Systems and methods provide for text normalization or inverse text normalization using a hybrid language system that combines rule-based processing with neural or learned processing. For example, a hybrid rule-based and neural approach identifies semiotic tokens within a textual input and generates a set of potential plain-text conversions of the semiotic tokens. The plain-text conversions are weighted and evaluated by a trained language model that rescores the plain-text conversion based on context to identify a highest scoring plain-text conversion for further processing within a language system pipeline.

Disclosed herein are systems, methods, and computer readable-media for providing an automatic synthetically generated voice describing media content, the method comprising receiving one or more pieces of metadata for a primary media content, selecting at least one piece of metadata for output, and outputting the at least one piece of metadata as synthetically generated speech with the primary media content. Other aspects of the invention involve alternative output, output speech simultaneously with the primary media content, output speech during gaps in the primary media content, translate metadata in foreign language, tailor voice, accent, and language to match the metadata and/or primary media content. A user may control output via a user interface or output may be customized based on preferences in a user profile.

A method and system is disclosed for speech synthesis of streaming text. At a text-to-speech (ITS) system, a real-time streaming text string having a starting point and an ending point may be received, and a first sub-string comprising a first portion of the text string received from an initial point to a first trigger point may be accumulated. The initial point is no earlier than the starting point and is prior to the first trigger point, and the first trigger point is no further than the ending point. A punctuation model of the ITS system may be applied to the first sub-string to generate a pre-processed first sub-string comprising the first sub-string with added grammatical punctuation as determined by the punctuation model. TTS synthesis processing may be applied to at least the pre-processed first sub-string to generate first synthesized speech, and audio play out of the first synthesized speech produced.

According to an embodiment, a text-to-speech device includes a receiver to receive an input text containing a peculiar expression; a normalizer to normalize the input text based on a normalization rule in which the peculiar expression, a normal expression of the peculiar expression, and an expression style of the peculiar expression are associated, to generate normalized texts; a selector to perform language processing of each normalized text, and select a normalized text based on result of the language processing; a generator generate a series of phonetic parameters representing phonetic expression of the selected normalized text; a modifier modifies a phonetic parameter in the normalized text corresponding to the peculiar expression in the input text based on a phonetic parameter modification method according to the normalization rule of the peculiar expression; and a output unit to output a phonetic sound synthesized using the series of phonetic parameters including the modified phonetic parameter.

Disclosed herein are systems, methods, and computer readable-media for providing an automatic synthetically generated voice describing media content, the method comprising receiving one or more pieces of metadata for a primary media content, selecting at least one piece of metadata for output, and outputting the at least one piece of metadata as synthetically generated speech with the primary media content. Other aspects of the invention involve alternative output, output speech simultaneously with the primary media content, output speech during gaps in the primary media content, translate metadata in foreign language, tailor voice, accent, and language to match the metadata and/or primary media content. A user may control output via a user interface or output may be customized based on preferences in a user profile.

A singing voice synthesis data editing method includes adding, to singing voice synthesis data, a piece of virtual note data placed immediately before a piece of note data having no contiguous preceding piece of note data, the singing voice synthesis data including: multiple pieces of note data for specifying a duration and a pitch at which each note that is in a time series, representative of a melody to be sung, is voiced; multiple pieces of lyric data associated with at least one of the multiple pieces of note data; and a sequence of sound control data that directs sound control over a singing voice synthesized from the multiple pieces of lyric data, and obtaining the sound control data that directs sound control over the singing voice synthesized from the multiple pieces of lyric data, and that is associated with the piece of virtual note data.

Systems and methods for providing non-lexical cues in synthesized speech are described herein. Original text is analyzed to determine characteristics of the text and/or to derive or augment an intent (e.g., an intent code). Non-lexical cue insertion points are determined based on the characteristics of the text and/or the intent. One or more nonlexical cues are inserted at insertion points to generate augmented text. The augmented text is synthesized into speech, including converting the non-lexical cues to speech output.

According to an embodiment, a transliteration work support apparatus include an input unit, an extraction unit, a presentation unit, a reception unit, and a correction unit. The input unit receives document information. The extraction unit extracts, as a correction part, a surface expression of the document information that matches a correction pattern expressing a plurality of surface expressions having the same regularity in way of correction in one form. The presentation unit presents a way of correction defined in accordance with the correction pattern used in the extraction of the correction part. The reception unit receives selection of the way of correction. The correction unit corrects the correction part based on the selected way of correction.

Systems and methods provide for text normalization or inverse text normalization using a hybrid language system that combines rule-based processing with neural or learned processing. For example, a hybrid rule-based and neural approach identifies semiotic tokens within a textual input and generates a set of potential plain-text conversions of the semiotic tokens. The plain-text conversions are weighted and evaluated by a trained language model that rescores the plain-text conversion based on context to identify a highest scoring plain-text conversion for further processing within a language system pipeline.

The techniques disclosed herein enable systems for spoken natural stylistic conversations with large language models. In contrast to many existing modalities for interacting with large language models that are limited to text, the techniques presented herein enable users to carry a fully spoken conversation with a large language model. This is accomplished by converting a user speech audio input to text and utilizing a prompt engine to analyze a sentiment expressed by the user. A large language model, having been trained on example conversations, by generating a text response as well as a style cue to express emotion in response to the sentiment expressed by speech audio input. A text-to-speech engine can subsequently interpret the text response and style cue to generate an audio output which emulates the sensation of human conversation.