Real-time acquisition and analysis system and method for telephone call voice based on Android platform

Abstract

The invention discloses a fixed-line telephone call voice real-time acquisition and analysis system and method based on an Android platform, and belongs to the technical field of telephone communication. A fixed-line telephone and an Android acquisition box are connected by using a one-to-two bridge to form an analog signal bypass, so that the Android acquisition box can record on the basis of not influencing the call; and the Android collection box only collects the effective call voice in real time, encrypts the effective call voice and uploads the effective call voice to the cloud server, so that non-effective call voice is eliminated, and the subsequent quality inspection efficiency and the text conversion accuracy of the cloud server are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of telephone communication technology, and in particular to a system and method for real-time acquisition and analysis of landline call voice based on the Android platform. Background Technology

[0002] In industries such as call centers, telephone customer service centers, financial and insurance telemarketing, hotel reservation centers, and government hotlines, where landline outbound / inbound calls are the core service channels, there will be a quality inspection process. The purpose is to ensure that the service process of customer service representatives (agents) conforms to the standard procedures and script specifications set by the company, or to promptly identify service defects that may lead to customer loss, complaints, or negative publicity. Quality inspection is mostly completed by detecting emotions and recognizing intentions in the voice based on the content of the call. In existing technologies, quality inspection is usually carried out either manually or by uploading the complete recording of the call for inspection, which results in serious deficiencies in the accuracy and efficiency of the quality inspection. Summary of the Invention

[0003] To address the technical problems existing in the prior art, this invention provides a method for real-time acquisition and analysis of landline call voice based on the Android platform, comprising the following steps: Connect the landline phone to the Android data acquisition box using a 1-to-2 splitter; The Android capture box identifies and captures valid call voice in real time, and encrypts and uploads it to the cloud server. After the cloud server decrypts the valid voice recordings, it uses automatic speech recognition technology to convert them into text, thus obtaining the final text. The final text is subjected to emotion detection and intent recognition.

[0004] Furthermore, the 1-to-2 bridge connector inserts its male plug into the handset port of the base unit, then connects its first female connector to the base unit's handset cable to complete the base unit's original call circuit, and finally connects its second female connector to the 3.5mm four-segment headphone plug of the Android acquisition box via a flexible flat cable.

[0005] Furthermore, the Android capture box identifies and captures valid call voice in real time, specifically as follows: Real-time monitoring determines whether the landline phone has been picked up; if so, recording begins. During the recording process, valid call audio is extracted in real time and written to the memory queue of the current call; During the recording process, the system monitors in real time to determine whether the landline has hung up. If so, the recording is stopped, and the valid call audio in the memory queue is packaged into an audio file.

[0006] Furthermore, the real-time extraction of valid call voice specifically includes: With a preset fixed time interval as one frame, the audio signal energy of each frame is obtained through an audio energy detection algorithm. Audio segments with a consecutive frame number greater than or equal to a preset frame number and an audio signal energy of each frame greater than or equal to a preset energy threshold are judged as valid call speech and extracted.

[0007] Furthermore, before writing the valid call audio to the memory queue, the following steps are also included: Insert a preset number of trailing frames at the end of each valid voice call.

[0008] Furthermore, the encrypted upload to the cloud server specifically includes: The Android capture box constructs a request body that conforms to a standard format, puts the audio file into it, and initiates a connection to the automatic speech recognition interface of the cloud server through the system network library; During the connection process, the Android capture box and the cloud server establish a secure encrypted channel through a preset encryption suite, and perform security verification on the cloud server's certificate. After completing the security verification, the Android data collection box generates authentication information for this request and puts it into the request header, which is then sent to the cloud server along with the request body. The cloud server verifies the authentication information, and returns an error signal if the verification fails.

[0009] Furthermore, the preset encryption suite specifically selects the TLS_AES_256_GCM_SHA384 encryption suite.

[0010] Furthermore, the cloud server verifies the identity verification information, specifically as follows: The authentication information is parsed from the request header, including digital signature, timestamp, random number, string used when concatenating signature, and ID identifier assigned to the Android collection box by the cloud server; The system detects whether the deviation between the timestamp and the time when the cloud server receives the request header exceeds a second preset duration. If yes, an error signal is returned; otherwise, a random number check is performed. The random number verification: The cloud server caches the random number after each successful verification, and the cache validity period is the same as the first preset period. If the current random number already exists in the cache, an error signal is returned; otherwise, a consistency check is performed. The consistency verification is as follows: The cloud server queries the corresponding key from the key database using the parsed ID identifier, concatenates the string used in the signature concatenation in the exact same order as the Android collection box to obtain the second string, uses the queried key to calculate the second string using the same signature algorithm as the Android collection box to obtain the second digital signature, and determines whether the second digital signature is consistent with the digital signature in the request header. If not, an error signal is returned.

[0011] Furthermore, the text conversion using automatic speech recognition technology specifically involves inputting the effective call speech into a pre-trained acoustic model, outputting a corresponding probability matrix, and then decoding it through a decoder to obtain the first text; the first text is then subjected to text normalization processing to obtain the final text. The acoustic model is trained as follows: a large amount of narrowband speech data is acquired as a dataset, and line distortion data augmentation technology is applied to the dataset to generate an expanded dataset containing various line noises. The acoustic model is trained using the expanded dataset as training data.

[0012] This invention also provides a real-time acquisition and analysis system for landline call voice based on the Android platform, which applies any of the above-described methods for real-time acquisition and analysis of landline call voice based on the Android platform, including: A 1-to-2 splitter for connecting a landline phone and an Android data acquisition box; An Android capture box is used to identify and capture valid call audio in real time and encrypt and upload it to a cloud server. The cloud server is used to decrypt the valid voice call, convert it into text using automatic speech recognition technology, and then perform emotion detection and intent recognition on the final text.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes a 1-to-2 bridge to connect a landline phone and an Android capture box, creating an analog signal bypass. This allows the Android capture box to record audio without affecting the call. The Android capture box then captures only valid call audio in real time and uploads it to the cloud server in encrypted form, eliminating invalid call audio. This effectively improves the efficiency of subsequent quality inspection and the accuracy of text conversion on the cloud server. Attached Figure Description

[0014] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for real-time acquisition and analysis of landline call voice based on the Android platform according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0019] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0020] Example 1 See Figure 1 As shown, the present invention provides a method for real-time acquisition and analysis of landline call voice based on the Android platform, which specifically includes the following steps: S1. Connect the landline phone to the Android data acquisition box using a 1-to-2 splitter; S2. The Android acquisition box identifies and acquires valid call voice in real time, and encrypts and uploads it to the cloud server; S3: After the cloud server decrypts the valid voice call, it uses automatic speech recognition technology to convert the voice call into text to obtain the final text. S4. Perform emotion detection and intent recognition on the final text.

[0021] S1. Connect the landline phone to the Android data acquisition box using a 1-to-2 splitter: The 1-to-2 bridge connector inserts its male plug into the handset port of the base unit, then connects its first female connector to the base unit's handset cable to complete the base unit's original call circuit, and finally connects its second female connector to the 3.5mm four-segment headphone plug of the Android acquisition box via a flexible flat cable.

[0022] The pins of the male plug, the first female socket, and the second female socket are in one-to-one correspondence. The MIC+ pin of the male plug is directly connected to the MIC+ pin of the first female socket, and the SPK+ pin of the male plug is directly connected to the SPK+ pin of the second female socket. The pins of the same type of the male plug and the second female socket are coupled through an audio transformer to achieve DC signal isolation of at least 1500V. The secondary output of the audio transformer is connected to the MIC bias circuit of the Android acquisition box through an RC low-pass network to complete high-impedance bridging acquisition.

[0023] A straight-through connection is achieved by directly shorting the connection with a single wire or PCB trace, allowing both uplink and downlink voice signals to reach the handset without obstruction, thus ensuring smooth communication.

[0024] The 1-to-2 bridge is a passive, high-fidelity voice signal coupling device. Physically, it bridges the handset and handset cable of a traditional analog landline phone to the landline host. It can bypass the bidirectional analog audio signal and provide it to the subsequent Android acquisition box for analysis without interrupting or changing the original call circuit.

[0025] In some embodiments, the 1-to-2 bridge can be a 1-to-2 RJ9 bridge. In this solution, the Android data acquisition box, located at the "edge" of the call (i.e., next to the landline), performs a high-quality, low-latency, and highly secure conversion from analog signals to analyzable data, and pushes the data to the cloud server in real time and reliably. It replaces the single-function DSP recording box or complex industrial control computer in traditional solutions, leveraging the maturity, openness, and powerful computing capabilities of the Android ecosystem to achieve an optimal balance between cost, performance, and functionality.

[0026] This solution connects the landline and the Android data acquisition box via a 1-to-2 bridge, forming a plug-and-play analog bypass. It does not require any changes to the original PSTN lines, number resources, or dialing habits, thus solving the problem that traditional IP-based solutions require replacing the landline or adding an IAD network.

[0027] S2. The Android capture box identifies and captures valid call audio in real time, and encrypts and uploads it to the cloud server. The Android capture box identifies and captures valid call voice in real time, specifically as follows: Real-time monitoring determines whether the landline phone has been picked up; if so, recording begins. During the recording process, valid call audio is extracted in real time and written to the memory queue of the current call; During the recording process, the system monitors in real time to determine whether a hang-up event has occurred on the landline. If so, the recording is stopped, and the valid call audio in the memory queue is packaged into an audio file.

[0028] The real-time monitoring and determination of whether the landline phone has been picked up specifically involves: The Android acquisition box uses an analog-to-digital converter to sample the DC level of the MIC pin at the 3.5mm four-segment headphone plug at a preset sampling rate. The average level is calculated by averaging multiple consecutive sampling points. If the average level is greater than the first preset level threshold and is maintained for a third preset duration, it is determined that the landline has been picked up.

[0029] The real-time monitoring and judgment of whether a call has been hung up specifically involves: If the average level is less than the second preset level threshold and remains continuously for the first preset duration, it is determined that a hang-up event has occurred on the landline.

[0030] The first preset level threshold is greater than the second preset level threshold.

[0031] It should be noted that the hang-up event in this solution does not mean the end of the call. After the recording stops, it continues to monitor and determine whether the landline has been picked up. By setting the corresponding first preset duration value, it is possible to capture each sentence in real time, encrypt and upload it to complete the subsequent testing operation, and realize the problem of delayed quality inspection results by recording and testing at the same time.

[0032] The real-time extraction of valid call voice is specifically as follows: With a preset fixed time interval as one frame, the audio signal energy of each frame is obtained through an audio energy detection algorithm. Audio segments with a consecutive frame number greater than or equal to a preset frame number and an audio signal energy of each frame greater than or equal to a preset energy threshold are judged as valid call speech and extracted.

[0033] Before writing valid call audio to the memory queue, the following steps are also included: Insert a preset number of trailing frames at the end of each valid voice message to avoid truncating the ending sound.

[0034] In some embodiments, the audio energy detection algorithm is the WebRTC APM Energy-VAD algorithm; WebRTC APM Energy-VAD Algorithm: WebRTC APM is the audio processing module in Google's Web Real-Time Communication project. Energy-VAD is an energy-based speech activity detection algorithm for this audio processing module. The audio signal energy during human speech is typically significantly higher than that of background noise and silent periods; therefore, this algorithm can effectively detect valid speech during a call.

[0035] In some embodiments, the preset energy threshold can be set to -50 dBFS; This scheme sets a threshold of audio signal energy greater than or equal to -50dBFS, meaning that only when the total energy of an audio frame reaches approximately 0.316% of full scale is it considered valid speech. This aims to filter out most stable low-energy background noise such as line noise and current hum. At the same time, the judgment condition of a consecutive frame number greater than or equal to a preset frame number (e.g., 6 frames, 60ms) is set because there may be some short-term pulses of noise on the telephone line, which may have high energy but extremely short duration. The duration condition can effectively filter out such interference and prevent it from being misjudged as valid speech. In normal conversation speech, the duration of a phoneme or syllable is usually greater than 60ms. This condition ensures that what is detected is a stable and continuous speech segment, rather than occasional noise.

[0036] The encrypted upload to the cloud server is specifically as follows: The Android capture box constructs a request body that conforms to a standard format, puts the audio file into it, and initiates a connection to the automatic speech recognition interface of the cloud server through the system network library; During the connection process, the Android capture box and the cloud server establish a secure encrypted channel through a preset encryption suite, and perform security verification on the cloud server's certificate. After completing the security verification, the Android data collection box generates authentication information for this request and puts it into the request header, which is then sent to the cloud server along with the request body. The cloud server verifies the authentication information, and returns an error signal if the verification fails.

[0037] Authentication information includes, but is not limited to, digital signatures, timestamps, random numbers, strings used in signature concatenation, and the ID identifier assigned to the Android data collection box by the cloud server.

[0038] In some embodiments, the request body conforming to the standard format may be a multipart / form-data format request body, with the audio file placed in the file field of the request body; The system network library initiates a connection to the cloud server's automatic speech recognition interface. Specifically, it can call HttpsURLConnection in the Java standard library to initiate a connection to the ASR interface configured on the cloud server. During the connection process, the Android capture box establishes a secure encrypted channel with the cloud server, which can be established through a preset encryption suite to create a TLS secure encrypted channel.

[0039] The cloud server's certificate is verified for security. Specifically, the cloud server sends a TLS certificate to the Android data acquisition box. The Android data acquisition box verifies, through its built-in CA root certificate store, whether the TLS certificate was issued by a trusted authority, whether it has expired, and whether the domain name matches.

[0040] The digital signature can be calculated by the Android data acquisition box using a signature algorithm for this request.

[0041] HttpsURLConnection is a core class in the Java standard library used to initiate HTTPS requests. It is a subclass of HttpURLConnection and is specifically designed to handle HTTP communication based on SSL / TLS encryption. It is built into the java.net package and allows for secure web communication in Java applications without additional dependencies.

[0042] The preset encryption suite specifically uses the TLS_AES_256_GCM_SHA384 encryption suite. Here, TLS indicates that it is a transport layer security protocol used for secure communication over the Internet; AES_256_GCM indicates that it is a symmetric key algorithm and mode used for encrypting data. AES is a symmetric encryption algorithm, meaning that encryption and decryption use the same key; 256 refers to the key length of 256 bits; GCM is one of the operating modes of AES; and SHA384 is the hash algorithm used to generate cryptographic hashes during the TLS handshake process.

[0043] ASR interface stands for Automatic Speech Recognition.

[0044] This solution uses Alibaba Cloud servers. Alibaba Cloud certificates are typically issued by internationally renowned CAs such as DigiCert, GlobalSign, or Symantec, and the root certificates of these CAs are pre-installed in the Android system.

[0045] Once the TLS certificate is verified, the TLS handshake is complete.

[0046] The cloud server verifies the identity verification information, specifically as follows: The authentication information is parsed from the request header, including digital signature, timestamp, random number, string used when concatenating signature, and ID identifier assigned to the Android collection box by the cloud server; The system detects whether the deviation between the timestamp and the time when the cloud server receives the request header exceeds a second preset duration. If yes, an error signal is returned; otherwise, a random number check is performed. The random number verification: The cloud server caches the random number after each successful verification, and the cache validity period is the same as the first preset period. If the current random number already exists in the cache, an error signal is returned; otherwise, a consistency check is performed. The consistency verification is as follows: The cloud server queries the corresponding key from the key database using the parsed ID identifier, concatenates the string used in the signature concatenation in the exact same order as the Android collection box to obtain the second string, uses the queried key to calculate the second string using the same signature algorithm as the Android collection box to obtain the second digital signature, and determines whether the second digital signature is consistent with the digital signature in the request header. If not, an error signal is returned.

[0047] After the S3 cloud server decrypts the valid voice recordings, it performs text conversion using automatic speech recognition technology. The text conversion using automatic speech recognition technology specifically involves inputting the effective call speech into a pre-trained acoustic model, outputting a corresponding probability matrix, and then decoding it through a decoder to obtain the first text; the first text is then subjected to text normalization processing to obtain the final text. The acoustic model is trained as follows: a large amount of narrowband speech data is acquired as a dataset, and line distortion data augmentation technology is applied to the dataset to generate an expanded dataset containing various line noises. The acoustic model is trained using the expanded dataset as training data.

[0048] The line noise includes, but is not limited to, G.711A quantization noise, frequency offset, and intermittent packet loss. Training the model on an expanded dataset containing various types of line noise effectively improves its robustness.

[0049] The narrowband voice data can be 8kHz narrowband voice data.

[0050] The acoustic model can be the TDNN-F acoustic model (Factorized Time Delay Neural Network), a deep learning model specifically designed for speech recognition. It can effectively capture the temporal dependencies of speech signals and is particularly suitable for processing continuous speech streams. Its output probability matrix usually exists in the form of log-posterior probability or negative log-likelihood. Essentially, for each frame of speech input, the model gives the confidence score of its belonging to each acoustic modeling unit (usually HMM state).

[0051] G.711A quantization noise refers to the signal distortion introduced by nonlinear quantization during the digitization of speech signals using the G.711 A-law coding standard. This distortion manifests as background noise similar to a "hissing" sound at the receiving end.

[0052] The decoder can be a WFST decoder (Weighted Finite-State TransducerDecoder), a search engine based on the WFST graph structure, a core component of ASR, which can efficiently search for the most likely word sequence (i.e., recognition result) from a huge language and pronunciation possibility space given the acoustic model output.

[0053] Text normalization is the process of converting non-standard and diverse original text formats into a unified, standardized, and machine-readable standard format, and converting numbers, currencies, dates, etc., into standard formats (e.g., "ten thousand twelve" is converted into 12000).

[0054] In this solution, 8kHz narrowband speech data is used as the dataset. Line distortion data augmentation technology is applied to the dataset to generate an expanded dataset containing various line noises. This expanded dataset is used as the training sample for the TDNN-F acoustic model, which effectively reduces the subsequent word error rate and improves the accuracy of speech-to-text conversion in landline call scenarios.

[0055] S4. Perform emotion detection and intent recognition on the final text: The emotion detection specifically involves segmenting the final text to obtain multiple word units, and then using word embedding technology to convert each word unit into a numerical vector of a preset length to construct a word embedding matrix. The word embedding matrix is ​​then used to calculate the emotion probability of each emotion type through a preset text convolutional neural network. The emotion types include negative, neutral, positive, and excited.

[0056] The intent recognition specifically involves outputting a sequence embedding matrix from the final text using a preset RoBERTa-wwm-ext model, then inputting the sequence embedding matrix into a preset bidirectional long short-term memory network, and finally outputting the intent probability of each intent type through a classification layer.

[0057] The RoBERTa-wwm-ext model is a Chinese pre-trained language model improved upon RoBERTa by the Harbin Institute of Technology (HIT) team, specifically optimized for Chinese semantic understanding. Through Whole Word Masking (WWM) technology and training on large-scale Chinese corpora, it significantly improves the performance of Chinese NLP tasks (such as intent recognition), taking Chinese text as input and outputting semantic embeddings.

[0058] The specific category of the intent type is determined by setting it according to requirements when training the bidirectional long short-term memory network.

[0059] After completing emotion detection and intent recognition, the process also includes: The final text is anonymized to obtain anonymized text; the anonymized text, along with the corresponding emotion detection results, intent recognition results, sensitive word list, and quality inspection score, are encapsulated into JSON data and stored in the database of the Android data collection box.

[0060] The de-identification process includes, for example, regular expression de-identification of mobile phone numbers, bank card numbers, and ID card numbers using preset masking rules. The sensitive word list is obtained by identifying sensitive words from the de-identified text based on a preset sensitive word library. The quality inspection score can be calculated based on emotion detection results and intent recognition results using preset scoring rules.

[0061] Example 2 This invention also provides a real-time voice acquisition and analysis system for landline calls based on the Android platform, specifically comprising: A 1-to-2 splitter for connecting a landline phone and an Android data acquisition box; An Android capture box is used to identify and capture valid call audio in real time and encrypt and upload it to a cloud server. The cloud server is used to decrypt the valid voice call, convert it into text using automatic speech recognition technology, and then perform emotion detection and intent recognition on the final text.

[0062] Example 3 The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods.

[0063] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0064] The beneficial effects of this invention are as follows: This invention utilizes a 1-to-2 bridge to connect a landline phone and an Android capture box to form an analog signal bypass, thereby enabling the Android capture box to record audio without affecting the call. The Android capture box then captures only valid call audio in real time and uploads it to the cloud server in encrypted form, eliminating invalid call audio, which effectively improves the efficiency of subsequent quality inspection and the accuracy of text conversion on the cloud server. By connecting the landline and Android acquisition box through a 1-to-2 bridge, a plug-and-play analog bypass is formed, which does not require modification of the original PSTN line, number resources and dialing habits, and solves the problem that traditional IP solutions must replace the landline or add an IAD network. By setting an audio signal energy threshold, the system ensures that only when the total energy of a single audio frame reaches a specific proportion of full scale is it considered valid speech. This mechanism aims to effectively filter out low-energy background interference commonly found in telephone lines, such as stable background noise and current hum. Simultaneously, the solution includes a duration-based judgment condition for consecutive frames to address occasional short-duration impulse noise that may occur on telephone lines. Although such interference may have high energy, its extremely short duration allows for effective exclusion through the duration condition, avoiding false positives. Considering that the duration of a single phoneme or syllable in a normal conversation usually exceeds this set duration, this condition reliably ensures that only stable and continuous speech segments are detected, thereby improving the accuracy of speech extraction.

[0065] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] 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. 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 multiple 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 in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0067] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for real-time acquisition and analysis of landline call voice based on the Android platform, characterized in that, Includes the following steps: Connect the landline phone to the Android data acquisition box using a 1-to-2 splitter; The Android capture box identifies and captures valid call voice in real time, and encrypts and uploads it to the cloud server. After the cloud server decrypts the valid voice recordings, it uses automatic speech recognition technology to convert them into text, thus obtaining the final text. The final text is subjected to emotion detection and intent recognition.

2. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 1, characterized in that, The 1-to-2 bridge connector inserts its male plug into the handset port of the base unit, then connects its first female connector to the base unit's handset cable to complete the base unit's original call circuit, and finally connects its second female connector to the 3.5mm four-segment headphone plug of the Android acquisition box via a flexible flat cable.

3. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 1, characterized in that, The Android capture box identifies and captures valid call voice in real time, specifically as follows: Real-time monitoring determines whether the landline phone has been picked up; if so, recording begins. During the recording process, valid call audio is extracted in real time and written to the memory queue of the current call; During the recording process, the system monitors in real time to determine whether the landline has hung up. If so, the recording is stopped, and the valid call audio in the memory queue is packaged into an audio file.

4. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 3, characterized in that, The real-time extraction of valid call voice is specifically as follows: With a preset fixed time interval as one frame, the audio signal energy of each frame is obtained through an audio energy detection algorithm. Audio segments with a consecutive frame number greater than or equal to a preset frame number and an audio signal energy of each frame greater than or equal to a preset energy threshold are judged as valid call speech and extracted.

5. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 4, characterized in that, Before writing valid call audio to the memory queue, the following steps are also included: Insert a preset number of trailing frames at the end of each valid voice call.

6. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 1, characterized in that, The encrypted upload to the cloud server is specifically as follows: The Android capture box constructs a request body that conforms to a standard format, puts the audio file into it, and initiates a connection to the automatic speech recognition interface of the cloud server through the system network library; During the connection process, the Android capture box and the cloud server establish a secure encrypted channel through a preset encryption suite, and perform security verification on the cloud server's certificate. After completing the security verification, the Android data collection box generates authentication information for this request and puts it into the request header, which is then sent to the cloud server along with the request body. The cloud server verifies the authentication information, and returns an error signal if the verification fails.

7. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 6, characterized in that, The preset encryption suite specifically uses the TLS_AES_256_GCM_SHA384 encryption suite.

8. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 6, characterized in that, The cloud server verifies the identity verification information, specifically as follows: The authentication information is parsed from the request header, including digital signature, timestamp, random number, string used when concatenating signature, and ID identifier assigned to the Android collection box by the cloud server; The system detects whether the deviation between the timestamp and the time when the cloud server receives the request header exceeds a second preset duration. If yes, an error signal is returned; otherwise, a random number check is performed. The random number verification: The cloud server caches the random number after each successful verification, and the cache validity period is the same as the first preset period. If the current random number already exists in the cache, an error signal is returned; otherwise, a consistency check is performed. The consistency verification is as follows: The cloud server queries the corresponding key from the key database using the parsed ID identifier, concatenates the string used in the signature concatenation in the exact same order as the Android collection box to obtain the second string, uses the queried key to calculate the second string using the same signature algorithm as the Android collection box to obtain the second digital signature, and determines whether the second digital signature is consistent with the digital signature in the request header. If not, an error signal is returned.

9. The method for real-time acquisition and analysis of landline call voice based on the Android platform according to claim 1, characterized in that, The text conversion using automatic speech recognition technology specifically involves inputting the effective call speech into a pre-trained acoustic model, outputting a corresponding probability matrix, and then decoding it through a decoder to obtain the first text; the first text is then subjected to text normalization processing to obtain the final text. The acoustic model is trained as follows: a large amount of narrowband speech data is acquired as a dataset, and line distortion data augmentation technology is applied to the dataset to generate an expanded dataset containing various line noises. The acoustic model is trained using the expanded dataset as training data.

10. A real-time acquisition and analysis system for landline call voice based on the Android platform, employing the real-time acquisition and analysis method for landline call voice based on the Android platform as described in any one of claims 1 to 9, characterized in that, include: A 1-to-2 splitter for connecting a landline phone and an Android data acquisition box; An Android capture box is used to identify and capture valid call audio in real time and encrypt and upload it to a cloud server. The cloud server is used to decrypt the valid voice call, convert it into text using automatic speech recognition technology, and then perform emotion detection and intent recognition on the final text.

Images