25 results about "Grapheme" patented technology

The invention provides a Suzhou dialect speech recognition system and method based on a tone track neural field, and the system comprises a tone track neural field module which is used for modeling the tone change of Suzhou dialects into a continuous space-time neural field; the bidirectional semantic memory network module comprises a forward prediction memory bank and a backward correction memory bank; the phoneme-font coupling error corrector is used for realizing polyphone disambiguation and homonym error correction by establishing association mapping between a phoneme sequence and a font sequence; the semantic entropy calculation module is used for evaluating the uncertainty of the recognition result; and the self-adaptive fusion decision module is used for generating a final recognition text. According to the invention, through an online learning mechanism, the system can continuously accumulate experience from actual use, automatically discover a new language mode and update an identification strategy. The self-improvement capability enables the system to adapt to the dynamic change of languages, and the performance is continuously improved along with the increase of the use time. Each use of the user helps the system to become more intelligent and accurate.

Methods, systems (126), and computer program products for domain-adaptive speech recognition using artificial intelligence are provided herein. A computer-implemented method comprises generating a set of speech data candidates, each speech data candidate having one or more graphemes, by processing a sequence of phonemes associated with input speech data using an artificial intelligence-based data conversion model (104) (302); determining a subset of graphemes from the set of speech data candidates for a target pair of one or more phonemes and one or more graphemes (304); generating a first speech recognition output by processing at least a portion of the subset of graphemes using at least one biased language model (108) and an artificial intelligence-based speech recognition model (306);Generating a second speech recognition output by replacing at least a portion of the subset of graphemes in the first speech recognition output with at least one of the one or more graphemes from the target pair (308); and performing automated operations based on the second speech recognition output (310).

Methods, systems (126), and computer program products for domain adaptive speech recognition using artificial intelligence are provided herein. A computer-implemented method comprising: generating a set of language data candidates (302) by processing a sequence of phonemes associated with input speech data using an artificial intelligence-based data transformation model (104), each language data candidate comprising one or more graphemes; determining, for a target pair of phonemes and graphemes, a subset of graphemes from the candidate set of language data (304); generating a first speech recognition output (306) by processing the subset of words using at least one bias language model (108) and an artificial intelligence-based speech recognition model; generating a second speech recognition output by replacing at least a portion of a subset of graphemes in the first speech recognition output with at least one grapheme from the target pair (308); and performing an automated action (310) based on the second speech recognition output.

An individualized decodable text system and non-transitory computer-readable medium, among other materials, are disclosed for generating customized reading materials tailored to students. The system includes processors configured to prompt a user to select subsets of phonemes and corresponding graphemes. Based on these selections, the system identifies or constructs a set of words containing only the selected phonemes and graphemes, excluding unselected ones, and generates phrases or sentences for student practice. Real-time feedback on pronunciation may be provided through word and phoneme-level scoring. The system may leverage a pre-indexed phoneme-grapheme database or rule-based word construction, using dynamic filtering and efficient querying techniques for adaptability and scalability. This approach facilitates individualized instruction by aligning reading materials with each student's developmental needs to improve decoding accuracy and promote reading fluency.

Techniques are described for recognizing a spoken wake word (WW) or a spoken command for a human-machine interface using a speech recognition system that does not require WW / command matching speech data for training. The system uses the text or grapheme representation of the WW or commands for pre-deployment training. The technique involves receiving a target phrase for recognition by a speech recognition model. It also involves analyzing a sequence of acoustic units representative of the target phrase as it is spoken to generate offline analysis data. Furthermore, the technique includes building the speech recognition model based on this offline analysis data to decode speech signals of the target phrase according to the acoustic units.The technique also includes processing speech based on the speech recognition model to detect the presence of the target phrase.

The present invention discloses a multi-style personalized Tibetan speech synthesis model for low-resource conditions, including a speaker style encoder and a sequentially connected grapheme-to-phoneme conversion module, a text encoder, a variance adapter, a mel spectrum decoder and a HifiGAN vocoder; the synthesis model is trained using a "pre-training + meta-learning" model algorithm; the model parameters that can be learned by the deep learning model are converted to i Divided into parameters related to rhythm i p , parameters related to the speaker i s and other remaining parameters #imgabs0#, i.e. #imgabs1# i p and i s The model includes learnable model parameters for the speaker style encoder, and #imgabs2# includes learnable model parameters for the text encoder, mel-spectrogram decoder, and variance adapter. This synthetic model can be applied to intelligent bilingual teaching products, as well as Tibetan spoken dialogue, multimedia information processing, and other human-computer interaction fields.

A method for encoding musical information includes accessing a database storing musical information for a song in the form of a string of notes and text. The text is divisible into a string of graphemes, each grapheme corresponding to a note in the string of notes. A reference note is selected from the string of notes. Each note in the string of notes is associated with a degree selected from among 12 degrees in an octave, relative to a reference degree corresponding to the reference note. Twelve different colors are associated with the 12 degrees. The different colors are selected according to a degree-color function linking each degree to a respective color. The notes in the string are divided into different groups corresponding to different musical octaves. One or more specific printing attributes are associated with each musical octave. For each grapheme, data is stored indicating the color associated with the degree of the note corresponding to the grapheme and one or more specific printing attributes associated with the musical octave to which the note corresponding to the grapheme belongs.

Systems and methods are described for grapheme-phoneme correspondence learning. In an example, a display of a device is caused to output a grapheme graphical user interface (GUI) that includes a grapheme. Audio data representative of a sound made by the human user is received based on the grapheme shown on the display. A grapheme-phoneme model can determine whether the sound made by the human corresponds to a phoneme for the displayed grapheme based on the audio data. The grapheme-phoneme model is trained based on augmented spectrogram data. A speaker is caused to output a sound representative of the phoneme for the grapheme to provide the human with a correct pronunciation of the grapheme in response to the grapheme-phoneme model determining that the sound made by the human does not correspond to the phoneme for the grapheme.

The application relates to the technical field of computer-aided design, in particular to a legend recognition method and device and electronic equipment, the method comprising the following steps: obtaining a target topology relationship of a target legend and a to-be-recognized legend, the target topology relationship being the positional relationship between a first target shape category formed by target graphemes in the target legend and other graphemes in the target legend; recognizing the shape category formed by each grapheme in the to-be-recognized legend, and determining all shape categories in the to-be-recognized legend; searching for a second target shape category identical to the first target shape category in all the shape categories; determining a to-be-recognized topology relationship based on the second target shape category and the positional relationship of other graphemes in the to-be-recognized legend; and matching the target topology relationship with the to-be-recognized topology relationship to determine a legend recognition result. The shape category formed by graphemes is used as the basis for legend recognition, which guarantees the accuracy of legend recognition, and the shape category is used to reduce the matching range, thereby improving the efficiency of legend recognition.

Techniques are described for recognizing a spoken wake word (WW) or a spoken command for a human-machine interface using a speech recognition system that does not require WW / command matching speech data for training. The technique trains a tokenizer and involves decomposing a word from a database into a multitude of combinations of unique subwords by splitting the word at a variety of different points for each combination. The word comprises one or more written units and one or more corresponding acoustic units. The technique maps the acoustic units that make up the word to the written units that make up the word to create an acoustic unit-to-written unit mapping.The technique involves assigning a subset of acoustic units to each of the unique subwords based on the acoustic unit-to-written unit mapping, in order to generate an acoustic unit-to-subword mapping for the word. The technique accumulates the acoustic unit-to-subword mappings for a large number of words from the database to create a subword probability dictionary.

Systems and methods are described for grapheme-phoneme correspondence learning. In an example, a display of a device is caused to output a grapheme graphical user interface (GUI) that includes a grapheme. Audio data representative of a sound made by the human user is received based on the grapheme shown on the display. A grapheme-phoneme model can determine whether the sound made by the human corresponds to a phoneme for the displayed grapheme based on the audio data. The grapheme-phoneme model is trained based on augmented spectrogram data. A speaker is caused to output a sound representative of the phoneme for the grapheme to provide the human with a correct pronunciation of the grapheme in response to the grapheme-phoneme model determining that the sound made by the human does not correspond to the phoneme for the grapheme.

A language-independent dictionary-trained grapheme-to-phoneme converter and text-to-speech engine for improved speech recognition is disclosed. Techniques are described for recognizing a spoken wakeup word (WW) or command using a speech recognition system that does not need to be trained with any speech data that matches the WW / command for a human machine interface. Techniques train a word segmentation device and include decomposing a word into a plurality of combinations by splitting the word from a database at a plurality of different points for each of the plurality of combinations of unique sub-words. The word includes one or more writing units and one or more corresponding acoustic units. Techniques map acoustic units constituting a word to writing units constituting the word to generate an acoustic unit-to-writing unit mapping. The technique includes assigning a subset of the acoustic units to each of the unique sub-words based on the acoustic unit-to-write unit map to generate an acoustic unit-to-sub-word assignment of the word. Techniques accumulate acoustic units to sub-word assignments for a plurality of words from a database to create a sub-word likelihood dictionary.

Techniques are described for recognizing a spoken wake word (WW) or a spoken command for a human-machine interface using a speech recognition system that does not require WW or command-matching speech data for training. The system uses the text or grapheme representation of the WW or commands for pre-deployment training. The technique involves the system receiving a text representation of a target phrase in a target language. It includes training an acoustic model based on a speech database to distinguish speech signals according to the acoustic units of the target language. The training of the acoustic model is independent of the target phrase.It involves creating a recognition model based on the textual representation of the target phrase and the acoustic model to recognize the target phrase in speech, and processing speech from a speaker based on the acoustic model and the recognition model to detect the presence of the target phrase.

Methods, systems, and computer program products for domain adaptive speech recognition using artificial intelligence are provided herein. A computer-implemented method includes generating a set of language data candidates, each language data candidate comprising one or more graphemes, by processing a sequence of phonemes related to input speech data using an artificial intelligence-based data conversion model; determining, for a target pair of phonemes and graphemes, a subset of graphemes from the set of language data candidates; generating a first speech recognition output by processing the subset of graphemes using at least one biasing language model and an artificial intelligence-based speech recognition model; generating a second speech recognition output by replacing at least a portion of the subset of graphemes in the first speech recognition output with at least one of the graphemes from the target pair; and performing automated actions based on the second speech recognition output.

Embodiments of the present application provide a training sample generation method and device, and a model training method, relating to the technical field of artificial intelligence, comprising: obtaining a plurality of original training samples; for each original training sample, obtaining video features and text features of the original training sample; determining a plurality of target image regions and a plurality of target graphemes in the original training sample based on the video features and the text features of the original training sample; calculating quality data of the original training sample based on the video features of the plurality of target image regions and the text features of the plurality of target graphemes; and selecting each original training sample based on a first number of original training samples with quality data less than a first preset threshold to obtain a target training sample. The embodiments can improve the training effect of the cross-modal model, save time and labor costs, improve the generation efficiency of the training sample, and further improve the training efficiency of the cross-modal model.

A computer-implemented system and method generate personalized text based on statistics derived from input received from a user representing the user's attempts to decode graphemes into phonemes. Such statistics may be measured and recorded at the grapheme-phoneme level, and may include substitutions, insertions, deletions, and correct utterances of phonemes by the user when reading text. A language model may be trained based on characteristics of the user, such as the user's age and / or reading grade level, and the personalized text may be generated after such training of the language model. Generating the personalized text may include generating a text creation prompt based on the statistics. The resulting text creation prompt may include a set of target words. The text creation prompt may be provided to the language model, which may generate the personalized text in response. The personalized text may include some or all of the target words.

A system and method for reading development in children with dyslexia using word-to-graphic transformations are disclosed. The method includes presenting words to the learner and converting the words into corresponding graphical representations that visually depict the meaning of the words. The method further includes associating phonemes with parts of the graphical representation to help the learner connect speech sounds with written letters. Next, the method includes progressively simplifying the graphical representation of words as the learner's reading proficiency improves, transitioning them toward recognizing standard written words. The method also includes tracking the learner's progress through a proficiency tracking module, which dynamically adjusts the complexity of future word transformations based on performance data. Additionally, the method provides real-time auditory and visual feedback during the learning process to reinforce phoneme-grapheme associations. The system operates on an interactive platform, such as a tablet, allowing personalized learning experiences based on the child's individual progress.

A kind of oracle bone script variant character classification method based on context mask prediction.This application relates to the technical field of ancient character digital processing and natural language processing, and aims at the problem that it is difficult to determine the standard character attribution by only relying on the appearance of the character shape in the existing oracle bone script variant character classification, and proposes the following solution: convert the oracle bone script sentence into a character shape classification code sequence, construct a mask training sample based on the character position with a determined standard character category, and train a context mask prediction model;Mask the target variant character position in the sentence to be classified, and output the candidate character shape classification code according to the context character position by the model, and merge it into the standard character category.The application is suitable for automatic classification of oracle bone script variant characters and ancient character corpus collation.

Methods, systems (126), and computer program products for domain adaptive speech recognition using artificial intelligence are provided herein. A computer-implemented method includes generating a set of language data candidates, each language data candidate comprising one or more graphemes, by processing a sequence of phonemes related to input speech data using an artificial intelligence-based data conversion model (104) (302); determining, for a target pair of phonemes and graphemes, a subset of graphemes from the set of language data candidates (304); generating a first speech recognition output by processing the subset of graphemes using at least one biasing language model (108) and an artificial intelligence-based speech recognition model (306); generating a second speech recognition output by replacing at least a portion of the subset of graphemes in the first speech recognition output with at least one of the graphemes from the target pair (308); and performing automated actions based on the second speech recognition output (310).

The system and method for generating realistic lip synchronization for an AI avatar. The lip synchronization process begins by receiving input data. The input data can either be text input or audio stream. If the input data is text input, a text-to-speech (TTS) module converts it into speech while generating word-level timestamps. If the input data is the audio stream, an automatic speech recognition (ASR) module transcribes the spoken content and provides word-level timestamps. The transcribed text is transformed into a sequence of phonemes using a grapheme-to-phoneme conversion system. The phonemes are mapped to visemes based on their corresponding mouth shapes using a predefined phoneme to viseme mapping table. Then the visemes are mapped to the blendshapes of the AI avatar using image similarity comparison. The selected blendshapes are animated to generate synchronized animation of the AI avatar, with transitions smoothed to ensure natural lip movements and facial expressions.

Techniques are described for recognizing a spoken wake word (WW) or a spoken command for a human-machine interface using a speech recognition system that does not require WW or command-matching speech data for training. The system uses the text or grapheme representation of the WW or commands for pre-deployment training. The technique includes generating a statistical matrix that characterizes a phonetic model of an acoustic model that distinguishes speech signals according to a variety of acoustic units of a language. It further includes creating a speech recognition model based on the statistical matrix to recognize a target phrase in the speech. It also includes processing an audio signal based on the acoustic model and the speech recognition model to detect the presence of the target phrase.

The application provides a speech synthesis method and device, electronic equipment and storage medium, and relates to the technical field of speech synthesis. The method first acquires a text to be synthesized, then determines pronunciation features of the text to be synthesized according to loaded language rules, and then performs speech synthesis on the text to be synthesized according to the pronunciation features through a speech synthesis model to obtain target synthesized speech. The pronunciation features of the text to be synthesized can be determined from the dimensions of word formation characteristics, context dependence between graphemes of the text to be synthesized and the like according to the language rules, so that the pronunciation features of the text to be synthesized are more in line with language characteristics, thereby increasing the accuracy of the obtained pronunciation features of the text to be synthesized. The speech synthesis model performs speech synthesis on the text to be synthesized according to the pronunciation features with higher accuracy, thereby improving the quality of speech synthesis.

The invention discloses data-free speech recognition. Techniques are described for recognizing a spoken wakeup word (WW) or command using a speech recognition system that does not need to be trained with any speech data that matches the WW or command for a human machine interface. The system is trained using a text or word representation of a WW or command prior to deployment. The technique includes receiving, by a system, a textual representation of a target phrase in a target language. The technique includes training an acoustic model based on a speech database to distinguish speech signals according to acoustic units of a target language. The training of the acoustic model is independent of the target phrase. The technique includes constructing a recognition model to recognize a target phrase in speech based on a textual representation of the target phrase and an acoustic model; and processing the speech from the speaker based on the acoustic model and the recognition model to detect the presence of the target phrase.

Embodiments of the present application provide a phoneme-based speech domain migration method, system and electronic device. The method comprises: performing grapheme-to-phoneme conversion on target domain text to obtain a target domain phoneme sequence; converting the target domain phoneme sequence into a plurality of target domain phoneme N-gram sequences according to a phoneme N-gram dictionary; generating target domain speech segments with speaker diversity using the plurality of target domain phoneme N-gram sequences; and generating synthesized audio for the target domain based on the target domain speech segments. Embodiments of the present application use a dictionary constructed from speech segments of basic phoneme n-grams to generate speech for the text of the target domain, the splicing and synthesis method guided by phonemes has the ability to model the connected reading between words, and because the audio segments for splicing and synthesis come from a large amount of real speech, the synthesized speech has better speaker diversity, and the required computing resources are reduced, avoiding overfitting of the ASR model on the synthesized data.

The invention provides a speech synthesis method and device, electronic equipment and a storage medium, and relates to the technical field of speech synthesis, and the method comprises the steps: firstly obtaining a to-be-synthesized text, then determining the pronunciation characteristics of the to-be-synthesized text according to a loaded language rule, and then carrying out the speech synthesis of the to-be-synthesized text according to the pronunciation characteristics through a speech synthesis model, according to the method, the pronunciation characteristics of the to-be-synthesized text can be determined from dimensions such as word formation characteristics and context dependence between the graphemes of the to-be-synthesized text according to language rules, so that the pronunciation characteristics of the to-be-synthesized text better conform to language characteristics, and the accuracy of the obtained pronunciation characteristics of the to-be-synthesized text is improved; the speech synthesis model carries out speech synthesis on the to-be-synthesized text according to the pronunciation features with higher accuracy, and the speech synthesis quality can be improved.