Chipintelli, the third-generation speech AI chip developed by Chipintelli, has achieved a record-breaking milestone with over 10 million units shipped in the first half of 2023.
The CEO, Yunpeng He, highlighted the company's success in developing the world's first deep neural network processor (BNPU) for offline speech processing, which offers high-performance and cost-effective speech interaction chips and solutions compared to traditional DSP or MCU architectures. With the emergence of ChatGPT, people have been impressed by the capabilities of large language models and the charm of artificial intelligence. As the gateway to human-machine interaction in the AI era, intelligent speech recognition has gained wider recognition. The BNPU architecture complements speech AI applications, and the company's independently developed on-device NLP algorithms and deep learning technologies enable truly intelligent speech interactions and provide users with a more natural and personalized communication experience. This is why our third-generation speech AI chip has gained recognition in the market.
BNPU Architecture Chip: A high-performance speech chip born for AI
Just like general-purpose computing chips such as CPUs and GPUs, each era has its own computing architecture, and the same goes for speech recognition chips. The use of DSP and MCU architectures in speech chips can no longer meet the computational power, processing speed, and intelligent algorithm requirements of the intelligent speech era. Chipintelli's third-generation BNPU architecture chip can support larger language models, deep learning algorithms, and on-device NLP techniques. Its parallel computing architecture greatly improves inference and computation efficiency, resulting in efficient speech processing capabilities, real-time performance, low latency, and improved speech recognition accuracy, thereby significantly enhancing the user's speech interaction experience.
Each era has its own computing architecture
"Reliable": High-quality speech chips that meet industrial standards
All Chipintelli chips undergo reliability and environmental standard testing. The chips operate within a temperature range of -40°C to +85°C, meet MSL3 moisture sensitivity level requirements, comply with the IEC 61000-4-2 4K contact discharge test standard, adhere to FCC electromagnetic compatibility standards, and comply with ROHS and REACH environmental standards. They fully meet the safety, stability, and environmental requirements of various household appliances. In addition, all Chipintelli chips are independently developed by Chipintelli and, as a direct customer of SMIC (Semiconductor Manufacturing International Corporation), can ensure maximum product and production capacity stability, avoiding issues such as supply shortages and delivery delays.
Algorithm-driven: "Intelligent algorithms" for better noise resistance and higher recognition rates
Chipintelli has pioneered four core technologies, including dual-microphone deep speech separation, on-device NLP, single-microphone deep learning noise reduction, and dialect self-learning. Among them, on-device NLP technology enables context association, multi-turn dialogues, and intent understanding. Deep learning noise reduction technology allows products to achieve over 90% recognition rates in noisy environments with 70±5dB of noise and human speech. "Scenario optimization" brings a more reliable speech interaction experience. Chipintelli has conducted extensive training for home scenarios, such as TV sounds, fan noise, water flow sounds, group sounds, and ambient sounds, improving the accuracy of speech recognition. In actual home environments, it can achieve far-field speech recognition up to 10 meters with an accuracy rate of over 97%. Furthermore, Chipintelli is equipped with a professional algorithm optimization team that can provide specialized training based on specific customer command words, delivering satisfactory language models to customers. With support for over 500 offline command words and on-device NLP technology supporting tens of thousands of phrases, Chipintelli provides customers with rich and flexible speech interaction options. Its strong word generalization ability allows multiple command words to be assigned to the same function, freeing users from being restricted to specific command words and greatly enhancing the user experience.
Full-stack Service: As a leader in the intelligent speech industry, we are committed to continuously innovating chips and algorithms. In order to provide a more user-friendly development environment, we have invested significant resources in building an all-in-one speech AI platform, offering hundreds of speech solutions to our customers. To quickly respond to customer needs, we also have an efficient and professional FAE (Field Application Engineer) team. By implementing an integrated "Chip-Algorithm-Platform-Solution-FAE" strategy, we aim to maximize the potential of our chips, reduce friction costs, and ensure that customers obtain the maximum value.
