Dual-channel speech timing analysis method, wearable device, medium and smart glasses

Abstract

The application discloses a double-channel speech timing analysis method, a wearable device, a medium and intelligent glasses. A microphone array is arranged on the wearable device, and two physically separated channel audio signals with a crosstalk suppression ratio not lower than a preset threshold are output on the hardware collection level by means of an acoustic isolation structure, so that the sound sources of a wearing user and a non-wearing user are distinguished without software sound source separation and voiceprint recognition. The start and end time points of a speech unit of the wearing user are detected from a first channel, and the end time point of the speech unit is taken as a trigger event. Only the second channel audio signal in a subsequent bounded time window is processed, acoustic response characteristics are measured, and an influence index representing the speech unit is output by weighted calculation. The method replaces single-channel software separation with hardware physical isolation and event-driven bounded window measurement, and solves the technical problems of single-channel device speech and reaction causal timing ambiguity, systematic timing error and high continuous calculation load.

Description

Technical Field

[0001] This application relates to the field of wearable smart device technology, and in particular to a dual-channel speech timing analysis method, wearable device, medium, and smart glasses. Background Technology

[0002] As human-computer interaction evolves from a single text modality to a multimodal paradigm that integrates voice and behavioral signals, dialogue systems are no longer satisfied with simply "understanding the content," but rather need to "understand the interaction process." In the specific form of smart glasses, the device continuously accompanies the user in face-to-face or remote dialogues. Its integrated microphone array, inertial sensors, and gyroscopes provide a unique hardware foundation for capturing the real-time reactions of dialogue participants. However, traditional single-channel recording tools have fundamental limitations in face-to-face interaction scenarios: the voices of all participants are mixed and recorded into a single audio stream, creating a chaotic acoustic superposition. The system cannot accurately determine "when the wearer ends speaking" or "in which time window does the other party's subsequent reaction begin," resulting in a complete blurring of the causal sequence between speech and reaction. In scenarios such as smart meetings, virtual assistants, remote collaboration, and social assistance, how to leverage the physical separation capabilities of dual-channel hardware to perceive, quantify, and model the dynamic reactions of dialogue participants in real time, and thereby assess changes in dialogue participation, has become one of the core requirements in the current wearable smart interaction field.

[0003] At the academic level, several fundamental research directions have emerged in the field of dialogue analysis. Firstly, "turn-taking dynamics" analyzes the impact of speaker role transitions on turn-taking rate using statistical methods, revealing macroscopic patterns in dialogue structure. Secondly, "listener response modeling" focuses on feedback signals in dialogue, quantifying and classifying backchannels such as nodding and brief responses to characterize listener engagement. At the industry and patent level, Meta's published multimodal speaker recognition patent integrates voiceprint features, directional audio, and camera visual information, achieving precise identification of "who is speaking." Simultaneously, numerous summary generation and dialogue sentiment analysis products for meeting scenarios exist on the market, enabling post-conference content extraction and sentiment labeling.

[0004] However, the aforementioned existing technologies still have significant shortcomings, making it difficult to directly support the technical patent application for the core innovation of "evaluation of changes in dialogue participant responses based on smart glasses." First, while existing frameworks cover the macroscopic quantification of dialogue characteristics, they are generally based on single-channel recording or software-based sound source separation, failing to address the fundamental problem of ambiguous causal timing between speech and response in "mixed" audio. Even with the most advanced AI algorithms, single-channel devices cannot achieve precise physical segmentation of the wearer's and the other party's speech, leading to systematic errors in the time window benchmark for participation evaluation. Second, if the solution is merely described as "analyzing changes in the other party's response using a Large Language Model (LLM)," it is easily deemed an "abstract idea" under the Alice Test framework of patent examination, lacking a close connection to the specific hardware architecture of smart glasses. Third, if existing NLP technologies are not explicitly written into the hardware perception prerequisites of smart glasses, they are easily regarded as conventional software applications of natural language processing, failing to demonstrate the substantial contribution of the technical solution in the closed loop of "physical isolation acquisition - precise temporal segmentation - edge response analysis - near-eye real-time feedback." Summary of the Invention

[0005] This application provides a dual-channel speech timing analysis method, a wearable device, and a medium to solve the aforementioned technical problems.

[0006] In view of this, the first aspect of this application provides a dual-channel speech timing analysis method, the method comprising: Step S1: Output at least two physically separated audio channels at the hardware acquisition level through a microphone array set on the wearable device. The first audio channel corresponds to the first voice content of the wearer, and the second audio channel corresponds to the second voice content of at least one non-wearer. The physical separation is achieved at the acquisition level by the acoustic isolation structure of the wearable device, so that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold, thereby distinguishing the sound source without performing software sound source separation and voiceprint recognition on the first voice content and the second voice content. Step S2: Identify at least one speech unit of the user from the first channel audio signal, and determine the start time and end time of each speech unit; Step S3: Using the end time point of the speech unit as the trigger event, process the second channel audio signal only within a bounded time window after the end time point, and extract acoustic response feature data from the second channel audio signal. The acoustic response feature data includes at least one of response delay, energy change, speech rate change rate, and fundamental frequency change. Step S4: Based on the acoustic response feature data, calculate and output the influence index representing the discourse unit through weighted combination.

[0007] Optionally, the acoustic isolation structure includes at least one of an acoustic isolation cavity, a shock-absorbing bracket, and a windproof cover; The microphone array of the first channel is positioned close to the user's mouth with its main flap facing the user's lips, while the microphone array of the second channel faces the front of the user's line of sight.

[0008] Optionally, the step of detecting speech units from the first channel audio signal and determining the start and end time points of each speech unit specifically includes: Acquire a reference signal characterizing the user's own voice, the reference signal being derived from a sensor that is synchronized with the user's body vibration or vocal cord activity and shares the same hardware clock source as the first channel audio signal; The correlation between the first channel audio signal and the reference signal is calculated, and the start and end times of the speech unit are determined based on the correlation results.

[0009] Optionally, the reference signal is at least one of bone conduction vibration signal, acceleration sign signal, or laryngeal pickup signal.

[0010] Optionally, a sliding window cross-correlation Rxr(τ) is performed between the first channel audio signal x(t) and the reference signal r(t) within the analysis window length W, where τ is the time delay compensation amount used to compensate for the propagation delay between the reference signal and the air conduction microphone; When the peak value of Rxr(τ) exceeds the preset correlation threshold, it is determined that the user wearing the device is making a sound at that moment. The starting point is determined by the moment when the reference signal first exhibits significant vibration characteristics, and the ending point is corrected by searching towards the moment when the vibration residue of the reference signal actually disappears, so that the labeling error of the ending time point does not exceed ±50ms.

[0011] Optionally, the measurement of the second channel audio signal is performed only within the bounded time window opened by the triggering event, and the measurement of the second channel audio signal is not performed outside the bounded time window; The length of the bounded time window is 5-60 seconds, and it is dynamically adjusted after the end time point.

[0012] Optionally, it also includes: Based on the energy ratio pattern of the first channel audio signal and the second channel audio signal, the rotation frequency of the speech unit, and the number of sound sources in the second channel audio signal, the dialogue scenario type is identified. The dialogue scenario type includes at least one of formal meeting, one-on-one conversation, multi-person discussion, and telephone communication. The estimation of the number of sound sources is based on the statistical characteristics of energy distribution, direction of arrival, and speech rhythm of each channel, without relying on voiceprint recognition; and according to the identified dialogue scenario type, a corresponding feature subset is selected from the acoustic response features for calculation.

[0013] Optionally, steps S1 to S4 are executed in real time on the microcontroller of the wearable device using fixed-point arithmetic. During operation, neural network inference is not performed, voiceprint recognition is not performed, and the audio signal is not uploaded to a remote server. The first channel audio signal and the second channel audio signal do not leave the wearable device and are discarded after measurement and calculation are completed. The wearable device only retains the acoustic response feature data and influence index obtained by measurement.

[0014] Optionally, the response delay is the time interval between the end time of the speech unit and the start time of the first valid response in the second channel audio signal; The energy change is the deviation of the short-time energy envelope of the second channel audio signal within the bounded time window from the baseline energy level before the end of the speech unit; The speech rate change rate is the ratio of the change in the syllable rate of the second channel audio signal within the bounded time window relative to the baseline speech rate of the speaker before the end of the speech unit. The fundamental frequency change is the offset of the statistical characteristics of the fundamental frequency trajectory of the second channel audio signal within the bounded time window relative to the fundamental frequency baseline before the end of the speech unit.

[0015] Optionally, the response feature data may also include semantic direction features; The semantic trend features include at least one of the following: changes in topic similarity, keyword replacement rate, changes in decision-making word density, and changes in sentiment polarity.

[0016] Optionally, it also includes: The influence index is calculated in time series over multiple time periods or multiple dialogue scenarios. Based on the statistical distribution changes of the influence index of the wearer in different time periods, the median drift trend of the influence index in specific dialogue scenarios, and the correlation analysis between the influence index and the wearer's communication strategy adjustment, a cross-time period influence trend report is generated in the user interface of a mobile terminal that is connected to the wearable device.

[0017] Optionally, it also includes: Based on the influence index, the discourse units are classified and labeled, and displayed in the mobile terminal user interface connected to the wearable device according to the level. The classification label includes at least three levels: high influence level, medium influence level and low influence level. The display according to the level includes visual differentiation by using different color codes or icon symbols.

[0018] Optionally, the microphone array includes at least two MEMS microphones with different polarity directional characteristics, wherein the at least two microphones include at least: Type I microphones have a polarity that is approximately unbiased. The second type of microphone is a dipole microphone, which has a fixed spatial response zero-point direction. This fixed spatial response zero-point direction is determined by the diaphragm structure and acoustic aperture layout during manufacturing and cannot be changed during operation.

[0019] A second aspect of this application provides a wearable device, the device including a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the steps of the dual-channel speech timing analysis method described in the first aspect above, according to the instructions in the program code.

[0020] A third aspect of this application provides a computer-readable storage medium for storing program code for executing the dual-channel speech timing analysis method described in the first aspect.

[0021] A fourth aspect of this application provides a smart glasses that utilizes the dual-channel speech timing analysis method described in the first aspect above. The smart glasses include: The first directional microphone array is located on the inside of the temple of the glasses, close to the mouth of the wearer, with its main lobe facing the lips of the wearer, forming the first channel; The second directional microphone array, located on the outer side of the temple or the edge of the front frame, has its main lobe facing forward of the wearer's line of sight, forming the second channel; An acoustic isolation structure located between the first directional microphone array and the second directional microphone array ensures that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold.

[0022] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: This application provides a dual-channel speech timing analysis method, a wearable device, a medium, and smart glasses. Through a microphone array installed on the wearable device, and with the aid of an acoustic isolation structure, two physically separated audio signals with a crosstalk suppression ratio not lower than a preset threshold are output at the hardware acquisition level. This distinguishes the sound sources of the wearer from those of the non-wearer without software sound source separation or voiceprint recognition. The method detects the start and end times of the wearer's speech units in the first channel, using the end time of the speech unit as a trigger event. The second channel audio signal is processed only within a bounded time window thereafter, measuring the acoustic response characteristics. Weighted calculations are then performed to output an indicator representing the influence of the speech unit. This method replaces single-channel software separation with hardware physical isolation and event-driven bounded window measurement, solving the technical problems of fuzzy causal timing of speech and response in single-channel devices, systematic timing errors, and high continuous computational load. Attached Figure Description

[0023] Figure 1 This is a flowchart of the dual-channel speech timing analysis method in the embodiments of this application; Figure 2 This is a schematic diagram of the wearable device in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the smart glasses in the embodiments of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0025] For easier understanding, please refer to Figure 1 , Figure 1 This is a flowchart of the dual-channel speech timing analysis method in the embodiments of this application, as follows: Figure 1 As shown, specifically: Step S1: Output at least two physically separated audio channels at the hardware acquisition level through a microphone array set on the wearable device. The first audio channel corresponds to the first voice content of the wearer, and the second audio channel corresponds to the second voice content of at least one non-wearer. The physical separation is achieved at the acquisition level by the acoustic isolation structure of the wearable device, so that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold, thereby distinguishing the sound source without performing software sound source separation and voiceprint recognition on the first voice content and the second voice content. Furthermore, the acoustic isolation structure includes at least one of an acoustic isolation cavity, a shock-absorbing bracket, and a windproof cover; The microphone array of the first channel is positioned close to the user's mouth with its main flap facing the user's lips, while the microphone array of the second channel faces the front of the user's line of sight.

[0026] It's important to note that the core of this step lies in solving the structural flaw of traditional single-channel recording—the "mess"—through physical channel separation at the hardware level, rather than sound source separation at the software algorithm level. In wearable devices shaped like glasses, the first directional microphone array (CH1) is typically positioned on the inner side of the temples near the wearer's mouth, employing directional beamforming technology with the main lobe pointing towards the wearer's lips to suppress oncoming ambient noise. The second directional microphone array (CH2) is positioned on the outer side of the temples or the front edge of the frame, with the main lobe pointing outwards, primarily capturing speech from the person speaking in front of the wearer's line of sight. The two microphones are physically designed with acoustic isolation chambers, shock-absorbing brackets, and windscreens to suppress crosstalk between channels (typically requiring a crosstalk suppression ratio ≥20dB).

[0027] Even when deploying state-of-the-art blind source separation (BSS) or deep learning sound source separation models, single-channel devices cannot achieve equivalent accuracy in the following scenarios: High reverberation environment: In scenarios such as conference rooms and coffee shops, reflected sound causes the sound source separation model to produce "artifacts", making it impossible to determine whether a certain segment of speech belongs to the wearer or the conversation partner; Voice overlap period: When the wearer and the other party in the conversation speak at the same time (interrupting, interrupting, echoing), the single-channel system cannot accurately cut off their respective time domain boundaries; Endpoint drift: The software separation model has an inherent processing delay (typically 50ms–500ms), which causes a systematic shift in the start and end times of discourse units, directly affecting the accuracy of subsequent causal windows.

[0028] Suppose a wearer (A) and a conversation partner (B) are having a one-on-one business negotiation. A says, "I think we should start this project at the beginning of next quarter." Traditional single-channel recording would produce a superimposed waveform of A and B's speech; however, in this solution, CH1 clearly records A's complete speech, while CH2 only captures B's weak background noise (or actively suppressed) during A's speech. After A finishes speaking, CH2 immediately captures B's response: "I agree, but regarding the budget..." Since the two signals are completely separated at the hardware level, the system does not need to perform speaker clustering or voiceprint discrimination. It can directly and accurately open the causal analysis window on CH2 based on the end time of CH1.

[0029] Step S2: Identify at least one speech unit of the user from the first channel audio signal, and determine the start and end time points of each speech unit; It should be noted that identifying utterance units (Utterance Units) from CH1 (wearer channel) and assigning their start and end timestamps is the baseline anchor for constructing the "speech-response" causal chain. This step typically employs a multi-level endpoint detection strategy: Primary detection: Coarse-grained speech activity detection (VAD) is performed based on short-time energy (STE) and zero-crossing rate (ZCR) to quickly filter out long silence segments; For more precise detection, a bone conduction vibration sensor (if integrated into the device) can be introduced as an auxiliary verification signal. Since the bone conduction signal is generated only by the wearer's vocal cord vibration and is not affected by environmental noise, the system can calculate the cross-correlation function between the CH1 audio and the bone conduction signal to accurately pinpoint the onset and offset of the voice. Semantic integrity verification: Streaming ASR can be used to verify the initial segmentation results, avoiding misjudging pauses (such as breathing or pauses in thought) as discourse unit boundaries. The system employs a sliding window semantic coherence model, confirming boundaries only when the semantic vector similarity between adjacent segments is below a threshold.

[0030] The accuracy of the end timestamp directly determines the reliability of subsequent response feature extraction. This solution requires the annotation error of the end timestamp to be controlled within ±50ms to ensure that in scenarios where the other party responds quickly (such as immediate agreement or short feedback), the other party's response is not mistakenly attributed to the end of the wearer's speech, or the wearer's drawn-out ending is misjudged as the other party's response.

[0031] For example: The wearer says, "I think this solution is feasible (pause 0.3 seconds)." A basic VAD might split the content into two utterance units due to the pause. The system recognizes through a semantic coherence model that "this solution" and "I think it's feasible" constitute a complete proposition semantically, and the pause duration is below the inter-sentence pause threshold (e.g., 800ms). Therefore, it merges the entire sentence into one utterance unit and marks the end timestamp as the end of the last sound of the word "feasible" (accurate to the sampling point level).

[0032] Step S3: Using the end time point of the speech unit as the trigger event, process the second channel audio signal only within a bounded time window after the end time point, and extract acoustic response feature data from the second channel audio signal. The acoustic response feature data includes at least one of response delay, energy change, speech rate change rate, and fundamental frequency change. It should be noted that within a preset time window ΔT (usually 5–60 seconds, dynamically adjustable) before and after the end timestamp, the system extracts response features from CH2 (the conversational channel): acoustic response features. This window is called the Causal Response Window, and its physical meaning is: to analyze only the changes in conversational behavior directly caused by the wearer's speech.

[0033] Acoustic response characteristics reflect the immediate physiological and psychological stress response of the interlocutor after hearing the wearer's words, and are characterized by low latency and difficulty in spoofing.

[0034] Furthermore, determining the start and end times of each discourse unit specifically includes: Acquire a reference signal characterizing the user's own voice. The reference signal comes from a sensor that is synchronized with the user's body vibration or vocal cord activity and shares the same hardware clock source as the first channel audio signal. The correlation between the first channel audio signal and the reference signal is calculated, and the start and end times of the speech unit are determined based on the correlation results.

[0035] Furthermore, the reference signal is specifically at least one of bone conduction vibration signal, acceleration sign signal, or laryngeal pickup signal.

[0036] Furthermore, a sliding window cross-correlation Rxr(τ) is performed between the first channel audio signal x(t) and the reference signal r(t) within the analysis window length W, where τ is the time delay compensation amount used to compensate for the propagation delay between the reference signal and the air conduction microphone; When the peak value of Rxr(τ) exceeds the preset correlation threshold, it is determined that the user wearing the device is making a sound at that moment. The starting point is determined by the moment when the reference signal first shows significant vibration characteristics, and the ending point is corrected by searching towards the moment when the vibration residue of the reference signal actually disappears, so that the labeling error of the ending time point does not exceed ±50ms.

[0037] It should be noted that in wearable audio acquisition devices (especially those in the form of glasses), although the first directional microphone array (CH1) is beamformed towards the wearer's mouth, it is still inevitably subject to the following interferences in real-world scenarios: ambient background noise (café noise, street traffic noise), leakage sound from the speaker's voice after reflection off walls, and the wearer's own non-speech vocalizations (such as coughing, throat clearing, and sighing). Traditional speech activity detection (VAD) algorithms based on energy thresholds or zero-crossing rates will produce a large number of false alarms (misjudging ambient noise as the start of speech) or missed detections (truncating the initial consonant of a weak speech syllable) in such scenarios.

[0038] This solution introduces a reference signal synchronized with the wearer's body vibrations or vocal cord activity as a physical anchor point for endpoint detection. This reference signal is acquired by a sensor independent of the air-conducting microphone, and its essential characteristic is that it responds only to the wearer's own vocal cord vibrations or craniofacial / larynx mechanical vibrations, and is naturally immune to the speech of others or background noise in the environment. By calculating the temporal correlation or frequency coherence between the CH1 audio signal and this reference signal, the system can physically confirm whether the acoustic event detected in the current CH1 truly originates from the wearer's own vocalization, thereby accurately pinpointing the start and end timestamps of the speech unit.

[0039] Synchronous signal acquisition: The CH1 microphone and the reference signal sensor share the same hardware clock source, ensuring that the time axis alignment accuracy of the two signals reaches the sampling point level (e.g., at a sampling rate of 48kHz, the time alignment error is <20.8μs).

[0040] Cross-correlation: Performing a sliding window cross-correlation between the CH1 audio signal x(t) and the reference signal r(t): ; Where W is the analysis window length (typically 20ms–40ms), and τ is the delay compensation amount (used to compensate for the propagation delay between bone conduction / larynx vibration and the air conduction microphone, typically in the range of 0.5ms–2ms). The peak value exceeds the preset threshold. At that moment, it was determined that the wearer's own voice was present.

[0041] Endpoint locking policy: Starting point lock: The moment when the CH1 energy first exceeds the mute threshold. Tracing back, find the moment when a significant vibration feature first appears in the reference signal. and with As a starting timestamp, this effectively solves the problem of weak initial sounds (such as breathy sounds like "um... I think") being missed by traditional energy VAD.

[0042] End point lock: The moment when the CH1 energy drops back to the mute threshold. Search backwards, if the reference signal is in If residual vibration persists within D milliseconds (e.g., 50ms), it is determined to be a prolonged trailing tone or voiced sound, and the end timestamp is corrected to the actual moment when the reference signal vibration disappears. This avoids misinterpreting the wearer's trailing tone (e.g., the prolonged sound of "okay—") as a response from the other party, thus preventing misalignment of the bounded time window.

[0043] Multi-reference signal fusion: When the device integrates both a bone conduction vibration sensor and an accelerometer, the system employs a weighted fusion strategy. Bone conduction signals are sensitive to mid-to-high frequency speech (voiceless consonants), while accelerometer signals are sensitive to low-frequency vibrations (voiced sounds, nasal resonance). Fusion of both can cover the full-frequency sound characteristics, improving the robustness of endpoint detection under extreme volume conditions such as whispers or shouts.

[0044] Furthermore, measurements of the second channel audio signal are performed only within the bounded time window opened by the triggered event, and no measurements of the second channel audio signal are performed outside the bounded time window. The length of the bounded time window is 5-60 seconds and is dynamically adjusted after the end time.

[0045] It should be noted that, unlike schemes that perform continuous analysis of the second channel throughout, this method adopts an event-driven processing architecture: The measurement of step S3 is initiated for the second channel only when the triggering event of the end time of the speech unit is detected in step S2 from the first channel. Outside of this window, the processor does not perform the measurement calculations on the second channel. Since user speech only accounts for a portion of the total time in face-to-face conversations, this mechanism significantly reduces the processor's average computational load and power consumption, allowing all processing to be completed in real-time on the wearable device's microcontroller using fixed-point computation, without uploading to the cloud or running neural network inference.

[0046] Furthermore, the response delay is the time interval between the end time of the speech unit and the start time of the first valid response in the second channel audio signal; The energy change is the deviation of the short-time energy envelope of the second channel audio signal within a bounded time window from the baseline energy level before the end of the speech unit. The rate of change of speech rate is the ratio of the rate of change of the syllables of the speaker's speech within a bounded time window to the baseline of the speaker's speech rate before the end of the utterance unit. The fundamental frequency change is the offset of the statistical characteristics of the fundamental frequency trajectory of the speaker's speech within a bounded time window relative to the fundamental frequency baseline before the end of the utterance unit.

[0047] It should be noted that response delay is a strong indicator of influence assessment. Experiments show that in business negotiation scenarios, a response delay of less than 0.5 seconds usually indicates strong agreement or an eager rebuttal; a delay of 2–5 seconds often signifies thinking, hesitation, or strategic silence; and a silence period of more than 10 seconds may indicate that the party is seeking external confirmation or internal resistance.

[0048] Examples of energy changes: If the wearer presents a challenging viewpoint, the other party may unconsciously raise their voice in response, indicating emotional agitation or increased defensiveness; if the other party responds at a significantly lower volume than usual, it may suggest a lack of confidence or a perfunctory attitude.

[0049] Examples of changes in speaking speed: A sudden increase in speaking speed is often accompanied by excitement, eagerness, or anxiety; a significant decrease in speaking speed may indicate careful thinking, hesitation under the pressure of authority, or the speaker organizing counterarguments.

[0050] Examples of pitch changes: When the other party responds to the wearer's question by raising the average fundamental frequency by more than 15Hz, it is usually acoustically associated with tension, uncertainty, or questioning emotions; a drop in fundamental frequency may indicate a calm, affirmative, or authoritative response.

[0051] Step S4: Based on the acoustic response feature data, calculate and output the influence index representing the discourse unit through weighted combination.

[0052] It should be noted that the influence index is obtained by weighting and combining the response features extracted in step S3.

[0053] Influence index I adopts a Min-Max normalization-scenario lookup table weighted model: ; in, The value of the i-th feature after Min-Max normalization; For scene-adaptive weights, satisfy The type is determined by looking up a table based on the identified dialogue scenario type.

[0054] Normalization rules (example): Example of scene weight representation: Furthermore, the response feature data also includes semantic direction features; Semantic trend features include at least one of the following: changes in topic similarity, keyword replacement rate, changes in decision-making word density, and changes in sentiment polarity.

[0055] Furthermore, the topic similarity change is obtained by calculating the semantic vector cosine similarity between the discourse unit and the response text of the dialogue party within the bounded time window; The keyword replacement rate is the proportion of the number of newly added keywords relative to the discourse unit in the dialogue response within a bounded time window, relative to the total number of keywords in the discourse unit. The change in the density of decision-making words is the change in the frequency of decision-making words in the dialogue party's response within a bounded time window relative to the density of decision-making words in the dialogue party's discourse before the end of the discourse unit. Emotional polarity change refers to the direction and magnitude of the shift in the emotional polarity score of the interlocutor's response within a bounded time window relative to the interlocutor's emotional polarity baseline before the end of the discourse unit.

[0056] It should be noted that the wearer's speech units and the responses of the other party in the window are encoded into semantic vectors respectively, and the cosine similarity is calculated. If the cosine similarity is greater than the first preset threshold, it indicates that the other party directly responds to the wearer's topic, which is a "continuation-type" response; if it is less than the second preset threshold, it indicates that the other party deliberately changes the topic, which may be an avoidance or defense strategy. In this case, the influence index should be significantly reduced.

[0057] Here's an example of keyword turnover rate: The wearer says, "We need to reduce this year's marketing budget." The other party responds, "I agree to control costs, but the ROI of online advertising is still rising." Here, the new keywords introduced in the other party's response, which are not part of the wearer's core keyword set, include "control costs," "online advertising," and "ROI." The keyword turnover rate is 3 / 2 = 1.5. A high turnover rate indicates that the other party is reframing the issue, possibly attempting to steer the discussion towards their own concerns. Impact assessment needs to consider this "negotiated shift."

[0058] Example of changes in the density of decision-making words: If, after the wearer proposes a solution, the density of decision-making words in the response jumps from 2% to 12% from the baseline, and the words are "can be promoted" and "execute immediately", it indicates that the wearer's words have triggered a clear decision-making behavior, and the influence indicator should receive a high weighting.

[0059] Using a pre-trained sentiment analysis model configured on a connected mobile terminal, the model outputs the sentiment polarity scores of the wearer's speech and the sentiment polarity scores of the dialogue partner's response, as well as the emotional baseline of the dialogue partner before the end timestamp. For example, the dialogue partner's baseline sentiment before the wearer speaks is neutral, but after hearing the wearer's encouraging speech, the response sentiment jumps to positive, indicating that the wearer has successfully triggered positive emotion transfer.

[0060] Furthermore, it also includes: Based on the energy ratio pattern of the first and second channel audio signals, the rotation frequency of the speech units, and the estimation of the number of sound sources in the second channel audio signal, the dialogue scenario type is identified. The dialogue scenario type includes at least one of formal meetings, one-on-one conversations, multi-person discussions, and telephone communication. The estimation of the number of sound sources is based on the statistical characteristics of energy distribution, direction of arrival and rhythm of speech in each channel, without relying on voiceprint recognition; and according to the type of dialogue scenario identified, the corresponding feature subset is selected from the acoustic response features for calculation.

[0061] It should be noted that scene recognition is a prerequisite for weight adaptation. The system performs multimodal discrimination by integrating the following features: For example: If the system detects that there are more than 3 active sound sources in the second channel at the same time (this estimate is based on the statistical characteristics of the energy distribution, direction of arrival and speech turn rhythm of each channel, without voiceprint recognition), and the speech turn frequency is higher than 15 times per minute, and the device gyroscope shows that the wearer's head has periodic looking around, then it is automatically determined to be a "multi-person discussion" scenario. The response delay weight is reduced accordingly (because the delay in multi-person scenarios is affected by the turn competition and the noise is relatively large), and the keyword replacement rate weight is increased (because the competition for topic control is more intense in multi-person scenarios).

[0062] Furthermore, steps S1 to S4 are executed in real time on the microcontroller of the wearable device using fixed-point arithmetic. During operation, neural network inference is not performed, voiceprint recognition is not performed, and audio signals are not uploaded to a remote server. The first channel audio signal and the second channel audio signal do not leave the wearable device and are discarded after measurement and calculation are completed. The wearable device only retains the acoustic response feature data and influence index obtained from the measurement.

[0063] It should be noted that all signal processing in this method is performed locally on the wearable device. The raw audio signals of the first and second channels do not leave the device, are not written to persistent storage, and are not uploaded to a remote server. After endpoint detection and acoustic feature measurement are completed, the raw audio frames are discarded, and the device only retains the measured acoustic response feature data (such as RL, ΔE, ΔSR, ΔF0 values) and the influence index calculated from them. This edge-side processing architecture reduces data transmission power consumption while preventing the leakage of raw voice content.

[0064] Furthermore, it also includes: The system calculates the time series of influence indicators across multiple time periods or multiple dialogue scenarios. Based on the statistical distribution changes of influence indicators among wearers in different time periods, the median drift trend of influence indicators in specific dialogue scenarios, and the correlation analysis between influence indicators and wearers' communication strategy adjustments, a cross-time period influence trend report is generated in the user interface of a mobile terminal that communicates with the wearable device.

[0065] It should be noted that the system accumulates time series of influence indicators of wearers at different times and in different scenarios over a long period of time, and generates trend reports, which are displayed on the user interface of the mobile terminal that communicates with the wearable device: Statistical distribution changes: Calculate box plots of the wearer's weekly / monthly influence index and observe the trends in the median and interquartile range. A continuous increase in the median indicates improved communication effectiveness; a widening interquartile range indicates unstable influence performance. Scene drift analysis: By comparing the median difference in influence between the same wearer in "formal meetings" and "one-on-one conversations", we can identify their strengths and weaknesses in certain scenarios. Strategic Correlation Analysis: Time-series correlation analysis is conducted between abrupt changes in influence indicators and adjustments to wearers' communication strategies (such as starting to use more decision-making vocabulary and reducing interruptions) to provide interpretable strategic recommendations.

[0066] For example, the report shows that the median influence of wearers in multi-person discussions was 0.45 in Q1, rising to 0.62 in Q2. Correlation analysis revealed that starting in Q2, the proportion of "closed-loop questions" (such as "So your conclusion is...") in wearers' discourse units increased by 18%, and the density of decision-making vocabulary in the corresponding responses also increased. Based on this, the system generated the suggestion: "The closed-loop questioning strategy significantly improved your influence in multi-person discussions; it is recommended to maintain this strategy." Furthermore, it also includes: Based on influence metrics, discourse units are categorized and labeled, and displayed in a mobile terminal user interface that communicates with wearable devices, with the categorization labeling including at least three levels: high influence level, medium influence level, and low influence level; the display by level includes visual differentiation using different color codes or icon identifiers.

[0067] It should be noted that the system sets three levels of influence thresholds, marking discourse units as follows: High Impact: This usually corresponds to rapid responses, positive emotional transfer, and high decision-making vocabulary density from the parties involved in the conversation. Medium Impact: Corresponds to regular dialogue follow-up or neutral feedback; Low Impact: Corresponds to long delays in dialogue, topic shifts, and no significant changes in energy / speech rate.

[0068] On the mobile terminal user interface that communicates with the wearable device in the form of glasses, the system uses different color codes or icon symbols to provide real-time feedback to the wearer: Color coding: High influence is marked with an emerald green halo, medium influence with amber, and low influence with light gray; Details Index: Wearers can expand the details card of a certain discourse unit through the mobile terminal user interface, displaying key response feature tags (such as "the other party responded quickly in 0.8 seconds" and "positive emotional transfer +0.6") and waveform thumbnails of the original CH2 response segment.

[0069] For example, after a wearer speaks in a meeting, their speech is marked in emerald green on the mobile device's user interface, indicating that the system has determined that the speech has a high impact. Clicking on the details reveals: "Decision-making vocabulary density increased by 10% | Response delay of 1.2 seconds | Topic similarity 0.85," instantly confirming that the viewpoint has been substantially accepted by the other party, allowing for a smooth transition to the next topic.

[0070] Furthermore, the microphone array includes at least two MEMS microphones with different polarity directional characteristics, wherein the at least two microphones include at least: Type I microphones have a polarity that is approximately unbiased. The second type of microphone is the dipole microphone. The dipole microphone has a fixed spatial response zero-point direction, which is determined by the diaphragm structure and acoustic aperture layout during manufacturing and cannot be changed during operation.

[0071] It should be noted that the audio signal is collected by a microphone array set on the wearable device. The key feature of this array is that it uses at least two MEMS microphones with different polarity directional characteristics to form a physical-level spatial filtering prior.

[0072] Type 1: Omnidirectional microphone This type of microphone exhibits consistent sensitivity in all directions, without any obvious directional bias. Its function is to uniformly capture sound from all directions, providing complete sound field information for subsequent processing. In glasses-shaped devices, omnidirectional microphones are typically positioned in the front array area, close to the wearer's face, enabling them to effectively capture the wearer's own voice as well as sound sources in front of them.

[0073] Type 2: Dipole microphone This is one of the core innovations of this solution. Dipole microphones have a typical figure-eight polar pattern, with a fixed spatial response null point determined during the manufacturing stage—meaning sound is almost impossible to pick up in certain specific directions. This null point direction is fixed at the factory by the diaphragm structure and acoustic aperture layout and cannot be changed during operation. For example, if the acoustic apertures are symmetrically distributed on both sides of the diaphragm, the null point direction is perpendicular to the diaphragm plane; if an asymmetrical layout is used, the null point direction will shift accordingly. This physical characteristic is not assigned by software algorithms but is an inherent property of the hardware itself.

[0074] Another embodiment of the present invention provides a wearable device for dual-channel audio acquisition, such as... Figure 2 As shown, device 10 includes: One or more processors 110 and memory 120, Figure 2 The following description uses a processor 110 as an example. The processor 110 and the memory 120 can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0075] Processor 110 is used to perform various control logics of device 10, and can be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), microcontroller, ARM (Acorn RISC Machine) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of these components. Furthermore, processor 110 can also be any conventional processor, microprocessor, or state machine. Processor 110 can also be implemented as a combination of computing devices, such as a combination of DSP and microprocessor, multiple microprocessors, one or more microprocessors combined with DSP and / or any other such configuration.

[0076] The memory 120, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions corresponding to the dual-channel speech timing analysis method in this embodiment of the invention. The processor 110 executes various functional applications and data processing of the device 10 by running the non-volatile software programs, instructions, and units stored in the memory 120, thereby implementing the dual-channel speech timing analysis method in the above method embodiment.

[0077] The memory 120 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created according to the use of the device 10. Furthermore, the memory 120 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include memory remotely located relative to the processor 110, and these remote memories may be connected to the device 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0078] One or more units are stored in memory 120, and when executed by one or more processors 110, perform the following steps: Step S1: Output at least two channels of audio signals that are physically separated at the hardware acquisition level through the microphone array set on the wearable device. The first channel audio signal corresponds to the first voice content of the wearer and the second channel audio signal corresponds to the second voice content of at least one non-wearer. The physical separation is achieved at the acquisition level by the acoustic isolation structure of the wearable device, so that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold, thereby distinguishing the sound source without performing software sound source separation and voiceprint recognition on the first and second voice content. Step S2: Identify at least one speech unit of the user from the first channel audio signal, and determine the start and end time points of each speech unit; Step S3: Using the end time of the speech unit as the trigger event, the second channel audio signal is processed only within a bounded time window after the end time. Acoustic response feature data is extracted from the second channel audio signal. The acoustic response feature data includes at least one of response delay, energy change, speech rate change rate, and fundamental frequency change. Step S4: Based on acoustic response feature data, calculate and output the influence index representing the discourse unit through weighted combination.

[0079] This invention provides a non-volatile computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are executed by one or more processors, they implement any one of the embodiments of the dual-channel speech timing analysis method described above.

[0080] As examples, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The memory components or memories disclosed in the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0081] like Figure 3 As shown in the illustration, this application also provides a smart glasses application using the dual-channel speech timing analysis method described in the method embodiment. The smart glasses include: The first directional microphone array 301, located on the inside of the temple of the glasses and close to the mouth of the user, has its main lobe facing the lips of the user, forming the first channel; The second directional microphone array 302, located on the outer side of the temple or the edge of the front frame, has its main lobe facing forward of the wearer's line of sight, thus forming a second channel. An acoustic isolation structure 303 located between the first directional microphone array 301 and the second directional microphone array 302 ensures that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold.

[0082] This application provides a dual-channel speech timing analysis method, a wearable device, a medium, and smart glasses. Using a microphone array mounted on the wearable device and an acoustic isolation structure, it outputs two physically separated audio signals with a crosstalk suppression ratio not lower than a preset threshold at the hardware acquisition level. This distinguishes the sound sources of the wearer from those of the non-wearer without software sound source separation or voiceprint recognition. The method detects the start and end times of the wearer's speech units from the first channel, using the end time of the speech unit as a trigger event. The second channel audio signal is processed only within a bounded time window after the end of the speech unit, measuring the acoustic response characteristics. Weighted calculations are then performed to output an indicator representing the influence of the speech unit. This method replaces single-channel software separation with hardware physical isolation and event-driven bounded window measurement, solving the technical problems of fuzzy causal timing of speech and response in single-channel devices, systematic timing errors, and high continuous computational load.

[0083] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0084] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0085] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0087] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0090] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0091] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

Claims

1. A dual-channel speech timing analysis method, characterized in that, include: Step S1: Output at least two physically separated audio channels at the hardware acquisition level through a microphone array set on the wearable device. The first audio channel corresponds to the first voice content of the wearer, and the second audio channel corresponds to the second voice content of at least one non-wearer. The physical separation is achieved at the acquisition level by the acoustic isolation structure of the wearable device, so that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold, thereby distinguishing the sound source without performing software sound source separation and voiceprint recognition on the first voice content and the second voice content. Step S2: Identify at least one speech unit of the user from the first channel audio signal, and determine the start time and end time of each speech unit; Step S3: Using the end time point of the speech unit as the trigger event, process the second channel audio signal only within a bounded time window after the end time point, and extract acoustic response feature data from the second channel audio signal. The acoustic response feature data includes at least one of response delay, energy change, speech rate change rate, and fundamental frequency change. Step S4: Based on the acoustic response feature data, calculate and output the influence index representing the discourse unit through weighted combination.

2. The dual-channel speech timing analysis method according to claim 1, characterized in that, The acoustic isolation structure includes at least one of an acoustic isolation cavity, a shock-absorbing bracket, and a windproof cover; The microphone array of the first channel is positioned close to the user's mouth with its main flap facing the user's lips, while the microphone array of the second channel faces the front of the user's line of sight.

3. The dual-channel speech timing analysis method according to claim 1, characterized in that, The step of detecting speech units from the first channel audio signal and determining the start and end time points of each speech unit specifically includes: Acquire a reference signal characterizing the user's own voice, the reference signal being derived from a sensor that is synchronized with the user's body vibration or vocal cord activity and shares the same hardware clock source as the first channel audio signal; The correlation between the first channel audio signal and the reference signal is calculated, and the start and end time points of the speech unit are determined based on the correlation results.

4. The dual-channel speech timing analysis method according to claim 3, characterized in that, The reference signal is specifically at least one of bone conduction vibration signal, acceleration sign signal, or laryngeal pickup signal.

5. The dual-channel speech timing analysis method according to claim 3, characterized in that, A sliding window cross-correlation Rxr(τ) is performed between the first channel audio signal x(t) and the reference signal r(t) within the analysis window length W, where τ is the time delay compensation amount used to compensate for the propagation delay between the reference signal and the air conduction microphone; When the peak value of Rxr(τ) exceeds the preset correlation threshold, it is determined that the user wearing the device is making a sound at that moment. The starting point is determined by the moment when the reference signal first exhibits significant vibration characteristics, and the ending point is corrected by searching towards the moment when the vibration residue of the reference signal actually disappears, so that the labeling error of the ending time point does not exceed ±50ms.

6. The dual-channel speech timing analysis method according to claim 1, characterized in that, The measurement of the second channel audio signal is performed only within the bounded time window opened by the triggering event, and the measurement of the second channel audio signal is not performed outside the bounded time window; The length of the bounded time window is 5-60 seconds, and it is dynamically adjusted after the end time point.

7. The dual-channel speech timing analysis method according to claim 1, characterized in that, Also includes: Based on the energy ratio pattern of the first channel audio signal and the second channel audio signal, the rotation frequency of the speech unit, and the number of sound sources in the second channel audio signal, the dialogue scenario type is identified. The dialogue scenario type includes at least one of formal meeting, one-on-one conversation, multi-person discussion, and telephone communication. The estimation of the number of sound sources is based on the statistical characteristics of energy distribution, direction of arrival, and speech rhythm of each channel, without relying on voiceprint recognition; and according to the identified dialogue scenario type, a corresponding feature subset is selected from the acoustic response features for calculation.

8. The dual-channel speech timing analysis method according to claim 1, characterized in that, Steps S1 to S4 are executed in real time on the microcontroller of the wearable device using fixed-point arithmetic. During operation, neural network inference is not performed, voiceprint recognition is not performed, and the audio signal is not uploaded to a remote server. The first channel audio signal and the second channel audio signal do not leave the wearable device and are discarded after measurement and calculation are completed. The wearable device only retains the acoustic response feature data and influence index obtained by measurement.

9. The dual-channel speech timing analysis method according to claim 1, characterized in that, The response delay is the time interval between the end time of the speech unit and the start time of the first valid response in the second channel audio signal; The energy change is the deviation of the short-time energy envelope of the second channel audio signal within the bounded time window from the baseline energy level before the end of the speech unit; The speech rate change rate is the ratio of the change in syllable rate of the second channel audio signal within the bounded time window relative to the baseline speech rate of the speaker before the end of the speech unit. The fundamental frequency change is the offset of the statistical characteristics of the fundamental frequency trajectory of the second channel audio signal within the bounded time window relative to the fundamental frequency baseline before the end of the speech unit.

10. The dual-channel speech timing analysis method according to claim 1, characterized in that, Response feature data also includes semantic direction features; The semantic trend features include at least one of the following: changes in topic similarity, keyword replacement rate, changes in decision-making word density, and changes in sentiment polarity.

11. The dual-channel speech timing analysis method according to claim 1, characterized in that, Also includes: The influence index is calculated in time series over multiple time periods or multiple dialogue scenarios. Based on the statistical distribution changes of the influence index of the wearer in different time periods, the median drift trend of the influence index in specific dialogue scenarios, and the correlation analysis between the influence index and the wearer's communication strategy adjustment, a cross-time period influence trend report is generated in the user interface of a mobile terminal that is connected to the wearable device.

12. The dual-channel speech timing analysis method according to claim 1, characterized in that, Also includes: Based on the influence index, the discourse units are classified and labeled, and displayed in the mobile terminal user interface connected to the wearable device according to the level. The classification label includes at least three levels: high influence level, medium influence level and low influence level. The display according to the level includes visual differentiation by using different color codes or icon symbols.

13. The dual-channel speech timing analysis method according to claim 1, characterized in that, The microphone array includes at least two types of MEMS microphones with different polarity pointing characteristics, wherein the at least two types of microphones include at least: Type I microphones have a polarity that is approximately unbiased. The second type of microphone is a dipole microphone, which has a fixed spatial response zero-point direction. This fixed spatial response zero-point direction is determined by the diaphragm structure and acoustic aperture layout during manufacturing and cannot be changed during operation.

14. A wearable device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the dual-channel speech timing analysis method according to any one of claims 1-13 according to the instructions in the program code.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the dual-channel speech timing analysis method according to any one of claims 1-13.

16. A type of smart glasses, characterized in that, The smart glasses, using the dual-channel speech timing analysis method according to any one of claims 1-13, comprise: The first directional microphone array is located on the inside of the temple of the glasses, close to the mouth of the wearer, with its main lobe facing the lips of the wearer, forming the first channel; The second directional microphone array, located on the outer side of the temple or the edge of the front frame, has its main lobe facing forward of the wearer's line of sight, forming the second channel; An acoustic isolation structure located between the first directional microphone array and the second directional microphone array ensures that the crosstalk suppression ratio between the first channel and the second channel is not lower than a preset threshold.

Images