The Journal of China Voice Research recently published an article "Reviewing Endangered Language with Language Copy Methods — Case of Xibabi"。This article proposes a new method of recording endangered language with artificial intelligence technology，From this we see the in -depth combination of large -scale corpus and AI technology、Language resource protection and the development of AI technology will have an important promotion role。

　bet365 Play online games

2017，American scientists proposed the Speaking Rosetta Plan，It is designed to record unknown language without text in parallel relationships in the text of the text of the unknown language (usually endangered language)。

Zu Yiqing and others in the "Intelligent Voice Technology and its industrialization of intelligent voice technology and its intelligent voice technology and its system" project of the Ministry of Industry and Information Technology in 2015，Carry out the voice synthesis of Xibabi。This work has given them a concept of language replication of endangered language: use bet365 Play online games voice synthesis、Skills such as voice recognition and machine translation, copy the language of endangered language。Multi -language、The experience of the voice recognition system of multiple pronunciation people in terms of language classification and language common research，You can use it to use it in the field of endangered language treatment，and help developers realize a language replication of an endangered language faster。

Copy language proposed by this idea，is a record of the basic appearance of a language。Conventional recording data can improve the language and sound effect at most，Once there is a language replication system that has this endangered language，You can enter any text or voice of mainstream language or reference language，Voice content corresponding to the endangered language by converting output。System framework based on mainstream language or reference language，Completion of language replication requires establishing a voice synthesis system for target language、The translation system between the voice recognition system and the language and the mainstream language。The voice output of the language replication system is not a natural voice，Instead of voice synthesized by voice synthesis technology。Bet365 lotto review When an endangered language is really lost，People can still interact through the voice replication system and this language。

　　Voice technology expands the space for endangered language research

The language replication system framework with text language is roughly as follows: Assume that the mainstream language or reference language is Chinese Mandarin，The target language is an endangered language，Enter any Chinese text，The system can output the voice of this endangered language。Enter the text of any endangered language，The system can also output Chinese voice。

The target language voice synthesis system is the basis of the language replication system。The voice synthesis system in language protection、The meaning of language research is far greater than practical meaning。The traditional voice synthesis method needs to be analyzed by text，Convert the text to the idiom voice unit，Then convert the voice unit sequence into a sound through a synthesizer。In the section of acoustic modeling，Need to define basic voice units (for example, phonetic、Sound Mother's Mother's Mother)，At the same time, you need to clarify the rhythm characteristics of these voice units in continuous discourse，Is it re -read、Location of the rhythm structure where the rhythmic Bet365 lotto review structure is。In addition，The basic voice unit also carries the function of the sentence、Active function、Emotional performance and other higher levels of linguistics characteristics。If these linguistics characteristics are marked correctly，Trained acoustic models carry richer linguistic content。When generating synthetic voice，can produce richer expression。

At the same time，The output effect of the voice synthesis system can also check whether the input of linguistic knowledge is correct。For mainstream language，For example, Chinese Mandarin，The definition of the basic voice structure has been very clear，It can reach the level of automation in terms of audio segments，including chapter、Dialogue、Emotions and other linguistics characteristics also have room for research。For endangered language，The basic voice structure has not been revealed clearly，Using voice synthesis technology can get a complete analysis。For example, the basic phonetic definition of Sibber is a large number of sound changes in continuous discourse，In the process of data labeling, you can relatively complete the law of sound change，And isolated words cannot show changes in these sounds。If you only rely on manual analysis，In order to eliminate the impact of other phonetum，Bet365 lotto review Usually only an isolated word can be used for analysis。During the data processing process of voice synthesis，Researchers have the opportunity to analyze the segment of each fragment of the continuous discourse，At the same time, analysis of other linguistic levels such as continuous discourse such as rhythm，The linguistic knowledge is conveyed to the voice synthesis system through data label，and the correctness of the output inspection of the output synthesis。In this research mode，You can promote linguistic research。

The discussion of the previous discussion is limited to the endangered language processing with words。The technical problems involved in endangered language records without text are more complicated，Difficulty is also greater，But voice technology can develop more space for language research without text。

　　Language resources protection and AI development complement each other

Chinese and ethnic minority language scholars，You can use this voice synthesis system to carry out your own research。We think，In addition to endangered language records，Linguist can collaborate with artificial intelligence engineers，The first two aspects are acting on the first two aspects: voiceologists and linguists use existing knowledge to finely label bet365 best casino games the data，Label content includes voice structure、Sentence structure，until the chapter information structure and supersonic segment; will be labeled，Use the intelligent voice synthesis system as the research platform，By synthetic verification, it carefully examines whether the input linguistics knowledge is correct。Research method that combines large -scale natural corpus with artificial intelligence AI，Its results can study the basic voice structure，You can also study the focus of the statement more in depth、Linguistics of rhythm and language，Of course，can also further improve the naturalness of the synthetic voice。

As the intelligent language technology has arrived，Linguist and voiceologists should actively act，Do a good job in the construction of voice and language data resources，Provide solid data support for the development of my country's AI industry。

(Author Unit: Institute of Language Research Institute of Chinese Academy of Social Sciences)