Equipped with new functions that contribute to further efficiency in systems introduced by local assemblies and private companies
Advanced Media Corporation (Head office: Toshima-ku, Tokyo; CEO: Kiyoyuki Suzuki; hereinafter, "Advanced Media") has launched "AmiVoice®The "Minutes Taking Support System" will be released in October 2007 as a system that will contribute to dramatically improving the efficiency of taking minutes by adding features such as a speaker estimation function that identifies individuals from the characteristics of the speaker's voice and automatically inputs the speaker's name.
"AmiVoice® The "Meeting Minutes Creation Support System" was introduced in June 2004 by the Numazu City Council in Shizuoka Prefecture, and has since been adopted by local assemblies such as the Hokkaido Assembly, as well as many companies.AmiVoice®The system has been well-received as a powerful support for creating minutes, with features such as "unspecified speaker" support that does not require prior learning of the user's voice or speaking characteristics, and high recognition rates achieved by customizing to each user. Based on requests from existing users, new functions have been added to further streamline the creation of minutes. In particular, the speaker estimation function, which uses voiceprint authentication technology to identify the speaker from the speaker's voice characteristics, is the first such function to be installed in a minutes creation system. With conventional systems, the minutes creator had to take the time to determine the speaker from the recorded voice and enter it, but with this speaker estimation function, minutes can be created more quickly, contributing to a significant reduction in creation time and costs.
Main Product Features
[1: Speaker identification function]
This function uses our company's "voiceprint authentication technology*1" to identify individuals based on the characteristics of the speaker's voice. By incorporating a speaker estimation function, the speaker's name is automatically entered, making it possible to create minutes more efficiently.
[2: Enhanced conversion function based on the standard character dictionary]
The previous system also provided a dictionary of correct spelling, but this time we have strengthened the recognition dictionary and language model*2 in accordance with the "Standard Character Example Dictionary" that is referenced by the minutes writers. With this improvement, the recognition results are automatically written in accordance with the character example.
(Example: greeting ⇒ in the standard character usage dictionary it is written as "greeting")
[3: Acoustic model adaptation to the installation equipment] (Optional service)
If the acoustic characteristics of the facility (such as a conference room) cause a decrease in speech recognition accuracy, this service aims to improve recognition accuracy by adapting the acoustic model*3 to the environment of the facility. By customizing the acoustic model to suit the facility, higher recognition accuracy can be achieved.
Advanced Media has added this new function to its "AmiVoice®We will expand sales of the "Minutes Preparation Support System" as a service that strongly supports the preparation of minutes in local assemblies and private companies. We are also aiming for sales of 1 million yen in the next year. We will continue to develop voice recognition technology.AmiVoice® We will expand new corporate services that utilize voice recognition technology and open up a new future by widely providing the convenience of voice recognition technology.
*1) Voiceprint authentication technology
Voiceprint authentication technology is a biometric authentication technology developed by Advanced Media that uses "voiceprints," which are the characteristics of an individual's voice. Like fingerprints, voiceprints are unique to each individual, so it is possible to identify individuals based on them. Our company offers products that utilize voiceprint authentication technology, such as the "AmiVoice Guard" office entrance and exit management system, the "Open Door System" entrance opening and closing system for apartment complexes, and the "AmiVoice Mobile Verification" voiceprint authentication system via mobile phone. These systems achieve a stronger level of security by combining keyword checking with voice recognition technology.
*2) "Language model"
A language model is a grammar that describes the context in which a word is used. Here, we use statistical grammar that analyzes large amounts of text and digitizes the occurrence and connection probability of each word. The connection probability is calculated based on the frequency of occurrence of three consecutive words in a sentence. For example, if someone says "I am a doctor," we estimate that the word that comes after "I" and "is" is more likely to be "doctor" than "stone" or "will."
*3) "Acoustic model"
An acoustic model is a mechanism for estimating phonetic symbols from acoustic "features" obtained by analyzing input speech. Here, we use speaker-independent acoustic models that are statistically trained from large amounts of speech signals and labeled phonetic symbols.
