. When the average speech frame energy of each speech frame in the target audio is determined, the audio processing device can Figure 4 calculate the total speech energy in the manner described above. As Figure 4 shown in the flowchart of another signal-to-noise ratio estimation method, this method is applied to an audio processing device. This method estimates the signal-to-noise ratio by directly calculating the total speech energy, and this method may include the following steps S401 - S405.

[0119] In step S401, determine the audio energy of each speech frame and superimposed frame in the audio frame sequence of the target audio.

[0120] For the target audio, the audio processing device can first divide it into multiple audio frames to form an audio frame sequence, and then determine the speech frames and superimposed frames therein according to the timing information of the reference speech and the reference transient noise: the speech frames only contain speech signals, and the superimposed frames contain speech signals and transient noise signals. Then, data points can be sampled for each audio frame respectively, and the audio energy of each audio frame can be determined using the aforementioned formula (1). Thereafter, the audio processing device can sum the energies of adjacent audio frames of the same type to respectively determine the audio energy of the speech segment and the superimposed segment.

[0121] In step S402, calculate the average speech frame energy of each speech frame in the audio frame sequence.

[0122] In one embodiment, the audio processing device can determine the speech frame energy of the target audio (i.e., the total audio energy of each speech frame in the speech segment), and determine the number of speech frames in the target audio. Furthermore, the audio processing device can calculate the first average speech frame energy = the speech frame energy of the target audio / the number of speech frames.

[0123] As Figure 3As shown, there are three speech segments in the target audio: speech segment 1 (containing seven speech frames), speech segment 2 (containing three speech frames), and speech segment 3 (containing four speech frames). The number of speech frames corresponding to the three is 14. Therefore, after calculating the audio energy of each speech frame through formula (1), the audio processing device can calculate the sum of the audio energies of each speech frame as the speech frame energy of the target audio. Then, the average energy of the first speech frame can be calculated as = the speech frame energy of the target audio / 14.

[0124] The inventors also found that in the case where the target audio contains multiple speech segments (especially when the target audio is relatively long), the closer the speech segment is to the superimposed segment, the closer its signal amplitude is usually to the amplitude of the speech signal in the superimposed segment. Therefore, in another embodiment, the audio processing device can also determine the corresponding average energy of the speech frame for each superimposed segment, hereinafter referred to as the average energy of the second speech frame. For example, for a plurality of sequentially adjacent superimposed frames (usually corresponding to the same superimposed segment), determine a plurality of adjacent speech frames to the left of the first superimposed frame and / or a plurality of adjacent speech frames to the right of the last superimposed frame; further determine the sum of the audio energies of the above-mentioned respective adjacent speech frames and the number of speech frames of the above-mentioned adjacent speech frames, and calculate the average energy of the second speech frame accordingly = the audio energy of each adjacent speech frame / the number of speech frames of the adjacent speech frames corresponding to the plurality of sequentially adjacent superimposed frames. The average energy of the second speech frame determined in this way can be closer to the average speech energy in the above-mentioned plurality of superimposed frames, thereby improving the estimation accuracy of the speech energy in the superimposed frames and further improving the estimation accuracy of the signal-to-noise ratio.

[0125] As Figure 3 shown, for superimposed segment 1, the speech segment 2 on its left and the speech segment 3 on its right together contain seven speech frames. Therefore, the average energy of the second speech frame Ep1 corresponding to superimposed segment 1 can be calculated as = (the audio energy of speech segment 2 + the audio energy of speech segment 3) / (the number of speech frames of speech segment 2 + the number of speech frames of speech segment 3) = (the audio energy of speech segment 2 + the audio energy of speech segment 3) / 7. For superimposed segment 2, the speech segment 3 on its left contains four speech frames and there is no speech segment on its right. Therefore, the average energy of the second speech frame Ep2 corresponding to superimposed segment 2 can be calculated as = (the audio energy of speech segment 3) / (the number of speech frames of speech segment 3) = the audio energy of speech segment 3 / 4.

[0126] In step S403, determine the number of superimposed frames and calculate the speech energy in the superimposed frames.

[0127] For the above-mentioned average energy of the first speech frame, the audio processing device can determine the number of superimposed frames of all superimposed frames in the audio frame sequence, and then calculate the speech energy in the superimposed frames = the average energy of the first speech frame * the number of superimposed frames of all superimposed frames.

[0128] As Figure 3 shown, there are two overlapping segments in the target audio: overlapping segment 1 and overlapping segment 2, and the two together contain five overlapping frames. Therefore, the speech energy in the overlapping frames of the target audio = the average energy of the first speech frame * 5.

[0129] For the average energy of the second speech frames corresponding to each of the foregoing overlapping segments, the audio processing device can respectively determine the number of overlapping frames within each overlapping segment. Furthermore, the audio processing device can respectively calculate the speech energy in the overlapping frames within each overlapping segment. Furthermore, the audio processing device can calculate the where N is the number of overlapping segments in the target audio, Epj is the average energy of the second speech frames corresponding to the j-th overlapping segment, and nj is the number of overlapping frames in the j-th overlapping segment.

[0130] As Figure 3 shown, overlapping segment 1 contains three overlapping frames and overlapping segment 2 contains two overlapping frames. Therefore, the speech energy in the overlapping frames of the target audio can be calculated as = (the average energy of the second speech frames corresponding to overlapping segment 1 * the number of overlapping frames contained in overlapping segment 1) + (the average energy of the second speech frames corresponding to overlapping segment 2 * the number of overlapping frames contained in overlapping segment 2) = 3 * Ep1 + 2 * Ep2.

[0131] In step S404, the noise energy in the overlapping frames is calculated according to the audio energy of the overlapping frames.

[0132] The audio processing device can calculate the noise energy in the overlapping frames = the audio energy of the overlapping frames - the speech energy in the overlapping frames.

[0133] In step S405, the signal-to-noise ratio of the target audio is calculated.

[0134] In the case where the noise energy in the overlapping frames and the speech energy in the overlapping frames are determined, the audio processing device can calculate the signal-to-noise ratio of the target audio according to Equation (2), which will not be elaborated here.

[0135] Through the above method, this solution can improve the accuracy of active signal-to-noise ratio estimation for audio containing transient noise signals.

[0136] Exemplary Medium

[0137] After introducing the method of the exemplary embodiments of the present disclosure, next, reference is made to Figure 5 describe the media of the exemplary embodiments of the present disclosure.

[0138] In this exemplary embodiment, the above method can be implemented by a program product. For example, a portable compact disc read-only memory (CD-ROM) can be used and includes program code, and this memory can run on a device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, the readable medium 50 can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device.

[0139] The program product can adopt any combination of one or more readable media. The readable medium 50 can be a readable signal medium or a readable medium. The readable medium 50 can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0140] The computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium can also be any readable medium other than the readable medium, and this readable medium can send, propagate, or transmit a program used by or in combination with an instruction execution system, apparatus, or device.

[0141] The program code contained on the readable medium 50 can be transmitted by any suitable medium, including but not limited to wireless, wired, optical fiber cable, RE, etc., or any suitable combination of the above.

[0142] The program code for performing the operations of the present disclosure can be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages, such as Java, C++, etc., and also include conventional procedural programming languages, such as the C language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0143] Exemplary Apparatus

[0144] After introducing the media of the exemplary embodiments of the present disclosure, next, reference will be made to Figure 6 describe the devices of the exemplary embodiments of the present disclosure. Regarding the following devices, the specific manner in which each functional module performs operations and the specific functions achieved after performing the operations have been described in detail in the foregoing embodiments of the signal-to-noise ratio estimation method, and will not be elaborated herein.

[0145] Figure 6 A block diagram of a signal-to-noise ratio estimation device according to an embodiment of the present disclosure is schematically shown. The signal-to-noise ratio estimation device may include:

[0146] A sequence division module 601, configured to divide a target audio into an audio frame sequence, where the target audio is obtained by superimposing a reference speech and a reference transient noise in the time domain;

[0147] An audio frame determination module 602, configured to determine a speech frame and a superimposed frame in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; the speech frame includes a speech signal, and the superimposed frame includes a speech signal and a transient noise signal, and the duration of the transient noise signal is less than a duration threshold;

[0148] An energy determination module 603, configured to respectively determine the speech energy and the noise energy in the superimposed frame according to the audio energy of each of the speech frame and the superimposed frame; the audio energy of the superimposed frame is the sum of the speech energy and the noise energy in the superimposed frame;

[0149] A signal-to-noise ratio calculation module 604, configured to calculate the signal-to-noise ratio of the target audio based on the speech energy and the noise energy in the superimposed frame.

[0150] Optionally, the energy determination module 603 is further configured to:

[0151] Determine the speech energy of the speech frames in the audio frame sequence, and determine the average speech energy of the speech frames according to the number of speech frames;

[0152] Determine the speech energy in the superimposed frame according to the average speech energy of the speech frames and the number of superimposed frames of the superimposed frame.

[0153] Optionally, the energy determination module 603 is further configured to:

[0154] Determine the audio energy of the superimposed frame;

[0155] Use the difference between the audio energy of the superimposed frame and the speech energy in the superimposed frame as the noise energy in the superimposed frame.

[0156] Optionally, the energy determination module 603 is configured to determine the audio energy of any audio frame in the following manner:

[0157] Collect the signal energy of the corresponding audio signal at multiple sample points in the any audio frame;

[0158] Use the accumulated value of the collected multiple signal energies as the audio energy of the any audio frame.

[0159] Optionally, the sequence division module 601 is further configured to:

[0160] Divide the time axis of the target audio into multiple time intervals according to a preset duration, and the audio frames respectively corresponding to the respective time intervals form the audio frame sequence in the order of precedence in the time axis.

[0161] Optionally, the audio frame determination module 602 is further configured to:

[0162] Determine a speech time period, a noise time period, and an overlapping time period between the speech time period and the noise time period in the time axis of the target audio according to the timing information of the reference speech and the reference transient noise;

[0163] Determine the audio frames in the audio frame sequence that belong to the speech time period and do not belong to the overlapping time period as speech frames; and determine the audio frames in the audio frame sequence that belong to the overlapping time period as superimposed frames.

[0164] Exemplary Computing Device

[0165] After introducing the methods, media, and devices of the exemplary embodiments of the present disclosure, next, reference is made to Figure 7 Describe the computing device of the exemplary embodiments of the present disclosure.

[0166] Figure 7 The computing device 70 shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.

[0167] As Figure 7 shown, the computing device 7 of the present disclosure is embodied in the form of a general-purpose computing device. The components of the computing device 70 may include, but are not limited to: at least one of the above processing units 701, at least one of the above storage units 702, and a bus 703 connecting different system components (including the processing unit 701 and the storage unit 702).

[0168] The bus 703 includes a data bus, a control bus, and an address bus.

[0169] The storage unit 702 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 7021 and / or cache memory 7022, and may further include a readable medium in the form of non-volatile memory, such as read-only memory (ROM) 7023.

[0170] The storage unit 702 may also include a program / utilities 7025 having a set (at least one) of program modules 7024. Such program modules 7024 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment.

[0171] The computing device 70 may also communicate with one or more external devices 704 (such as a keyboard, a pointing device, etc.).

[0172] Such communication may be through an input / output (I / O) interface 705. Also, the computing device 70 may further communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 706. As Figure 7 shown, the network adapter 706 communicates with other modules of the computing device 70 through a bus 703. It should be understood that, although not shown in the figures, other hardware and / or software modules may be used in conjunction with the computing device 70, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0173] It should be noted that although several units / modules or sub-units / modules of the signal-to-noise ratio estimation device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-described units / modules may be embodied in one unit / module. Conversely, the features and functions of one unit / module described above may be further divided and embodied by multiple units / modules.

[0174] In addition, although the operations of the method of the present disclosure are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all of the shown operations must be performed to achieve the desired result. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution.

[0175] Although the spirit and principles of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the specific embodiments disclosed, and the division of each aspect does not mean that the features in these aspects cannot be combined for benefit. Such division is only for the convenience of expression. The present disclosure aims to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A signal-to-noise ratio estimation method, characterized in that, Including: Dividing the target audio into a sequence of audio frames, where the target audio is obtained by superposing a reference speech and a reference transient noise in the time domain; Determining speech frames and superposition frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; the speech frames contain speech signals, the superposition frames contain speech signals and transient noise signals, the duration of the transient noise signal is less than a duration threshold, and the duration of any superposition frame is less than the duration of a single-frame transient noise signal; Respectively determining the speech energy and noise energy in the superposition frames according to the audio energies of the speech frames and the superposition frames; the audio energy of the superposition frame is the sum of the speech energy and the noise energy in the superposition frame; Calculating the signal-to-noise ratio of the target audio based on the speech energy and noise energy in the superposition frames.

2. The method according to claim 1, wherein The determining the speech energy in the superposition frame according to the audio energies of the speech frames and the superposition frames includes: Determining the speech energy of the speech frames in the audio frame sequence and determining the average speech energy of the speech frames according to the number of speech frames; Determining the speech energy in the superposition frame according to the average speech energy of the speech frames and the number of superposition frames of the superposition frames.

3. The method according to claim 2, characterized in that The determining the noise energy in the superposition frame according to the audio energies of the speech frames and the superposition frames includes: Determining the audio energy of the superposition frame; Taking the difference between the audio energy of the superposition frame and the speech energy in the superposition frame as the noise energy in the superposition frame.

4. The method according to claim 2 or 3, characterized in that, Determining the audio energy of any audio frame includes: Collecting the signal energies of the corresponding audio signals at multiple sample points in the any audio frame; Taking the cumulative value of the collected multiple signal energies as the audio energy of the any audio frame.

5. The method according to claim 1, wherein The dividing the target audio into a sequence of audio frames includes: Dividing the time axis of the target audio into multiple time intervals according to a preset duration, and the audio frames corresponding to each time interval form the audio frame sequence in the order of precedence in the time axis.

6. The method according to claim 1, wherein The determining the speech frames and the superposition frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise includes: Determining a speech time period, a noise time period, and an overlapping time period between the speech time period and the noise time period in the time axis of the target audio according to the timing information of the reference speech and the reference transient noise; Determining the audio frames in the audio frame sequence that belong to the speech time period and do not belong to the overlapping time period as speech frames; and determining the audio frames in the audio frame sequence that belong to the overlapping time period as superposition frames.

7. A signal-to-noise ratio estimation device, characterized in that Including: A sequence division module, configured to divide a target audio into a sequence of audio frames, where the target audio is obtained by superposing a reference speech and a reference transient noise in the time domain; An audio frame determination module, configured to determine speech frames and superimposed frames in the audio frame sequence according to the timing information of the reference speech and the reference transient noise; the speech frames contain speech signals, the superimposed frames contain speech signals and transient noise signals, the duration of the transient noise signal is less than a duration threshold, and the duration of any superimposed frame is less than the duration of a single-frame transient noise signal; An energy determination module, configured to respectively determine the speech energy and the noise energy in the superimposed frames according to the audio energy of each of the speech frames and the superimposed frames; the audio energy of the superimposed frame is the sum of the speech energy and the noise energy in the superimposed frame; A signal-to-noise ratio calculation module, configured to calculate the signal-to-noise ratio of the target audio based on the speech energy and the noise energy in the superimposed frame.

8. The device according to claim 7, characterized in that, The energy determination module is further configured to: Determine the speech energy of the speech frames in the audio frame sequence, and determine the average speech energy of the speech frames according to the number of speech frames; Determine the speech energy in the superimposed frames according to the average speech energy of the speech frames and the number of superimposed frames of the superimposed frames.

9. The device according to claim 8, characterized in that, The energy determination module is further configured to: Determine the audio energy of the superimposed frame; Use the difference between the audio energy of the superimposed frame and the speech energy in the superimposed frame as the noise energy in the superimposed frame.

10. The device according to claim 8 or 9, characterized in that The energy determination module is configured to determine the audio energy of any audio frame in the following manner: Collect the signal energy of the corresponding audio signal at multiple sample points in the any audio frame; Use the accumulated value of the collected multiple signal energies as the audio energy of the any audio frame.

11. The device according to claim 7, characterized in that, The sequence division module is further configured to: Divide the time axis of the target audio into multiple time intervals according to a preset duration, and the audio frames corresponding to each time interval respectively form the audio frame sequence in the order of precedence in the time axis.

12. The device according to claim 7, characterized in that, The audio frame determination module is further configured to: According to the timing information of the reference speech and the reference transient noise, determine a speech time period and a noise time period in the time axis of the target audio, and the overlapping time period between the speech time period and the noise time period; Determine the audio frames in the audio frame sequence that belong to the speech time period and do not belong to the overlapping time period as speech frames; and determine the audio frames in the audio frame sequence that belong to the overlapping time period as superimposed frames.

13. A medium on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-6.

14. A computing device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor realizes the method according to any one of claims 1-6 by running the executable instructions.

Images