How to customize your voice with speech synthesis technology?

Nexdata
4 min readJun 18, 2021

At the 2020 Xiaomi Developer Conference (MIDC), Xiaomi announced to launch the smart voice assistant Xiaoai 5.0. Xiaoai 5.0 has made a lot of innovations in sound experience, such as cute children’s voice, multi-emotional voice, Cantonese synthesis, customized voice, etc.

Behind the upgrade of Xiaoai is the continuous innovation of Xiaomi’s speech synthesis technology.

What is speech synthesis?

Speech synthesis, also known as TTS(Text to Speech), is a technology that artificially generates human speech and converts arbitrary text information into standard and smooth speech read aloud in real time.

TTS involves multiple disciplines such as acoustics, linguistics, digital signal processing, and computer science. It is a cutting-edge technology in the field of information processing. The main problem to be solved is how to convert text information into audible sound information, that is, to make the machine talk like human.

Speech synthesis is very popular in recent years. Well-known AI companies such as iFLYTEK, AISpeech, Google, and Huawei have made their efforts in the field of speech synthesis and developed voice assistants, smart speakers, voice translation and other applications, which have penetrated into our daily life.

Although TTS technology has made considerable development, there is still a lot of room for improvement.

At present, the naturalness and intelligibility of TTS is basically satisfied, but the naturalness of sentences and texts is still a big problem. Also, human speech has different emotions, tones and speeds, and styles, its richness is a big challenge which TTS needs make further efforts.

As professional AI data service provider, Datatang is committed to overcoming technical bottlenecks and promoting TTS technology for wider application. Datatang launched TTS data solutions.

Based on massive voice and text data annotation experience and leading TTS technology, Datatang supports the rapid synthesis of customized voices for different scenarios, timbres, sound quality, types and other requirements, so that machines can speak as well as human.

Datatang’s TTS data solution:

Datatang has abundant data resources, outstanding technical advantages and rich data processing experience, and supports customized voice data collection according to scenes, languages, ages, genders, and speakers.

Security compliance

In order to ensure safe and compliant data services, Datatang has established a security compliance system for company business in accordance with the data laws and policies of major countries around the world.

In compliance to the security compliance system, all the data collection is authorized by the person to be collected.

Professional recording environment

Datatang has a professional audio recording studio equipped with professional vocal condenser microphones and monitoring equipment. The recording studio complies with NR15 acoustic standards: the reverberation time is less than 0.1 second, the background noise is less than 20dB. The studio has been certified by the Building Physics Laboratory of Tsinghua University.

Abundant resources

Datatang has a resource of thousands of professional speakers around the world and professional team of hundreds of people.

Datatang supports speech synthesis in multiple languages, such as Mandarin, Chinese dialects and English, mixed Chinese-English, etc. Besides, Datatang has a variety of voice resources such as male voice, female voice, and child voice. Each timbre has different types of speakers, fully meeting the needs of diverse speech synthesis tasks.

Quality assurance

In the recording process, Datatang has professional monitors to ensure the recording quality. Through consulting experts, research papers, words pronunciation dictionaries, Google translation and Baidu translation, Datatang compiled a complete set of pronunciation rules and produced a pronunciation dictionary.

Datatang’s TTS data solution application scenarios:

Datatang provides TTS data solutions for various application scenarios, such as customer service, audio books, in-car voice interaction, music synthesis, etc.

Customer service

Datatang has a rich speech synthesis sound library, which can simulate the real status of the speaker and help create a conversational customer service, so as to improve customer experience and realize the transformation of marketing effects.

Audio books

Datatang’s speech synthesis data solution supports reading scenarios such as novels, news, books, etc., which helps people liberate their eyes, guarantees the smoothness and clarity of content, and lower the level of audio content creation.

In-car voice interaction

In-car interactive systems such as voice navigation, voice control, and in-car entertainment systems, which can create a convenient and entertaining driving experience while freeing the hands of driver.

Music synthesis

The music synthesis system learns from the data, provides intuitive control of the changes in timbre and music intensity, and can create music that cannot be achieved by artificial methods.

Datatang records music in accordance with TTS standards, including sheet music production, phonetic character labeling, intonation proofreading, etc.

End

The current speech synthesis technology has been applied to various scenarios, meeting most of the needs in the market, and is a relatively mature product. The main problem lies in the specific needs of different scenes, such as different digit readings, how to intelligently judge which broadcast method the current scene should be, and what tone and mood are more suitable for the current scene.

Datatang has been deeply involved in the field of AI data services for 10 years, always maintaining a sense of innovation, actively exploring new fields and new applications, in order to improve TTS data solutions. We are committed to transforming more research results into practical applications.

If you need data services, please feel free to contact us:info@datatang.com

--

--

Nexdata

Off-the-shelf AI training data, on-demand data collection & annotation services