Speech recognition - Voicetune Info.Tech.Co. Ltd

communicating with robot through words they can understand is always a dream to people. Speech recognition technology is a high-tech that is capable to transfer the voice signal into correspondent text or command through the machine’s recognition and understanding. Speech recognition is a crossing discipline with broad conceptions, including acoustics, phonetics, linguistics, information theory, pattern recognition and neural biology and others. Speech recognition technology is gradually becoming the key technology of computer information processing technology, and the application of speech technology has become an emerging competitive high-tech industry. For example, the recent application of voice recognition on the mobile terminal is the most popular, with increasing voice conversations robot, voice assistant, interactive tools emerging, so many Internet companies have invested in human, material and financial resources to expand research and application in this regard, which purpose is to quickly occupy the customer by innovative and convenient mode of the voice interactivities.

The technology is divided into specific recognition and non-specific recognition. In the non-specific recognition, it also includes small vocabularies and large vocabularies recognition. Voicetune improved speech recognition rate by improving the algorithm structure, with the advantages of small resources, fast speed, high identification rate, etc., which adapts to all types of application development in middle and low class chip. It is also widely applied in the field of voice toys, small appliances, PDA voice input and inquiry, voice name dialing phones of cellphone and fixed telephones, vehicle-mounted voice control system and voice data query.

	multi-sensor intelligent interactive robot 1) Implement the relevant action based on voice commands, such as walking, rolling, dancing, etc. 2) Speaker supported, which can play different content based on customer requirements 3) LED dot matrix screen, which can display different expressions 4) Infrared remote control and barrier function 5) Support touching function
	Balancing robot 1) Self-regulated center of gravity to maintain balance. 2) Switch mode by rotating the wheel, and companying with light color change. 3) Dance Mode: dance with the rhythm of the music, which contains music pieces with different rhythm modes and cool actions. 4) Hand gestures response: respond to hands gestures, drawback with hand closing. Spin movements with gesture left / right. 5) Weight-bearing function: maximum care raised the same weight with their objects. 6) Speech recognition: the voice commands to control the robot to do flexible forward, backward, left, right, etc., reach any place you want with your orders. 7) Custom Mode: customize the actions through gestures or speech, memorize the custom actions, and edit the action sequence. 8) Free walk mode: walk freely, can automatically avoid obstacles during walking, with the left and right, front and rear, circles and other activities. 9) Bright Eyes light, reflecting a variety of robot’s emotions. 10) Remote Controller (2.4G module)
	Toy Car 1) Support MP3 port for MP3 music playback 2) USB port supports for music in the driver 3) TF card port supports for music in TF card 4) Support 2.4G Module 5) Support for voice remote control
	Hand raising dog 1) Capable to a series of actions, including walking, sitting down, nodding, raising his hand 2) Support voice command functions 3) Support touch function 4) According to customer requirements, do different voice commands
	Hypal 1) The magic Hypal, can understand voice commands, and communicate with children 2) Support for touch technology 3) Built-brain teasers and other fun games
	Bear Miffy learning machine 1) Intelligent Voice Dialogue 2) Remote on-demand feature 3) Position game 4) Guess mind game 5) Record a message 6) Sleep timer 7) USB downloads 8) Containing massive early childhood content 9) Support 4G 8G TF card expansions, etc.

1. Introduce the innovated noise template technology, which effectively removes noise leading voice input, the speaker's tone tie mouth, tail and tail noise through the grammatical structure of a noise template.

2. Adopt the independent anti-grammar network rejects technology that fully considers of the lift-right structure features in the language model, and bring a classification method of anti-grammar network model. With this reject method, it can connect with almost all speech recognition engine to realize all kinds of recognition tasks from small vocabularies to large vocabularies in user related or user unrelated.

3. Adopt the secondary HMM technology, removal the mute in the first HMM, which can also compress the length of the input speech, and complete recognition in the second HMM.

4. High recognition rate, low cost, low power consumption:

a) Background noise: 60db

b) Dynamic memory (RAM): 2KB (Training mode) / 6KB (adaptive mode)

c) Read-only memory (ROM): 2KB / 10KB (related to the hardware platform)

d) Rejection rate: 5%

e) Recognition rate: 96% of the fixed model, 98% of self-learning model

5. Support all kinds of mainstream embedded platform to support a variety of mainstream processors, support for multiple languages recognition

"Voicetune noise recognition software" won the fund of Shanghai National Innovation Fund support in 2006, the conclusion from Shanghai Novelty Consulting Center of Chinese Academy of Sciences: The technology has generally reached the leading domestic and international advanced level.