70 results about "Audio segmentation" patented technology

The invention relates to the technical field of artificial intelligence, and provides a risk content identification method based on a multi-modal large model. The method comprises the following steps: identifying a forged part in audio; extracting bottom noise features and high-frequency features of the target face image, inputting the features into the image segmentation model, and positioning a forged area; dicing the target face image, and inputting a local area image and a global image obtained by dicing into a visual encoder to extract visual features; calculating the attention of the text features and the global image features on the local image features, and discarding the local image features with small attention; and inputting the output of the audio segmentation model and the image segmentation model, image features and questions into a large language model, and summarizing risk points. According to the method, multiple data sources are integrated, risk identification is carried out based on the multi-modal large model, the accuracy and robustness of identification are improved, multiple fraud means can be effectively coped with, and the problems that in the prior art, data of multiple modals cannot be coped with and interpretability is lacked are solved.

The invention discloses an audio stream adaptive decryption method, system and device and a storage medium. The method comprises the following steps: acquiring an MPEG-DASH list file and analyzing the MPEG-DASH list file; the standard DRM interface is initialized; performing feature extraction on the identified audio segmentation information in the audio segmentation mode, and dynamically adjusting a segmentation request and processing logic according to the extracted features to perform adaptive segmentation downloading and preprocessing; extracting the encryption parameter to identify an encryption mark of the target audio stream platform, and selecting a corresponding decryption algorithm according to the encryption mark to decrypt the current audio segment; and performing audio frame reconstruction on the decrypted data to form an updated audio stream, caching the updated audio stream to a preset cache region based on a self-adaptive buffer strategy, and outputting the final audio stream to ensure the continuity of the final audio stream and realize efficient parallel processing of decryption and playing, so that the decryption efficiency is improved, and the user experience is improved. And particularly, in a high-concurrency scene, the user experience is improved.

A system for audio segmentation for context change detection in speech is disclosed. The system accesses a speech signal of a user and identifies the user based on the voice print associated with the user. The system detects a potential context change between a first audio frame and a second audio frame of the speech signal. The system splits the speech signal into a first set of split audio frames based on the potential context change. The system generates a noisy speech signal by modulating a flicker noise signal into the speech signal. The system splits the noisy speech signal into a second set of split audio frames. The system detects a difference between the first and second sets of split audio frames. The system reconfigures a modulation of the speech signal with the flicker noise signal to reduce the difference between the first and second sets of audio frames.

A system for an audio slicing window selection for contextually splitting a speech signal is disclosed. The system identifies a first audio processing software algorithm that is assigned to a user. The system identifies a set of audio processing software algorithms and configures each of them with a respective audio slicing window. The system selects a second audio processing software algorithm, from among the set of audio processing software algorithms. The system selects one of the first and second audio processing software algorithms that is assigned an audio slicing window associated with the context of the speech signal. The system splits the speech signal using the selected audio processing software algorithm. The system determines whether the speech signal is split contextually. In response to determining that the speech signal is not split contextually, the selected audio processing software algorithm and / or the audio slicing window may be updated.

The invention relates to the technical field of speech recognition, and provides a multi-language subtitle generation method based on a Whisper and FunASR two-way speech recognition large model, and the method comprises the steps: inputting original audio segmentation fragments into the two-way speech recognition large model through a sliding window mode; based on the Pinyin similarity, comparing two output results of the two-way speech recognition large model to obtain a first recognition result and an existing result; based on the pinyin similarity, splicing the first recognition result and the existing result into a second recognition result; setting a next sliding window at the ending position of the second recognition result, and continuing to recognize the next audio clip until all audio clips are recognized, thereby obtaining a voice recognition result; and inputting a speech recognition result into the translation model to generate multilingual subtitle contents. The method is high in recognition accuracy, is high in robustness, is good in long audio processing effect, and is high in expandability.

The invention relates to an audio processing method and apparatus, a terminal device, a storage medium and a program product. The audio processing method comprises the steps of obtaining a first audio; segmenting the first audio into a plurality of second audios; determining playing parameters of each second audio; wherein the playing parameters are different, and sound effects obtained by playing the same second audio are different; and according to the playing parameters of the second audios, playing the corresponding second audios so as to realize playing of the first audios after audio processing. According to the embodiment of the invention, the sound effect after each second audio is played is improved, the situation that part of audio in the first audio is not matched with the playing parameter due to the fact that the first audio is played according to the same playing parameter is reduced, the situation that the sound effect is poor due to the fact that the playing parameter is not matched with the audio is also reduced, and the use experience of a user is improved. Manual operation of a user is not needed, so that operation steps of the user are reduced, the method is more intelligent and convenient, and the use experience of the user is improved.

The invention discloses an OSAHS screening method and related equipment based on sleep breath sounds, and the method comprises the steps: inputting an obtained audio signal into an OSAHS screening model, outputting an AHI value, and then obtaining an OSAHS symptom level; the training steps of the model include: performing audio segmentation and automatic labeling on sleep recording signals, and dividing tags into reliable tags and unreliable tags according to the reliability of the tags; training a noise label classifier by using a label correction algorithm, and iteratively identifying and correcting potential error labels in the unreliable labels in the training process of the noise label classifier; after the tag correction is completed, training an audio clip classifier by adopting the corrected tag; and fitting a linear regression model to predict an AHI value based on a classification result of the whole sleep recording signal. The label correction link is introduced in the model training process, potential wrong labels in unreliable labels are corrected, the accuracy of the model is effectively improved, and the method can be widely applied to the technical field of disease diagnosis equipment.

A method includes obtaining initial transcription for input natural speech; performing segmentation of initial transcription into text portions, based on punctuation marks in initial transcription; determining segment-level timestamps for text portions based on the input natural speech; performing audio segmentation on input natural speech, by cutting input natural speech based on segment-level timestamps, to obtain audio chunks; generating transcription portions for each of the audio chunks; merging transcription portions to form re-transcription; determining word-level timestamps for re-transcription, by aligning input natural speech against re-transcription; calculating silence time periods, each corresponding to silence between each two adjacent words of input natural speech, based on word-level timestamps; performing a final segmentation on input natural speech and re-transcription, based on silence time periods, to generate final audio segments and corresponding final transcription portions. The final audio segments and corresponding final transcription portions may be included in training dataset for training a model.

The invention relates to the technical field of rail traffic safety management, in particular to an audio and video and image fusion monitoring analysis system and method based on deep learning in a rail scene, and the system comprises a data collection module which is used for collecting audio data and video data in the rail scene; the audio processing module is used for receiving the collected audio data and carrying out noise removal and segmentation on the audio data to obtain a plurality of audio segments; the video processing module is used for receiving the collected video data and performing signal enhancement and splicing on the video data; and the data analysis module is used for keeping the plurality of audio segments and the spliced video consistent in time and completing a classification function based on a neural network. According to the invention, noise removal and audio segmentation are carried out on the audio data, image enhancement and splicing are carried out on the video data, and the audio data and the video data are fused, so that the classification accuracy of various events in the track scene is improved.

A method includes obtaining initial transcription for input natural speech; performing segmentation of initial transcription into text portions, based on punctuation marks in initial transcription; determining segment-level timestamps for text portions based on the input natural speech; performing audio segmentation on input natural speech, by cutting input natural speech based on segment-level timestamps, to obtain audio chunks; generating transcription portions for each of the audio chunks; merging transcription portions to form re-transcription; determining word-level timestamps for re-transcription, by aligning input natural speech against re-transcription; calculating silence time periods, each corresponding to silence between each two adjacent words of input natural speech, based on word-level timestamps; performing a final segmentation on input natural speech and re-transcription, based on silence time periods, to generate final audio segments and corresponding final transcription portions. The final audio segments and corresponding final transcription portions may be included in training dataset for training a model.

The invention discloses a psychological state monitoring method and system based on a deep learning voice emotion vector, which can accurately measure the psychological state of a pilot in a simulated flight training process, and overcome the problems of strong subjectivity, complex equipment, easy interference, non-deep voice emotion analysis and the like in the existing measurement method. And a powerful guarantee is provided for flight safety. The method comprises the following steps: acquiring voice data of a tested person, and performing audio processing on the acquired voice data to obtain an effective voice audio file set after audio segmentation and de-muting; adopting a pre-trained deep voice emotion calculation model to extract a high-dimensional voice emotion vector of each audio clip; and sequentially inputting the extracted voice emotion vectors into the trained psychological state evaluation model to generate psychological characteristic indexes.

The invention relates to the technical field of audio synchronous playing, in particular to an audio synchronous playing method and device based on remote WiFi. The method comprises the following steps: acquiring an audio uploading risk value of audio source equipment and an audio receiving risk value of audio receiving equipment based on an audio synchronous playing request, and judging whether to respond to the audio synchronous playing request based on the audio uploading risk value and the audio receiving risk value; according to the audio segment management sequence, performing content segment labeling on target audio data to be uploaded to obtain audio labeling segments, calculating audio feature parameters of each audio labeling segment, forming an information frame by the audio labeling segments, the audio feature parameters corresponding to the audio labeling segments and a preset playing timestamp, and sending the information frame to the server; the audio synchronization control system generates the information frame instruction and sends the information frame instruction to the audio source device through WiFi so as to instruct the audio source device to send the information frame to the audio receiving device, and the accuracy of audio synchronization playing can be improved.

A method and an electronic device for generating an anonymized audio output are provided. The method, executable by the electronic device, comprises acquiring an audio recording of a speaker; stochastically determining a base pitch value based on at least a first probabilistic function; segmenting the original audio input into a plurality of audio segments, each of the plurality of audio segments being associated with a respective pitch. For each audio segment, the method further comprises generating a pitch adjustment value using a combination of the base pitch value of the segment and a value determined using a second probabilistic function; generating an adjusted audio segment by adjusting the pitch of the audio segment using the pitch adjustment value, the adjusted audio segment having an adjusted pitch that is different from the original pitch; generating the anonymized audio output by combining the adjusted audio segments.

This invention relates to the field of artificial intelligence technology and provides a risk content identification method based on a multimodal large model. The method includes: identifying forged parts in audio; extracting background noise features and high-frequency features from a target face image and inputting them into an image segmentation model to locate the forged region; segmenting the target face image into blocks, and inputting the resulting local region images and global images into a visual encoder to extract visual features; calculating the attention of text features and global image features to local image features, discarding local image features with low attention; and inputting the outputs of the audio segmentation model and image segmentation model, image features, and questions into a large language model to summarize the risk points. This invention improves the accuracy and robustness of risk identification by integrating multiple data sources and using a multimodal large model for risk identification. It can also effectively counter various fraud methods and solves the problems of existing technologies being unable to handle multimodal data and lacking interpretability.

The present application relates to the technical field of audio processing, and specifically provides a harmony recognition and its model training method, program product, device, and storage medium. The harmony recognition model training method segments the audio data to be trained based on beat data, extracts features from multiple audio segmentation segments obtained by segmentation, and inputs the extracted audio feature data to be trained into the harmony recognition model to be trained to obtain the model recognition result output by the harmony recognition model to be trained; according to the model recognition result and the harmony annotation result of the audio data to be trained, the internal parameters of the harmony recognition model to be trained are optimized to obtain a trained harmony recognition model. Since the beat data can better characterize the chord change positions of the audio data to be trained, based on the beat data, the segmentation and positioning of the audio data to be trained can be more accurately achieved, thereby improving the harmony recognition accuracy of the trained harmony recognition model.

The application provides an audio segmentation method and device, electronic equipment and storage medium, wherein the method comprises: obtaining audio to be segmented; extracting acoustic features of each frame in the audio to be segmented, and based on the acoustic features of each frame, performing semantic boundary sequence labeling on the audio to be segmented to obtain semantic boundary labeling results of each frame; and based on the semantic boundary labeling results of each frame, performing segmentation on the audio to be segmented. The method, device, electronic equipment and storage medium provided by the application can assist semantic segmentation based on tone and pause information in the acoustic features of each frame, retain complete semantic information of the audio, and avoid punctuation recognition errors, thereby improving the accuracy and reliability of audio segmentation. Furthermore, the method can be applied to a cascaded speech translation system and an end-to-end speech translation system, thereby expanding the application range of audio segmentation.

The invention relates to the technical field of audio recognition, and relates to an audio transcription method and system based on intelligent analysis, and the method comprises the steps: carrying out the file verification and file slicing uploading of a to-be-transcribed file, and obtaining an uploaded audio file; performing audio segmentation on the uploaded audio file, and extracting a voice text after audio segmentation to obtain a primary transcription text; calculating an analysis transcription progress and a real-time transcription progress of the uploaded audio file, performing maximum screening on the analysis transcription progress and the real-time transcription progress to obtain a transcription progress, and updating the primary transcription text into a standard transcription text according to the transcription progress; performing feature activation on the text context feature of the standard transcription text to obtain a text abstract feature; performing feature style decoding on the text abstract features according to a style keyword input by a user to obtain a decoded text abstract; and combining and splicing the standard transcriptional text and the decoded text abstract to obtain an audio transcriptional file. According to the invention, the audio transcription efficiency can be improved.

The present disclosure provides a speech processing method and device, the method: obtaining a syllable in the audio to be processed as the current speech segment, judging whether the current speech segment meets the audio segmentation condition and the interactive response condition, the audio segmentation condition is set based on the ASR recognition process, and the interactive response condition is set based on the NLP processing process and the matching degree of the NLP processing result and the feedback message. If the current speech segment does not meet the audio segmentation condition and the interactive response condition, the next syllable in the audio to be processed is merged into the current speech segment, and the step of judging whether the current speech segment meets the audio segmentation condition and the interactive response condition is performed until the target speech segment that meets the audio segmentation condition and the interactive response condition is determined. Output a feedback message that matches the target speech segment. The method greatly improves the audio segmentation efficiency, improves the reaction speed of robots or intelligent devices in human-computer communication, and improves the human-computer interaction experience.

This application discloses a method and system for marine mammal audio retrieval based on biohash ciphertext, relating to the field of audio processing. The method comprises: extracting sub-band CQT energy-entropy ratio features from the original audio, constructing a key-address index table, and constructing a biohash sequence based on the key-address index table; segmenting the original audio using a dual-threshold segmentation method based on short-time energy and spectral distribution variance to obtain short-time audio segments, mapping the short-time audio segments to the key-address index table, and reconstructing the biohash sequence; performing an improved AES encryption on the original audio to obtain ciphertext audio; constructing a hash index table based on the mapping relationship between the reconstructed biohash sequence, key, and ciphertext audio, and uploading the ciphertext audio and hash index table to the cloud for retrieval. This application uses an audio segmentation algorithm to reconstruct the biohash sequence, improving retrieval efficiency, while also using an improved AES encryption algorithm to ensure data security.

Disclosed in the present invention are a computer-implemented method for adaptive decryption of an audio stream, and a device and a storage medium. The method comprises: acquiring an MPEG-DASH manifest file and parsing same; initializing a standard DRM interface; performing feature extraction on audio segment information in an identified audio segmentation mode, and on the basis of the extracted features, dynamically adjusting a segment request and processing logic, so as to perform adaptive segment downloading and pre-processing; extracting encryption parameters to identify an encryption flag of a target audio stream platform, and on the basis of the encryption flag, selecting a corresponding decryption algorithm to decrypt a current audio segment; and performing audio frame reconstruction on decrypted data to form an updated audio stream, buffering the updated audio stream into a preset buffer on the basis of an adaptive buffer policy, and outputting a final audio stream, so as to ensure the continuity of the final audio stream and realize efficient parallel processing of decryption and playback, thereby improving the decryption efficiency, and enhancing the user experience especially in high-concurrent scenarios.

The present invention relates to the field of data processing technology, and in particular to a multimodal fusion system based on audiovisual signs during movement, comprising: an audiovisual module, which is used to feed back three-dimensional scene images and audio to users; a binding module, which is used to receive audio fed back to users from the audiovisual module, segment the audio, and bind each group of segmented sub-audio to a three-dimensional model in the three-dimensional scene image fed back to users by the audiovisual module; and a configuration module, which is used to configure a three-dimensional scene space rectangular coordinate system, perceive user movement parameters in real time, and identify user position information in the three-dimensional scene. The present invention provides feedback direction fusion for audio feedback of users in virtual scenes by segmenting audio and binding sub-audio obtained by segmentation to virtual models in virtual scenes, so that when users walk in virtual scenes, the virtual scene audio can be fed back to users more realistically and vividly as the virtual scenes change, thereby improving the user's virtual scene experience.

The application provides an audio segmentation method and device, electronic equipment and storage medium, wherein the method comprises: determining a double-channel audio to be segmented; respectively performing mute section labeling on a first channel audio and a second channel audio in the double-channel audio to obtain a mute section in the first channel audio and a mute section in the second channel audio; determining a common mute separation point in the double-channel audio based on the mute section in the first channel audio and the mute section in the second channel audio, and performing segmentation on the first channel audio based on the common mute separation point to obtain a plurality of first segmented audio sections; performing mute section removal on each first segmented audio section to obtain each second segmented audio section; and performing customer audio combination based on a voiceprint feature of each second segmented audio section to obtain customer audio in units of customers, thereby overcoming the defect that a fixed-time segmentation cannot distinguish customers, realizing audio segmentation in units of customers, and providing assistance for different service quality inspection and service evaluation.

The present invention relates to the field of human-computer technology, and provides a method for quickly extracting telephone channel data, which includes the steps of: obtaining valid audio in the channel data to be extracted, where the channel data to be extracted is the channel data collected during a call between at least two people; splitting the valid audio into several segments; determining the segments belonging to the first speaker and the speaker change segments by comparing the voiceprint features of adjacent segments; clustering the segments belonging to the first speaker in the valid audio according to the segments of the first speaker and the change segments to obtain the valid audio of the first speaker. Based on the method provided by the embodiments of the present invention, the valid audio of the speaker can be automatically and quickly extracted without relying on manual labor, saving costs and improving the data collection efficiency. In addition, a device for quickly extracting telephone channel data and a computer-readable storage medium are also provided.

A system for audio segmentation for context change detection in speech is disclosed. The system accesses a speech signal of a user and identifies the user based on the voice print associated with the user. The system detects a potential context change between a first audio frame and a second audio frame of the speech signal. The system splits the speech signal into a first set of split audio frames based on the potential context change. The system generates a noisy speech signal by modulating a flicker noise signal into the speech signal. The system splits the noisy speech signal into a second set of split audio frames. The system detects a difference between the first and second sets of split audio frames. The system reconfigures a modulation of the speech signal with the flicker noise signal to reduce the difference between the first and second sets of audio frames.

The invention provides a piano keystroke force detection system, which relates to the field of electronic digital data processing, and comprises an audio acquisition module, an audio segmentation module, a force detection module and a standard correction module. The audio acquisition module is used to collect audio information generated when keystrokes are made, the audio segmentation module is used to segment and extract audio information of each keystroke from the audio information, the force detection module is used to analyze the independent keystroke audio information to measure the corresponding keystroke force, and the standard correction module is used to provide a standard keystroke force and correct the parameters in the force detection module. The system detects the piano keystroke force in a non-contact manner, so that the detection process does not affect or interfere with the playing of the piano, and can be suitable for the piano practice process.

Methods, apparatus, systems, and articles of manufacture to fingerprint an audio signal. An example apparatus disclosed herein includes an audio segmenter to divide an audio signal into a plurality of audio segments, a bin normalizer to normalize the second audio segment to thereby create a first normalized audio segment, a subfingerprint generator to generate a first subfingerprint from the first normalized audio segment, the first subfingerprint including a first portion corresponding to a location of an energy extremum in the normalized second audio segment, a portion strength evaluator to determine a likelihood of the first portion to change, and a portion replacer to, in response to determining the likelihood does not satisfy a threshold, replace the first portion with a second portion to thereby generate a second subfingerprint.

A system and method for enhancing a call characterization of voice calls is disclosed. The method comprises: receiving, at the platform (1), the voice call (30) segmented in audio packets; generating a Call Detail Record "CDR" (12) comprising call parameters; processing, by the Digital Signal Processing "DSP" engine (300), the audio packets, wherein the processing step in turn comprises: computing audio spectral representations; and, accumulating a predetermined number of audio packets (30.1). The method further comprises: evaluating, by the Audio Segmentation "AS" engine (301), an audio type present in the accumulated audio packets; generating temporal marks (31), by the Audio Segmentation "AS" engine (301), indicating which audio segment correspond to each audio type; computing metrics (32), by the Metrics Computation "MC" engine (302), using the accumulated audio packets and the temporal marks; generating an enhanced Call Detail Record "eCDR" (33) by adding the metrics (32) to the "CDR".

The present invention provides a method, system and device for transcribing multilingual long recordings with adaptive noise suppression, including: performing audio segmentation on the multilingual long recording to obtain multiple audio segments; performing spectrum analysis on each audio segment to obtain corresponding spectrum features; performing noise detection on each audio segment according to the spectrum features to determine the noise segment; performing adaptive noise suppression on each noise segment in the multilingual long recording to obtain denoised multilingual audio; performing voice activity detection on the denoised multilingual audio to determine voice segments and non-voice segments; performing feature extraction on the voice segments to obtain voice features; inputting the voice features into a pre-trained multilingual transcription model for transcription to obtain corresponding text results. In the present invention, by performing adaptive noise suppression on each noise segment in the multilingual long recording, the interference of noise is effectively reduced and the voice quality is improved.

Embodiments are disclosed for generating an audio segmentation of an audio sequence using deep embeddings. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving an input including an audio sequence and extracting features for each frame of the audio sequence, where each frame is associated with a beat of the audio sequence. The method may further comprise clustering frames of the audio sequence into one or more clusters based on the extracted features and generating segments of the audio sequence based on the clustered frames, where each segment includes frames of the audio sequence from a same cluster. The method may further comprise constructing a multi-level audio segmentation of the audio sequence and performing a segment fusioning process that merges shorter segments with neighboring segments based on cluster assignments.

The embodiment of the invention provides a pathological voice detection model training method based on multi-level data modeling. The method comprises the following steps: inputting a dialogue audio into a pathological voice detection model; in the session layer, an encoder is used for encoding continuous audio clips obtained through dialogue audio segmentation, extracting depth features of the audio clips, capturing global context information, obtaining pathological pseudo-labels of the audio clips through depth feature classification, and determining session level loss based on the pathological pseudo-labels; in the fragment layer, modeling is carried out by taking the pathology pseudo labels as supervision signals so as to position audio fragments containing symptom features, depth features of the audio fragments are classified, a predicted pathology classification result is obtained, and fragment-level loss is determined based on the predicted pathology classification result; and performing end-to-end training on the pathological voice detection model based on the loss. According to the embodiment of the invention, the training remarkably improves the performance of the model under a very small amount of annotation data, and achieves the generalization of cross-language / disease species.