Advancing the Speech Recognition Platform through DNNs, Multilingual Recognition and Conversational AI
First written in FY2017. 18 of 23 entries say "Advanced", first in FY2018-Q3. Latest entry: 2026-06-29.
1 other company's latest entry also says "Advanced".
- 2026-06-29AdvancedManagement planning
「Basic development of speech recognition has been completed, and we are already developing and commercializing a variety of products.」
After stating that basic speech recognition development had been completed, the company identified advancing voice AI technology and developing new products as its next development priorities. - FY2026AdvancedAnnual report
「For AI speech recognition AmiVoice®, the company advanced research into and implementation of the latest deep-learning technologies and improved the accuracy, expanded the scope of application, and added and enhanced functions of its End-to-End speech-recognition engine.」
The company advanced its speech-recognition platform toward the End-to-End model and moved peripheral technologies, including local LLMs, multi-stage AI reasoning and generative-AI summarization, into the implementation phase. - FY2025AdvancedAnnual report
「The AI speech-recognition API service AmiVoice® API was equipped with a next-generation speech-recognition engine using the End-to-End approach, achieving higher speech-recognition accuracy than the conventional hybrid engine.」
The company equipped AmiVoice API with an End-to-End next-generation speech-recognition engine, achieving higher recognition accuracy than the conventional model. - FY2025-H1AdvancedAnnual report
「The company released AOI LLM for AmiVoice® Communication Suite, enabling secure use of large language models (LLMs) in a local environment.」
To enhance its speech-recognition platform, the company released a product enabling secure use of LLMs in a local environment and advanced the incorporation of generative-AI technology into actual services. - FY2024AdvancedAnnual report
「During the consolidated fiscal year, the company worked to further improve recognition accuracy for AI speech recognition AmiVoice® and conduct research and development into peripheral and related technologies for AI speech recognition.」
The company expanded its research and development scope to include End-to-End speech recognition, Transformer acoustic and language models, generative-AI-based dialogue, key-sentence extraction and automatic summarization, and speaker identification. - FY2024-Q2AdvancedAnnual report
「The company released AI Summary (β) powered by GPT-3.5, which enables automatic summarization using Microsoft Azure OpenAI Service in the AI speech-recognition transcription support application AmiVoice® ScribeAssist.」
To enhance its speech-recognition platform, the company released an automatic summarization function using GPT-3.5 and moved generative AI into actual services. - FY2024-Q1AdvancedAnnual report
「Development also focused on functions that enhance the value of products and services, including the development of a summarization system using OpenAI's GPT-3.5/4 large language models and the start of a pilot program in Toride City.」
To enhance its speech-recognition platform, the company moved to developing and trialing a summarization function using OpenAI's large language model. - FY2023-Q1AdvancedAnnual report
「The company worked to develop functions that would enhance the added value of its products and services, including the trial introduction at Toride Municipal Assembly of a parliamentary minutes visualization system using AI speech recognition and natural-language processing, which was the first such implementation by a local government (according to the company). It also began offering a new emotion-analysis engine.」
The company expanded the application areas of its speech-recognition platform through the trial introduction of a parliamentary minutes visualization system using natural-language processing and the launch of an emotion-analysis engine. - FY2022AdvancedAnnual report
「For AI speech recognition AmiVoice®, the company advanced research into and implementation of the latest deep-learning technologies, including the End-to-End approach, which is attracting attention as a next-generation speech-recognition technology. It also improved the training method for Bi-LSTM (Bidirectional Long Short-Term Memory), a bidirectional deep-learning technology, thereby improving natural-speech recognition accuracy.」
In addition to improving the Bi-LSTM training method, the company expanded its efforts to research and implement the latest deep-learning technologies, including the End-to-End approach. - FY2022-Q3AdvancedAnnual report
「The new version improved the immediate visualization of call trends through an AI topic-extraction function and enhanced communication functions to support work-from-home operators. It also improved recognition accuracy by incorporating a Bi-LSTM engine using the latest deep-learning technology.」
To enhance its speech-recognition platform, the company incorporated a Bi-LSTM engine into Communication Suite4, achieving improved recognition accuracy. - FY2022-Q2AdvancedAnnual report
「The company expanded the business-oriented speech-recognition engines of its flagship product AmiVoice® ScribeAssist by adding medical, parliamentary and English speech-recognition engines. This enabled users to select the optimal engine for each meeting and use the product in English-language meetings.」
For its speech-recognition platform, the company added medical, parliamentary and English engines to ScribeAssist, enabling selection by use case and support for English-language meetings. - FY2021AdvancedAnnual report
「For AI speech recognition AmiVoice®, the company advanced research into and implementation of the latest deep-learning technologies and Bi-LSTM (Bidirectional Long Short-Term Memory), a bidirectional deep-learning technology, to improve natural-speech recognition accuracy. It also expanded the application of Bi-LSTM to Chinese, Korean, English and Thai and improved speech-recognition accuracy.」
The company advanced implementation of Bi-LSTM, enhancing natural-speech recognition and expanding its scope of application to Chinese, Korean, English and Thai. - FY2021-Q2AdvancedAnnual report
「In addition to the general-purpose ACP already released, the company released 5 domain-specific engines: medical ACP, business-reporting ACP for finance, business-reporting ACP for insurance, business-reporting ACP for pharmaceuticals, and meeting and minutes ACP. These engines incorporate Bi-LSTM (Bidirectional Long Short-Term Memory), a bidirectional deep-learning technology, producing a maximum error-improvement rate of 25% (according to the company) and substantially improving recognition accuracy, particularly for natural speech.」
The speech-recognition platform released 5 domain-specific engines and achieved a maximum error-improvement rate of 25% through Bi-LSTM. - FY2021-Q1AdvancedAnnual report
「AmiVoice® Cloud Platform (ACP), a development platform providing speech-recognition APIs, began offering APIs for 3 domain-specific engines, in addition to the previously available general-purpose engine: a medical general-purpose engine, a business-reporting pharmaceutical engine, and a business-reporting financial engine.」
The speech-recognition platform began offering APIs for 3 domain-specific engines covering medical, pharmaceutical and financial applications, in addition to the general-purpose engine. - FY2020AdvancedAnnual report
「For AI speech recognition, the company advanced research and implementation of Bi-LSTM (Bidirectional Long Short-Term Memory), a bidirectional deep-learning technology, to improve natural-speech recognition accuracy. It also improved speech-recognition accuracy in Chinese, Korean, English and Thai and expanded the scope of application.」
The company advanced LSTM research and implementation to Bi-LSTM, thereby enhancing natural-speech recognition and expanding the scope of application to Chinese, Korean, English and Thai. - FY2020-Q3ReframedAnnual report
「The Overseas Business Division worked to further improve the recognition accuracy of the Chinese-language AI speech recognition AmiVoice® engine and develop business in new fields beyond call centers.」
To enhance its speech-recognition platform, the company specified plans to improve the accuracy of its Chinese engine and develop applications in new fields beyond call centers. - FY2020-Q2RestatedAnnual report
「The Overseas Business Division focused on research and development, including further improving the recognition accuracy of the Chinese-language AI speech recognition AmiVoice® engine and developing call center products for China.」
The company continues to enhance its Chinese speech-recognition engine and develop products for China as measures to strengthen its speech-recognition platform. - FY2020-Q1RestatedAnnual report
「The Overseas Business Division focused on research and development, including further improving the recognition accuracy of the Chinese-language AI speech recognition AmiVoice® engine and developing call center products for China.」
The company continues to improve recognition accuracy for its Chinese AI speech-recognition engine and develop call-center products for China as part of its efforts to enhance its speech-recognition platform. - FY2019AdvancedAnnual report
「The company advanced research and implementation of recurrent neural networks, an advanced deep-learning technology, to improve natural-speech recognition accuracy. It also improved speech-recognition accuracy for Chinese, Korean, English and Thai and expanded the scope of application.」
The company advanced RNN research and implementation to improve natural-speech recognition accuracy, expanded support and the scope of application for Chinese, Korean, English and Thai, and implemented emotion analysis and LSTM in call-center products. - FY2019-Q3AdvancedAnnual report
「The company equipped AmiVoice® Communication Suite, an AI speech-recognition solution for contact centers, with emotion-analysis functionality and LSTM technology, a recurrent neural network and advanced deep-learning technology, achieving visualization of emotions and a substantial improvement in recognition accuracy.」
The implementation of LSTM technology progressed from research and development and phased implementation to deployment in call-center products and improved recognition accuracy. - FY2018RestatedAnnual report
「In basic technologies, the company worked to improve recognition accuracy by implementing recurrent neural networks, an advanced deep-learning technology, and to improve multilingual speech-recognition accuracy.」
The company continues to focus its research and development on improving natural-speech recognition accuracy through RNN research and implementation, and advancing multilingual recognition including Chinese, Korean and English. - FY2018-Q3AdvancedAnnual report
「The company implemented LSTM, a recurrent neural network and an advanced deep-learning technology, in the AmiVoice® speech-recognition engine. This produced a maximum error-improvement rate of 20% (according to the company) and substantially improved recognition of natural speech. The technology will initially be applied in the minutes and call-center fields and then implemented sequentially in products and services in each field.」
The company implemented LSTM in its speech-recognition platform and recorded a maximum error-improvement rate of 20%. It stated that implementation would proceed sequentially across products and services, beginning with the minutes and call-center fields. - FY2017First statedAnnual report
「In basic technologies, the company worked to improve recognition accuracy for natural speech using DNNs (deep neural networks) and to improve multilingual speech-recognition accuracy. In applied technologies, it worked on developing an intent-extraction engine for conversational agents and interactive information input/output applications.」
The company identified natural-speech recognition using DNNs, multilingual recognition, intent-extraction engines and interactive applications as key research and development areas.