Voice processing device, method and equipment

By designing the reception, detection and filtering components in the voice processing device, converting the pulse density modulated signal with high sampling rate into a pulse coded modulated signal, the power consumption and delay problems caused by digital decimation filters in the prior art are solved, and the power consumption and delay problems are realized. More efficient voice interaction synchronization and real-time.

CN118982983APending Publication Date: 2024-11-19SMARTER SILICON (SHANGHAI) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411190453.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In the existing voice processing technology, digital decimation filters involve more modules in the signal processing process, resulting in large power consumption and delay, and cannot guarantee the synchronization of voice interaction.

Method used

Design a speech processing device, including a receiving component, a detection component and a first filtering component, reduces data by converting a high sampling rate pulse density modulated signal into a pulse coded modulated signal, and quickly wake up the voice processor when the wake-up condition is met Delay and power consumption.

Benefits of technology

It improves the speed of voice processing and the synchronization of interaction, reduces the power consumption of voice processing devices, and ensures the real-timeness of voice interaction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982983A_ABST
    Figure CN118982983A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a voice processing device, method and equipment, and the device at least comprises a receiving assembly which is used for receiving an audio signal, and converting the received audio signal into a pulse density modulation signal; the detection component is used for detecting whether the pulse density modulation signal meets a wake-up condition or not, and sending a wake-up signal to the voice processor under the condition that the wake-up condition is met; the first filtering component is used for converting the pulse density modulation signal into a first pulse code modulation signal; and the voice processor is used for obtaining the first pulse code modulation signal in response to the received wake-up signal, and performing voice processing on the first pulse code modulation signal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of speech conversion technology, and relate to but are not limited to a speech processing device, method and equipment. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent voice human-computer interaction is widely used in various smart devices. In the process of voice human-computer interaction, the voice signal is converted into a digital pulse code modulation signal (PCM) by an analog-to-digital converter (ADC). The digital PCM signal is sent to the hardware voice activity detection module to determine whether there is voice in the PCM signal.

[0003] The related technology realizes the conversion of voice signals through the digital extraction filter in the ADC. However, the digital extraction filter involves many modules in the signal processing process, resulting in large power consumption and delay, and the synchronization of voice interaction cannot be guaranteed. Summary of the invention

[0004] In order to solve the problems existing in the related art, the embodiments of the present application provide a speech processing device, method and equipment.

[0005] In a first aspect, the present application provides a speech processing device, the device comprising:

[0006] A receiving component, configured to receive an audio signal and convert the received audio signal into a pulse density modulation signal;

[0007] A detection component, used to detect whether the pulse density modulation signal meets the wake-up condition, and send a wake-up signal to the voice processor if the wake-up condition is met;

[0008] A first filtering component, configured to convert the pulse density modulation signal into a first pulse code modulation signal;

[0009] The voice processor is used for acquiring the first pulse code modulation signal in response to receiving the wake-up signal, and performing voice processing on the first pulse code modulation signal.

[0010] In some embodiments, the detection component includes a second filtering component and a detection unit;

[0011] A second filtering component, configured to convert the pulse density modulation signal into a second pulse code modulation signal; wherein the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal;

[0012] The detection unit is used to detect whether the second pulse code modulation signal meets the wake-up condition, and send a wake-up signal to the voice processor if the wake-up condition is met.

[0013] In some embodiments, the device further comprises a buffer;

[0014] The first filtering component is further used to send the first pulse code modulation signal to the buffer;

[0015] The buffer is used for buffering the first pulse code modulation signal.

[0016] In some embodiments, the speech processor is further configured to send a start instruction to the first filter component in response to receiving the wake-up signal;

[0017] The first filtering component is further configured to receive a pulse density modulation signal sent by the receiving component in response to the start instruction;

[0018] The speech processor is further used to obtain the first pulse code modulation signal from the buffer and perform speech processing on the first pulse code modulation signal.

[0019] In some embodiments, the speech processor is further configured to:

[0020] In response to receiving the wake-up signal, a target first pulse code modulation signal is acquired from the buffer based on the time when the wake-up signal is received, and speech processing is performed on the target first pulse code modulation signal.

[0021] In some embodiments, the second filtering component includes a second cascaded integrator comb decimation filter, and the first filtering component includes a first cascaded integrator comb decimation filter and a signal optimizer;

[0022] The first cascaded integrator comb decimation filter is used to perform a first data extraction on the pulse density modulation signal sent by the receiving component to obtain an initial first pulse code modulation signal having a signal-to-noise ratio lower than that of the pulse density modulation signal, and the signal optimizer is used to perform signal optimization on the initial first pulse code modulation signal sent by the first cascaded integrator comb decimation filter to obtain the first pulse code modulation signal;

[0023] The second cascaded integrator comb extraction filter is used to perform a second data extraction on the pulse density modulation signal sent by the receiving component to obtain the second pulse code modulation signal with a signal-to-noise ratio lower than that of the pulse density modulation signal.

[0024] In some embodiments, the signal optimizer includes a cascaded integrator comb compensation filter and a finite impulse response filter;

[0025] The cascaded integrator comb compensation filter is used to compensate for the signal loss in the initial first pulse code modulation signal sent by the first cascaded integrator comb extraction filter to obtain a first compensation signal, and the finite impulse response filter is used to perform low-pass filtering on the first compensation signal sent by the first cascaded integrator comb extraction filter to obtain the first pulse code modulation signal.

[0026] In a second aspect, an embodiment of the present application provides a speech processing method, the method comprising:

[0027] The receiving component receives the audio signal and converts the received audio signal into a pulse density modulation signal;

[0028] The detection component sends a wake-up signal to wake up the speech processor in response to detecting that the pulse density modulation signal meets the wake-up condition;

[0029] A first filtering component converts the pulse density modulation signal into a first pulse code modulation signal;

[0030] In response to receiving the wake-up signal, the voice processor acquires the first pulse code modulation signal and performs voice processing on the first pulse code modulation signal.

[0031] In some embodiments, the detection component includes a second filtering component and a detection unit; the detection component sends a wake-up signal to wake up the voice processor in response to detecting that the pulse density modulation signal meets the wake-up condition, including:

[0032] The second filtering component converts the pulse density modulation signal into a second pulse code modulation signal; wherein the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal;

[0033] The detection unit detects whether the second pulse code modulation signal meets a wake-up condition, and sends a wake-up signal to the speech processor if the wake-up condition is met.

[0034] In a third aspect, an embodiment of the present application provides a speech processing device, comprising a speech processing apparatus; the speech processing apparatus comprises: a receiving component for receiving an audio signal and converting the received audio signal into a pulse density modulation signal; a detection component for detecting whether the pulse density modulation signal satisfies a wake-up condition, and sending a wake-up signal to a speech processor if the wake-up condition is met; a first filtering component for converting the pulse density modulation signal into a first pulse code modulation signal; and the speech processor for acquiring the first pulse code modulation signal in response to receiving the wake-up signal, and performing speech processing on the first pulse code modulation signal.

[0035] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a signal processing schematic diagram of the intelligent voice human-computer interaction process provided by an embodiment of the present application;

[0037] Figure 2 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 1 ;

[0038] Figure 3 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 2 ;

[0039] Figure 4 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 3 ;

[0040] Figure 5 This is an optional flowchart of the speech processing method provided in the embodiment of the present application;

[0041] Figure 6 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 4 ;

[0042] Figure 7 This is a schematic diagram of the structure of the voice processing device provided in the embodiment of the present application. Figure 1 ;

[0043] Figure 8 This is a schematic diagram of the structure of the voice processing device provided in the embodiment of the present application. Figure 2 . DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0045] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the technical field of the embodiments of the present application. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0046] Figure 1 is a signal processing schematic diagram of the intelligent voice human-computer interaction process provided by the embodiment of the present application, such as Figure 1 As shown, one of the current means of intelligent voice human-computer interaction is that the voice signal (voice signal) passes through an analog-to-digital converter 101 (ADC), and ADC101 is used to convert the analog voice signal into a digital PCM signal. The digital PCM signal is sent to a voice activity detection module 102 (HW-VAD, Hardware-based Voice Activity Detection), and HW-VAD102 determines whether there is voice in the PCM signal. If there is voice, the keyword recognition module 103 (Keyword Spotting) is awakened to perform keyword recognition. In the case of the presence of keywords, speech recognition is further performed through a speech recognition module 104 (Speech Recognition) and natural speech processing is performed through a natural language processing module 105 (NLP, Natural Language Processing).

[0047] In some implementations, the conversion of a pulse density modulation (PDM) signal into a PCM signal in the analog-to-digital converter 101 is implemented through a digital decimation filter module 1012 (Decimation filter). However, when a digital decimation filter having a cascaded integrator-comb (CIC) filter, a CIC compensation filter, and a finite impulse response (FIR) filter processes a voice signal, more modules are involved, resulting in greater power consumption and delay, and the synchronization of voice interaction cannot be guaranteed.

[0048] Based on the problems existing in the related technology, an embodiment of the present application provides a speech processing device, wherein a receiving component receives an audio signal and converts the received audio signal into a PDM signal; a detection component detects whether the signal meets the wake-up condition, and sends a wake-up signal to a speech processor if the wake-up condition is met; a first filtering component converts the PDM signal into a PCM signal; in response to receiving the wake-up signal, the speech processor obtains the PCM signal converted by the first filtering component, and performs speech processing on the PCM signal, such as keyword recognition, feature extraction, and other processing.

[0049] In this way, the high-sampling-rate PDM signal passes through the detection component to detect whether the signal meets the wake-up condition. If the wake-up condition is met, the corresponding PDM is converted into PCM through the first filtering component. At this time, the embodiment of the present application only needs to convert the PDM signal that meets the wake-up condition (for example, a PDM signal with voice) into a PCM signal, which improves the voice processing rate and the synchronization of voice interaction. At the same time, the detection component in the embodiment of the present application directly detects whether the signal meets the wake-up condition, without waiting for the digital extraction filter to process the PDM signal before detecting, which reduces data delay, can quickly wake up the voice processor, and reduces the power consumption of the voice processing device.

[0050] In some embodiments, the speech processing device provided by the present application can be set in a speech processing device to realize human-computer interaction of intelligent speech based on the speech processing device. The speech processing device can be a physical device independent of the terminal, placed in a home, office, vehicle or other place, and establishes a connection with the terminal through short-range communication methods such as Bluetooth, infrared, and wired. The terminal performs human-computer interaction through the digital signal output by the speech processing device and performs corresponding operations, such as controlling a smart home; the speech processing device can also be any smart device such as a mobile phone, a laptop, a tablet computer, a TV, etc.

[0051] Here, the voice processing device can be a mobile phone terminal, which can collect the user's voice signal, recognize the collected voice signal, and perform corresponding operations according to the recognition result, such as playing music, reading short messages, and recording. When the voice processing device is inside the car, the voice processing method provided in the embodiment of the present application can be used to enable functions such as navigation, music playback, and radio station control.

[0052] Figure 2 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 1 ,like Figure 2As shown, the speech processing device at least includes a receiving component 201, a detection component 202, a first filtering component 203 and a speech processor 204, wherein the receiving component 201 is used to receive an audio signal and convert the received audio signal into a pulse density modulation signal; the detection component 202 is used to detect whether the pulse density modulation signal meets the wake-up condition, and if the wake-up condition is met, send a wake-up signal to the speech processor; the first filtering component 203 is used to convert the pulse density modulation signal into a first pulse code modulation signal; the speech processor 204 is used to obtain the first pulse code modulation signal in response to receiving the wake-up signal, and perform speech processing on the first pulse code modulation signal.

[0053] Here, the receiving component 201 can be an analog-to-digital converter (e.g., Sigma-Delta ADC), which converts the audio signal into a digital signal, i.e., a pulse density modulation signal (i.e., PDM), after receiving the audio signal input by the user through a microphone. The receiving component 201 can continuously receive the audio signal, convert it into a digital signal and send it to the detection component, that is, the PDM signal sent by the receiving component can be continuous.

[0054] In some embodiments, the detection component 202 is used to detect whether there is voice in the signal output by the receiving component 201. The detection component 202 may include a signal conversion component and a VAD. The signal conversion component converts the PDM signal into a PCM signal. The VAD can distinguish between voice segments, silence or noise segments in the converted PCM signal to detect whether there is voice in the signal output by the receiving component 201, that is, whether the signal meets the wake-up condition. When the wake-up condition is met, a wake-up signal is sent to the voice processor. Here, the wake-up signal can be an interrupt signal int, which is used to indicate that voice has been detected in the PDM signal. Here, the wake-up signal can also include the position where the voice appears in the signal, so that the voice processor obtains the first pulse code modulation signal after the position for voice processing to reduce the amount of data processing.

[0055] Here, since PDM uses binary numbers 0 and 1 to represent analog signals, VAD cannot directly determine whether there is voice in the PDM signal. Therefore, the detection component 202 includes a signal conversion component that converts the PDM signal into a PCM signal. The signal conversion component can be a chip that realizes PDM to PCM conversion, or a component such as a digital signal processor (DSP, Digital Signal Processing) or a CIC filter.

[0056] In some embodiments, the first filter component 203 can be any filter such as a digital decimation filter that can realize PDM to PCM conversion. In some embodiments, the first filter component 203 can include a CIC filter, a CIC compensator and a FIR filter to realize PDM to PCM conversion. Among them, the CIC filter is used to downsample the PDM signal, and can maintain the spectrum information of the input PDM signal while reducing the sampling rate; the CIC compensator is used to adjust the frequency response distortion introduced by the CIC filter, that is, to compensate for the signal loss caused by the CIC filter, and compensate for the frequency response drop caused by the CIC filter; the FIR filter filters the impulse response value of the output signal of the CIC compensator, and further downsamples to obtain a first pulse code modulation signal, that is, a PCM signal.

[0057] like Figure 2 As shown, the first filter component 203 and the detection component 202 are parallel structures, that is, the first filter component 203 and the detection component 202 can run in parallel, or the detection component 202 detects that there is voice in the PDM signal, and after sending a wake-up signal to the voice processor 204, the voice processor 204 can send an opening instruction to the first filter component 203, and the opening instruction can include the position where the voice appears in the PDM signal. The first filter component 203 starts to run in response to the opening instruction and performs downsampling filtering on the PDM signal. Here, the first filter component 203 can be based on the position where the voice appears, and downsamples and filters the PDM signal after the position to obtain a PCM signal, and send the PCM signal to the voice processor 204.

[0058] In some embodiments, the speech processor 204 may be a digital signal processor (DSP). After the DSP obtains the first pulse code modulation signal, it may perform low bit rate speech encoding on the input first pulse code modulation signal to achieve more efficient transmission and storage. The DSP may also filter the first pulse code modulation signal, remove noise and interference, and perform operations such as feature extraction and speech recognition. Finally, the signal processed by the speech processor 204 may be converted into an analog audio signal through a digital-to-analog converter (DAC), and then output to a speaker or earphone through an amplification circuit; human-computer interaction may also be performed based on the signal processed by the speech processor 204, for example, the speech processor 204 performs keyword extraction, and the speech processing device performs corresponding operations based on the keyword, for example. Manage the smart home connected to the speech processing device, or determine the user's question based on the keyword, retrieve and determine the answer to inform the user through the speaker.

[0059] In an embodiment of the present application, a PDM signal with a high sampling rate is detected by a detection component to determine whether the signal meets the wake-up condition. If the wake-up condition is met, the corresponding PDM is converted into PCM through a first filtering component. At this time, the embodiment of the present application only needs to convert the PDM signal that meets the wake-up condition (for example, a PDM signal with voice) into a PCM signal, thereby improving the rate of voice processing and the synchronization of voice interaction. At the same time, the detection component in the embodiment of the present application directly detects whether the signal meets the wake-up condition without waiting for the digital extraction filter to process the PDM signal before detecting, thereby reducing data delay, being able to quickly wake up the voice processor, and reducing power consumption of the voice processing device.

[0060] In some embodiments, Figure 3 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 2 ,like Figure 3 As shown, the detection component 202 may include a second filtering component 301 and a detection unit 302. The second filtering component 301 is used to convert the pulse density modulation signal into a second pulse code modulation signal for subsequent detection; the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal; the detection unit 302 is used to detect whether the second pulse code modulation signal meets the wake-up condition, and if the wake-up condition is met, send a wake-up signal to the voice processor.

[0061] Here, the second filter component 301 may be a decimation filter including only a CIC filter, which converts a high sampling rate PDM signal into a second pulse code modulation signal with a low sampling rate. Compared with PCM conversion by the first filter component, the second filter component 301 enables the PCM signal to be sent to the detection unit 302 more quickly, and the detection unit 302 determines whether there is speech in the received audio signal based on the second pulse code modulation signal. Here, the detection unit 302 may be a voice activity detection module (VAD), which can identify whether there is a speech component in the audio signal and distinguish a speech segment from a non-speech segment (such as silence or noise).

[0062] It should be noted that, since the second filter component 301 only includes a CIC filter, although the conversion speed is improved during conversion, the audio quality is sacrificed, so that the second pulse code modulation signal converted by the second filter component 301 has a lower signal-to-noise ratio than the first pulse code modulation signal converted by the first filter component.

[0063] In the embodiment of the present application, since the detection component only includes a CIC filter, it has fewer structures than the first filter component and can quickly realize the conversion between the PDM signal and the PCM signal, so that after receiving the audio signal, it can be quickly converted into a PCM signal and detect whether there is voice through the detection component, and the audio signal with detected voice is converted through the first filter component, so that the PCM signal obtained by the voice processor is realized through the first filter component, and the PCM signal obtained by the voice processor not only reduces the sampling rate, but also ensures the signal-to-noise ratio of the converted signal, so that the PCM signal obtained by the voice processor does not lose important information, thereby improving the accuracy of voice processing; at the same time, the embodiment of the present application does not need to convert each audio signal through the first filter component, thereby reducing the amount of calculation and the power consumption of the voice processing device; and the PCM signal detected by the detection component is converted through the second filter component, and the modules in the second filter component are fewer than those in the first filter component, so that the conversion of the second filter component is faster, the delay is reduced, and the real-time nature of human-computer interaction is improved.

[0064] In some embodiments, Figure 4 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 3 ,like Figure 4 As shown, the speech processing device may further include a buffer 401 for buffering the first pulse code modulation signal converted by the first filtering component.

[0065] In some embodiments, the buffer 401 may be a buffer using a First Input First Output (FIFO) algorithm, and the first PCM signal that enters is read out from the FIFO buffer first. The buffer 401 may also be a buffer that is read and written using other algorithms, which is not limited by the embodiments of the application.

[0066] Here, the path of the first filtering component and the path of the detection component can be carried out simultaneously, that is, after the receiving component converts the audio signal into a PDM signal, the first filtering component can convert the PDM signal in real time to obtain a first pulse code modulation signal, and send the obtained first pulse code modulation signal to the FIFO buffer for storage. When the voice processor receives the wake-up signal, it can directly obtain the first pulse code modulation signal of the detected voice in the FIFO buffer for subsequent feature extraction or voice recognition.

[0067] Here, when the voice processor obtains the first pulse code modulation signal of the detected voice from the FIFO buffer, it can be obtained based on the time when the detection component sends the wake-up signal. The time of sending the wake-up signal is the time when the voice is detected in the signal. Since the path of the first filtering component and the path of the detection component can be carried out simultaneously, the voice processor can determine the corresponding first pulse code modulation signal in the FIFO buffer according to the time when the wake-up signal is sent. For example, the time of sending the wake-up signal is a certain moment, and the voice processor can obtain the first pulse code modulation signal that comes in after the moment in the FIFO buffer for processing, or obtain the first pulse code modulation signal within a time period (for example, one minute) after the moment. The embodiment of the present application does not limit this.

[0068] In an embodiment of the present application, the path of the first filter component and the path of the detection component may not be performed simultaneously. That is, when the detection component detects that there is voice in the second pulse code modulation signal output by the second filter component, it sends a wake-up signal to the voice processor. At this time, the path corresponding to the first filter component starts to convert the PDM signal to obtain a first pulse code modulation signal. At this time, the PDM signal before the conversion is a signal with voice detected by the detection component. Therefore, the voice processor can directly obtain the first pulse code modulation signal with voice after being processed by the first filter component, thereby improving the speed of voice processing.

[0069] In some embodiments, based on Figure 4 The voice processor 204 is also used to send a start instruction to the first filter component 203 in response to receiving a wake-up signal from the detection unit 302. The first filter component 203 is also used to receive a pulse density modulation signal sent by the receiving component in response to the start instruction. The voice processor 204 is also used to obtain a first pulse code modulation signal from the buffer and perform voice processing on the first pulse code modulation signal.

[0070] Here, the start instruction sent by the voice processor can be used to instruct the first filter component to receive and convert the PDM signal, or to turn on the power of the first filter component to open the path corresponding to the first filter component.

[0071] In some embodiments, when the voice processor receives the wake-up signal from the detection unit 302, that is, when it detects the presence of voice, it sends an on instruction to the first filter component 203, so that the path corresponding to the first filter component 203 starts working, converts the PDM converted by the receiving component, obtains a first pulse code modulation signal, and sends the first pulse code modulation signal to the buffer.

[0072] In an embodiment of the present application, the path where the first filter component is located and the path where the detection component is located may not operate synchronously. When the detection component detects voice and sends a wake-up signal to the voice processor, the voice processor sends a start instruction to the first filter component. At this time, the first filter component starts to process the PDM signal with voice. In other words, the first filter component does not need to process all PDM signals, but only processes the PDM signal with voice, which reduces the amount of data processed and reduces the power consumption of the voice processing device.

[0073] In some embodiments, a plurality of first pulse code modulation signals may be stored in the buffer, so the speech processor may also be used to obtain a target first pulse code modulation signal from the buffer based on the time when the wake-up signal is received in response to receiving the wake-up signal, and perform speech processing on the target first pulse code modulation signal. Here, the target first pulse code modulation signal is a PCM signal with speech.

[0074] Here, when the path of the first filter component and the path of the detection component are parallel, for the continuously received audio signal, the first filter component converts the PDM signal output by the receiving component in real time, continuously outputs the first pulse code modulation signal, and sends the first pulse code modulation signal to the buffer, and the voice processor can obtain the corresponding first pulse code modulation signal in the buffer for voice recognition according to the time when the wake-up signal is received (i.e., the time when the voice appears). Here, the time of sending the wake-up signal is a certain moment, and the voice processor obtains the first pulse code modulation signal based on the time, which can be to obtain the PCM signal in the FIFO buffer after that moment for processing, or to obtain the PCM signal within a time period (e.g., one minute) after that moment.

[0075] Here, when the path of the first filter component and the path of the detection component are non-parallel, when the detection component detects voice and sends a wake-up signal to the voice processor, the voice processor sends a start instruction to the first filter component. At this time, the first filter component starts to process the current PDM signal and outputs a first pulse code modulation signal. At this time, the first pulse code modulation signal is a signal in the presence of voice, and the voice processor directly obtains the first pulse code modulation signal for voice processing.

[0076] In some embodiments, the first filtering component includes a first cascaded integrator comb extraction filter and a signal optimizer; the first cascaded integrator comb extraction filter is used to perform a first data extraction on the pulse density modulated signal sent by the receiving component to obtain an initial first pulse code modulation signal having a signal-to-noise ratio lower than that of the pulse density modulated signal, and the signal optimizer is used to perform signal optimization on the initial first pulse code modulation signal sent by the first cascaded integrator comb extraction filter to obtain a first pulse code modulation signal.

[0077] Here, the first cascaded integrator comb extraction filter can be a CIC filter, which performs a first data extraction on the pulse density modulation signal sent by the receiving component to obtain an initial first pulse code modulation signal with a signal-to-noise ratio lower than that of the pulse density modulation signal. Here, the first data extraction can be a multi-stage extraction to reduce the sampling rate of the PDM signal, and each stage of the multi-stage extraction will reduce the data amount of the PDM signal, and multi-stage filtering can be implemented, thereby improving the quality of the output PCM signal.

[0078] After completing CIC filtering, data type conversion can also be performed, such as converting the processed signal from single-precision floating point numbers to double-precision floating point numbers or other required data types to ensure data accuracy and compatibility.

[0079] Here, the multi-stage extraction of the CIC filter has the problem of passband attenuation, which makes the signal amplitude-frequency response in the passband not meet the requirements. Therefore, the signal output by the CIC filter can be optimized by a signal optimizer to compensate for the passband attenuation caused by the CIC filter to achieve a smooth low-ripple passband. The signal optimizer can be any compensator that can compensate for the passband attenuation.

[0080] The embodiment of the present application realizes data extraction and avoids passband attenuation caused by data extraction through a CIC filter and a signal compensator. While realizing downsampling, it makes the passband response smoother, reduces the frequency response distortion caused by the CIC filter, and improves the signal quality of the first pulse code modulation signal.

[0081] In some embodiments, the signal optimizer includes a cascaded integrator comb compensation filter and a finite impulse response filter; the cascaded integrator comb compensation filter is used to compensate for the signal loss in the initial first pulse code modulation signal sent by the first cascaded integrator comb extraction filter to obtain a first compensation signal, and the finite impulse response filter is used to perform low-pass filtering on the first compensation signal sent by the first cascaded integrator comb extraction filter to obtain a first pulse code modulation signal.

[0082] Here, the cascaded integrator comb compensation filter in the signal optimizer may be a CIC compensation filter, and the finite impulse response filter may be a FIR filter. It may also be other filters and compensators that can achieve compensation and filtering, and this application does not limit this.

[0083] Here, the CIC compensation filter can compensate for the passband attenuation caused by the CIC filter and compensate for the signal loss in the initial first pulse code modulation signal sent by the CIC filter to achieve a relatively smooth low-ripple passband; the FIR filter can perform low-pass filtering on the signal output by the CIC compensation filter, remove noise, and extract frequency components of interest, thereby achieving signal enhancement and reconstruction.

[0084] The embodiment of the present application cascades the CIC compensation filter and the FIR filter, which can effectively compensate for the passband attenuation problem of the CIC filter, and at the same time utilize the high precision and stability of the FIR filter to improve the filtering effect of the entire system.

[0085] In some embodiments, the second filtering component may include a second cascaded integrator comb extraction filter, which is used to perform a second data extraction on the pulse density modulated signal sent by the receiving component to obtain a second pulse code modulation signal with a signal-to-noise ratio lower than the pulse density modulation signal.

[0086] Here, the second cascaded integrator comb decimation filter may be a CIC filter. Here, the second cascaded integrator comb decimation filter may be the same as or different from the first cascaded integrator comb decimation filter. The second data extraction may also be a multi-stage extraction to obtain a second pulse code modulation signal.

[0087] Here, since the second filtering component only includes a CIC filter and there is no signal optimizer, the second pulse code modulation signal is not subjected to signal optimization processing, so that the signal-to-noise ratio of the first pulse code modulation signal is higher than that of the second pulse code modulation signal.

[0088] In the embodiment of the present application, the second filter component only includes a CIC filter, and the first filter component has fewer modules, so that the second filter component converts the PDM signal more quickly, reduces delays, and improves the real-time performance of human-computer interaction.

[0089] Figure 5 is an optional flow chart of the speech processing method provided in the embodiment of the present application. The execution subject of the speech processing method provided in the embodiment of the present application is a speech processing device, such as Figure 5 As shown, the speech processing method provided in the embodiment of the present application can be implemented through steps S501 to S504:

[0090] Step S501: The receiving component receives an audio signal and converts the received audio signal into a pulse density modulation signal.

[0091] In an embodiment of the present application, the audio signal can be acquired based on a microphone of a voice processing device. The microphone continuously acquires the audio signal and sends it to a receiving component. The receiving component receives the audio signal in real time and converts the audio signal to generate a PDM signal.

[0092] Step S502: In response to detecting that the pulse density modulation signal meets the wake-up condition, the detection component sends a wake-up signal to wake up the voice processor.

[0093] Here, the detection component detects whether there is voice in the pulse density modulation signal. If there is voice, the wake-up condition is met, and a wake-up signal is sent to the voice processor, so that the voice processor starts to obtain the first pulse code modulation signal for voice recognition and other processing.

[0094] Step S503: The first filtering component converts the pulse density modulation signal into a first pulse code modulation signal.

[0095] In the embodiment of the present application, the first pulse code modulation signal may be a PCM signal, and the first filter component converts the PDM signal into a PCM signal. Here, the first filter component may be any filter that realizes the conversion of PDM into PCM.

[0096] In some embodiments, the first filtering component and the detection component can be operated synchronously, that is, when the receiving component converts the audio signal into a PDM signal, the first filtering component can synchronously convert the PDM signal to obtain a first pulse code modulation signal, and send the obtained first pulse code modulation signal to the FIFO buffer for storage. When the voice processor receives the wake-up signal, it can directly obtain the first pulse code modulation signal of the detected voice in the FIFO buffer for subsequent feature extraction or voice recognition.

[0097] In some embodiments, the first filtering component and the detection component may also be operated asynchronously, that is, when the detection component detects the presence of voice in the PDM signal and sends a wake-up signal to the voice processor, the voice processor sends a start instruction to the first filtering component. At this time, the first filtering component starts to process the current PDM signal and outputs a first pulse code modulation signal. At this time, the first pulse code modulation signal is a signal indicating the presence of voice, and the voice processor directly obtains the first pulse code modulation signal for voice processing.

[0098] Step S504: In response to receiving the wake-up signal, the voice processor acquires the first pulse code modulation signal and performs voice processing on the first pulse code modulation signal.

[0099] Here, the voice processor can obtain the corresponding first pulse code modulation signal in the buffer for voice recognition according to the time when the wake-up signal is received (i.e., the time when the voice appears in the PDM signal). For example, the time when the wake-up signal is sent is a certain moment, and the voice processor obtains the first pulse code modulation signal based on the time, which can be a PCM signal entering the FIFO buffer after that moment, or a PCM signal within a time period (e.g., one minute) after that moment, for voice processing.

[0100] In some embodiments, the speech processor can perform low-bit-rate speech encoding on the input PCM signal to achieve more efficient transmission and storage. The PCM signal can also be filtered to remove noise and interference, and perform operations such as feature extraction and speech recognition. The speech-processed signal can be converted into an analog audio signal through a digital-to-analog converter (DAC) and then output to a speaker or earphone through an amplification circuit; human-computer interaction can also be performed based on the recognized speech.

[0101] It should be noted that steps S501 to S504 are not intended to limit the order of implementation of the steps, and the order of the steps can be adjusted if feasible.

[0102] In the embodiment of the present application, a PDM signal with a high sampling rate is detected by a detection component to detect whether the PDM signal meets the wake-up condition. If the wake-up condition is met, the corresponding PDM is converted into PCM through the first filtering component. At this time, the embodiment of the present application only needs to convert the PDM that meets the wake-up condition into a PCM signal, and there is no need to convert each voice signal into a PCM signal, which reduces the amount of data, increases the rate of voice processing, and improves the synchronization of voice interaction. At the same time, the detection component in the embodiment of the present application directly detects whether the PDM signal meets the wake-up condition, without waiting for the digital extraction filter to process the PDM signal, reducing data delay, being able to quickly wake up the voice processor, and reducing power consumption.

[0103] In some embodiments, the detection component includes a second filtering component and a detection unit; step S502 may include: the second filtering component converts the pulse density modulation signal into a second pulse code modulation signal; wherein the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal; the detection unit detects whether the second pulse code modulation signal meets the wake-up condition, and if the wake-up condition is met, sends a wake-up signal to the voice processor.

[0104] In some embodiments, the speech processing device further includes a buffer, and the method further includes: the first filtering component sends the first pulse code modulation signal to the buffer; and the buffer buffers the first pulse code modulation signal.

[0105] In some embodiments, the method also includes: the voice processor sends a start instruction to the first filter component in response to receiving a wake-up signal; the first filter component receives the pulse density modulation signal sent by the receiving component in response to the start instruction; the voice processor obtains the first pulse code modulation signal from the buffer and performs voice processing on the first pulse code modulation signal.

[0106] In some embodiments, the method further includes: in response to receiving the wake-up signal, the voice processor obtains the target first pulse code modulation signal from the buffer based on the time when the wake-up signal is received, and performs voice processing on the target first pulse code modulation signal.

[0107] In some embodiments, the second filtering component includes a second cascaded integral comb decimation filter, and the first filtering component includes a first cascaded integral comb decimation filter and a signal optimizer; the first filtering component converts the pulse density modulated signal into a first pulse code modulated signal, including: the first cascaded integral comb decimation filter performs a first data extraction on the pulse density modulated signal sent by the receiving component to obtain an initial first pulse code modulated signal with a signal-to-noise ratio lower than the pulse density modulated signal, and the signal optimizer performs signal optimization on the initial first pulse code modulated signal sent by the first cascaded integral comb decimation filter to obtain the first pulse code modulated signal; the second filtering component converts the pulse density modulated signal into a second pulse code modulated signal, including: the second cascaded integral comb decimation filter performs a second data extraction on the pulse density modulated signal sent by the receiving component to obtain a second pulse code modulated signal with a signal-to-noise ratio lower than the pulse density modulated signal.

[0108] In some embodiments, the signal optimizer includes a cascaded integrator comb compensation filter and a finite impulse response filter; the signal optimizer performs signal optimization on the initial first pulse code modulation signal sent by the first cascaded integrator comb decimation filter to obtain a first pulse code modulation signal, including: the cascaded integrator comb compensation filter compensates for the signal loss in the initial first pulse code modulation signal sent by the first cascaded integrator comb decimation filter to obtain a first compensation signal, and the finite impulse response filter performs low-pass filtering on the first compensation signal sent by the first cascaded integrator comb decimation filter to obtain the first pulse code modulation signal.

[0109] It should be noted that the description of the speech processing method in the embodiment of the present application is similar to the description of the above-mentioned device embodiment, and has similar beneficial effects as the device embodiment, so it is not repeated. For technical details not disclosed in the embodiment of the speech processing method, please refer to the description of the device embodiment of the present application for understanding.

[0110] Next, an application of a speech processing device in a practical scenario is provided.

[0111] Digital decimation filters generally include CIC filters, CIC compensation filters, and FIR filters. Among them, the FIR filter can be a half-band filter, a general FIR filter, etc. In a low-power scenario, the PDM signal is filtered through a CIC filter, CIC compensation, and FIR filter in the digital decimation filter module, and the filtered signal is sent to VAD. At this time, the conversion of the PDM signal to a PCM signal consumes a lot of power, and there are many modules in the digital decimation filter, which occupies a large circuit area; at the same time, due to the large number of data processing processes, the delay is large, and there is a delay in VAD receiving the converted PCM signal, resulting in a slow interrupt (ie, wake-up signal) after detecting the voice, resulting in insufficient real-time voice human-computer interaction.

[0112] The embodiment of the present application proposes a simplified digital decimation filter module, which only has a CIC filter and can convert PDM signals into PCM signals more quickly, and is used for VAD to detect whether there is voice in the PCM signal. If there is voice, the high signal-to-noise ratio PCM signal output by the digital decimation filter including the CIC filter, CIC compensation filter and FIR filter is taken out from the FIFO (i.e., buffer) for voice wake-up (i.e., keyword recognition). In this way, the digital decimation filter module used for VAD to perform voice recognition only includes a CIC filter, which can quickly convert PDM signals into PCM signals and send them to VAD, with small delay and faster VAD interrupt startup.

[0113] Figure 6 This is an optional structural diagram of the speech processing device provided in the embodiment of the present application. Figure 4 ,like Figure 6 As shown, the second down-sampling filter 602 only includes a CIC filter, which is used to convert a high-sampling-rate PDM signal into a low-sampling-rate PCM signal, and send it to the voice activity detection module 603 more quickly. The voice activity detection module 603 can be a VAD module. The VAD determines whether there is voice in the PCM signal. If there is voice, it sends an int interrupt (i.e., a wake-up signal) to the signal processor 605. The signal processor 605 can be a DSP. The DSP obtains the PCM signal in the FIFO buffer 604. The PC M signal here is obtained by converting the PDM signal by the first down-sampling filter 601 and stored in the FIFO buffer 604.

[0114] In the embodiment of the present application, path 1 and path 2 may run in parallel or in non-parallel.

[0115] In the case of parallel execution of paths 1 and 2, the first down-sampling filter 601 and the second down-sampling filter 602 simultaneously convert the PDM signal to obtain PCM1 and PCM2 signals respectively, the PCM1 signal is sent to the FIFO buffer 604 for storage, the PCM2 signal is sent to the voice activity detection module 603 for voice activity detection, and when the PCM 2 signal is detected to have voice, an int interrupt is sent to the signal processor 605, and the signal processor 605 obtains a PCM signal with a higher signal-to-noise ratio from the FIFO buffer. Here, after the signal processor 605 is awakened by the int interrupt, since the FIFO buffer may store irrelevant PCM signals (for example, PCM signals without voice that were previously processed), the required PCM signals can be obtained from the FIFO buffer based on the wake-up time.

[0116] In the case of non-parallel operation, path 2 can be executed first, and the PCM2 signal is sent to the voice activity detection module 603 for voice activity detection. When it is detected that there is voice in the PCM2 signal, an int interrupt is sent to the signal processor 605, and the signal processor 605 instructs to start executing path 1. At this time, a PCM signal with a higher signal-to-noise ratio is obtained, and the signal has voice.

[0117] Figure 7 This is a schematic diagram of the structure of the voice processing device provided in the embodiment of the present application. Figure 1 , Figure 7 The speech processing device 700 shown includes any one of the above-mentioned speech processing means 701 .

[0118] Figure 8 This is a schematic diagram of the structure of the voice processing device provided in the embodiment of the present application. Figure 2 , Figure 8 Another structure of a speech processing device is provided. The speech processing device 800 may include: at least one processor 801, a speech processing device 802, and a user interface 803. The various components in the speech processing device are coupled together via a bus system 804. It is understood that the bus system 804 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 804 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 804 is not described in detail. Figure 8 Various buses are labeled as bus system 804 .

[0119] The voice processing device 802 is used to receive an audio signal and convert the received audio signal into a pulse density modulation signal; detect whether the pulse density modulation signal meets the wake-up condition, and if the wake-up condition is met, send a wake-up signal to the voice processor; convert the pulse density modulation signal into a first pulse code modulation signal; in response to receiving the wake-up signal, the voice processor obtains the first pulse code modulation signal, and performs voice processing on the first pulse code modulation signal to obtain a voice instruction.

[0120] The processor 801 may be an integrated circuit chip with signal processing capability, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor may be a microprocessor or any conventional processor, etc. The processor 801 is used to perform voice interaction control based on the voice instruction output by the voice processing device 802. For example, the voice processing device 802 recognizes the audio signal and obtains the instruction to turn on the player. At this time, the processor 801 turns on the player in the voice processing device based on the instruction.

[0121] The user interface 803 includes one or more output devices 8031 ​​that enable presentation of media content, and one or more input devices 8032. The output device 8031 ​​may be a speaker; the input device 8032 may be a microphone for receiving audio signals.

[0122] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

[0123] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0124] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, an element defined by the statement "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed.

[0125] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A speech processing device, comprising: A receiving component, configured to receive an audio signal and convert the received audio signal into a pulse density modulation signal; A detection component, used to detect whether the pulse density modulation signal meets the wake-up condition, and send a wake-up signal to the voice processor if the wake-up condition is met; A first filtering component, configured to convert the pulse density modulation signal into a first pulse code modulation signal; The voice processor is used for acquiring the first pulse code modulation signal in response to receiving the wake-up signal, and performing voice processing on the first pulse code modulation signal.

2. The device according to claim 1, wherein: The detection component includes a second filter component and a detection unit; A second filtering component, configured to convert the pulse density modulation signal into a second pulse code modulation signal; wherein the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal; The detection unit is used to detect whether the second pulse code modulation signal meets the wake-up condition, and send a wake-up signal to the voice processor if the wake-up condition is met.

3. The device according to claim 2, wherein: The device also includes a buffer; The first filtering component is further used to send the first pulse code modulation signal to the buffer; The buffer is used for buffering the first pulse code modulation signal.

4. The device according to claim 3, wherein: The voice processor is further configured to send a start instruction to the first filter component in response to receiving the wake-up signal; The first filtering component is further configured to receive a pulse density modulation signal sent by the receiving component in response to the start instruction; The speech processor is further used to obtain the first pulse code modulation signal from the buffer and perform speech processing on the first pulse code modulation signal.

5. The device according to claim 3 or 4, wherein: The speech processor is also used for: In response to receiving the wake-up signal, a target first pulse code modulation signal is acquired from the buffer based on the time when the wake-up signal is received, and speech processing is performed on the target first pulse code modulation signal.

6. The device according to claim 5, wherein: The second filtering component includes a second cascaded integrator comb decimation filter, and the first filtering component includes a first cascaded integrator comb decimation filter and a signal optimizer; The first cascaded integrator comb decimation filter is used to perform a first data extraction on the pulse density modulation signal sent by the receiving component to obtain an initial first pulse code modulation signal having a signal-to-noise ratio lower than that of the pulse density modulation signal, and the signal optimizer is used to perform signal optimization on the initial first pulse code modulation signal sent by the first cascaded integrator comb decimation filter to obtain the first pulse code modulation signal; The second cascaded integrator comb extraction filter is used to perform a second data extraction on the pulse density modulation signal sent by the receiving component to obtain the second pulse code modulation signal with a signal-to-noise ratio lower than that of the pulse density modulation signal.

7. The device according to claim 6, wherein: The signal optimizer includes a cascaded integrator comb compensation filter and a finite impulse response filter; The cascaded integrator comb compensation filter is used to compensate for the signal loss in the initial first pulse code modulation signal sent by the first cascaded integrator comb extraction filter to obtain a first compensation signal, and the finite impulse response filter is used to perform low-pass filtering on the first compensation signal sent by the first cascaded integrator comb extraction filter to obtain the first pulse code modulation signal.

8. A speech processing method, comprising: The receiving component receives the audio signal and converts the received audio signal into a pulse density modulation signal; The detection component sends a wake-up signal to wake up the speech processor in response to detecting that the pulse density modulation signal meets the wake-up condition; A first filtering component converts the pulse density modulation signal into a first pulse code modulation signal; In response to receiving the wake-up signal, the voice processor acquires the first pulse code modulation signal and performs voice processing on the first pulse code modulation signal.

9. The speech processing method according to claim 8, wherein: The detection component includes a second filtering component and a detection unit; the detection component sends a wake-up signal to wake up the voice processor in response to detecting that the pulse density modulation signal meets the wake-up condition, including: The second filtering component converts the pulse density modulation signal into a second pulse code modulation signal; wherein the signal-to-noise ratio of the first pulse code modulation signal is higher than the signal-to-noise ratio of the second pulse code modulation signal; The detection unit detects whether the second pulse code modulation signal meets a wake-up condition, and sends a wake-up signal to the speech processor if the wake-up condition is met.

10. A speech processing device, comprising a speech processing device; the speech processing device comprising: A receiving component, configured to receive an audio signal and convert the received audio signal into a pulse density modulation signal; A detection component, used to detect whether the pulse density modulation signal meets the wake-up condition, and send a wake-up signal to the voice processor if the wake-up condition is met; A first filtering component, configured to convert the pulse density modulation signal into a first pulse code modulation signal; The voice processor is used for acquiring the first pulse code modulation signal in response to receiving the wake-up signal, and performing voice processing on the first pulse code modulation signal.