EE Times海外深度解析：破局音频“玄学”，炬芯端侧AI先行

当端侧AI从概念走向大规模商用，音频无疑是端侧设备最核心、最高频的交互入口，高保真音频处理的重要性愈发显著。炬芯科技凭借低功耗大算力端侧AI技术、25年积淀的音频全链路技术及先进的无线通信技术融合，正在被全球市场重新认识。

国际权威技术媒体EE Times近日刊发深度报道，从技术的角度详细分析了音频信号从采集到还原的完整链路，解读炬芯科技如何以硬核芯片、自研算法及“金耳朵”团队等技术验证体系，让好声音有据可依。文章亦进一步揭示了炬芯科技将自研模数混合SRAM存内计算技术率先落地于AI音频场景，破局端侧AI落地难题，同时将继续以声音感知为基向外延伸AI全域的底层逻辑。

 

目前，炬芯科技新一代存内计算端侧AI芯片累计出货量达数百万颗，已在无线麦克风、游戏耳机、派对音箱、直播声卡、桌面麦克风等各类主流消费电子音频终端实现规模落地；与此同时，面向覆盖AI穿戴、AI办公场景的AI眼镜、AI Note、桌面助理机器人等多模态端侧AI设备，炬芯科技正与多家品牌客户联合推进方案落地，相关终端产品正陆续推向市场。

正如EE Times的深度观察所呈现：经过海量量产与实测数据打磨的全链路音频处理能力，无法依靠短期投入或外购IP简单复制。相比于不可捉摸的音频”玄学”，炬芯科技以低功耗高能效存内计算技术持续拉高音频处理的性能上限，并且可量化、可量产的特质更加稀缺。

以下为EE Times原文分享

Proven: Actions Technology's End-to-End Audio Architecture Tames Sound’s "Black Magic"

Whether in the mass consumer market or the high-end professional audio segment, sound quality remains the defining factor that separates one product from another. Consumers’ appetite for better audio has never waned.

According to Qualcomm's 2023 State of Sound report, 73% of consumers prioritize improved sound quality when upgrading to a new device, while 69% consider lossless audio a key purchasing factor. A 2026 report from Futuresource further found that across the three major speaker use cases—home speaker, party speaker, and portable speaker—sound quality consistently ranks as the feature users value most. Terms such as 32-bit floating-point recording, studio-grade sound, and Hi-Res Audio certification have become some of the most prominent selling points in product marketing.

Meanwhile, the Bluetooth Special Interest Group (SIG) continues to advance technologies such as LE Audio, High Data Throughput (HDT), and the High-resolution Lossless Codec (HLC), all of which aim to make high-definition, lossless audio transmission over Bluetooth a reality.

The problem, however, is that neither Bluetooth protocol upgrades nor optimizations to individual signal-processing stages can solve everything. While these improvements can certainly enhance audio performance, each remains constrained by its own limitations and can easily be undermined by bottlenecks elsewhere in the signal chain. As a result, the sound that ultimately reaches the listener often falls short of expectations.

This challenge of "localized optimization" is particularly evident in audio SoC design. Different audio-processing modules are often developed by separate teams using different toolchains, or integrated from third-party IP vendors. Each module may perform exceptionally well in isolation, yet interface losses, asynchronous clock domains, wasted computing resources, and additional power consumption arising from module interaction are rarely optimized at the system level.

Treating the entire audio signal path as a unified engineering system is the only way to truly unlock great sound.

With 25 years of experience in audio technology and a long-standing focus on low-power AIoT chip design, Actions Technology has built a mature, end-to-end, high-fidelity audio platform by combining high-performance hardware with advanced algorithms. The company has integrated every stage of the audio pipeline—including signal acquisition, preprocessing, encoding and decoding, transmission, post-processing, and signal reconstruction—into a single, unified architecture, enabling audio solutions across a wide range of application scenarios.

 

Today, Actions Technology's products cover virtually every category of audio devices across three major ecosystems: personal audio, home audio, and PC peripherals.
 

From ADC to DAC: A Closer Look at Actions Technology's End-to-End Audio Architecture

Following the flow of the audio signal itself, this article examines the key technologies that make up Actions Technology's end-to-end audio solution.

• High-SNR ADC: Building a Clean Foundation for Digital Audio

The analog-to-digital converter (ADC) is the first stage in the audio signal chain. Its signal-to-noise ratio (SNR) and total harmonic distortion (THD) determine the fundamental characteristics of a signal once it enters the digital domain. The noise floor defines the upper limit of achievable SNR, while ultra-low distortion ensures that every nuance of the original signal is faithfully preserved.

Actions Technology's audio SoCs incorporate a fully in-house, high-performance ADC capable of delivering an SNR exceeding 120 dB and a total harmonic distortion plus noise (THD+N) below -100 dB.

These specifications rival the professional-grade performance of high-end discrete components from industry leaders such as TI, ADI, and AKM. Integrating the ADC directly into the SoC simplifies external circuitry and reduces potential sources of variation. The result is a stable and reliable source of low-noise, low-distortion, high-fidelity digital audio—laying a rock-solid foundation for the final acoustic experience.

 

• Low-Distortion ASRC: Near‑Lossless Cross‑Clock‑Domain Resampling

Asynchronous clocks between transmitting and receiving devices are a common engineering challenge in audio systems. In typical Bluetooth, USB, and I²S applications, clock discrepancies can lead directly to buffer overflows or data dropouts.

The key to solving this problem lies in the Asynchronous Sample Rate Converter (ASRC). The THD+N performance of an ASRC directly reflects the purity of the converted signal: the lower the THD+N, the more transparent the conversion.

Actions Technology's proprietary ASRC module is fully integrated into the SoC and maintains THD+N below -140 dB. This not only eliminates the routing complexity and clock-synchronization risks associated with external ASRC chips, but also provides highly transparent, virtually colorless sample-rate conversion.

For comparison, most conventional ASRC implementations in the industry typically achieve THD+N values between -120 dB and -130 dB. Reaching -140 dB pushes signal conversion significantly closer to the theoretical limit of lossless performance, placing it among the industry's best.

 

• High-Precision AI Noise Reduction: Filtering Out the Chaos to Restore the Original Audio

Environmental noise cancellation (ENC) has become one of the defining features of intelligent audio devices. Unlike traditional DSP solutions using lightweight models for basic filtering, or cloud-based schemes with high latency and power consumption, Actions Technology adopts a dedicated audio‑processing architecture. It combines an in‑house audio DSP and AI NPU built upon Mixed‑Mode SRAM‑based Computing‑In‑Memory (MMSCIM) technology, even under extremely tight power constraints, the architecture provides sufficient computing headroom to support stable execution of high-precision neural network models.

Meanwhile, Actions Technology's proprietary AI noise-reduction model underwent more than a year of continuous, around-the-clock training across diverse real-world environments.

Leveraging comprehensive software and hardware capabilities, Actions Technology breaks through the application limitations of traditional noise-reduction technologies. While industry-standard solutions with 16kHz/32kHz sampling rates are primarily tailored for voice calls, Actions Technology natively supports advanced 48kHz AI ENC full-band high-fidelity real-time audio processing, extending high-precision noise reduction capabilities to professional audio scenarios such as gaming voice chat, live streaming interaction, and audio/video creation — effectively suppressing complex background noise while preserving every nuance and texture of the original audio.

 

• 32-Bit Floating-Point Codec Technology: Preserving Every Detail of the Original Recording

Audio codecs are responsible for compressing, packaging, and reconstructing audio signals. The codec algorithm itself ultimately determines how much information survives this process. When codec precision is insufficient, dynamic range is compressed, causing both subtle details and powerful transients to suffer.

Compared with the industry's more common 16-bit and 24-bit solutions, Actions Technology has developed its own 32-bit high-dynamic-range floating-point codec, dramatically expanding the range of audio signals that can be accurately processed.

From the barely perceptible sound of falling snow to the explosive force of thunder, from a whisper in your ear to the full impact of a live rock concert, the technology is designed to capture and preserve the entire sonic landscape.

At the same time, the floating-point architecture provides ample headroom for post-processing. Whether boosting weak signals or attenuating overloaded waveforms, engineers can make substantial adjustments without introducing additional noise or sacrificing resolution.

From an efficiency standpoint, Actions Technology's codec also employs a resource-optimized design strategy, achieving high compression and reconstruction efficiency while minimizing computational overhead.

The computing resources saved can then be redirected toward extending battery life or enabling more sophisticated audio-processing algorithms.

 

• Independent Dual-RF Wireless Transmission: Building a Stable, High-Fidelity Audio Link

After passing through the ADC, ASRC, AI-powered noise reduction, and 32-bit encoding stages, the audio signal ultimately reaches the wireless transmission link before being delivered to the receiving device. At this point, interference, packet loss, and bandwidth limitations become the primary obstacles to preserving a high-definition listening experience.

To address these challenges, Actions Technology integrates a comprehensive set of anti-interference technologies, including Adaptive Frequency Hopping (AFH), Forward Error Correction (FEC), and Packet Loss Concealment (PLC). Working in concert, these technologies effectively mitigate common issues in the 2.4 GHz spectrum, such as signal interference and attenuation, ensuring stable transmission of high-quality audio streams.

In addition, Actions Technology's proprietary first-generation independent dual-RF architecture creates a dedicated physical-layer channel for high-bitrate audio. This design isolates the audio stream from auxiliary communications such as remote-control commands, status reporting, and firmware updates, preventing different types of data traffic from competing for bandwidth.

Under standard Bluetooth operation, the dual-RF architecture improves the stability of concurrent connections across multiple usage scenarios. When operating under the company's proprietary 2.4 GHz protocol, the architecture unlocks even greater bandwidth, supporting throughput of up to 4 Mbps—double the peak throughput of classic Bluetooth at 2 Mbps—while reliably carrying 48 kHz/32-bit high-bitrate audio streams.

The company's second-generation independent dual-RF architecture pushes bandwidth even further, reaching 12 Mbps while incorporating more advanced error-correction and coding technologies to deliver even stronger wireless performance.

Meanwhile, the Bluetooth Special Interest Group is introducing High Data Throughput (HDT) technology as part of the next-generation Bluetooth Low Energy core specification. HDT is expected to increase peak throughput from 2 Mbps to approximately 7.5 Mbps, making experiences such as lossless multichannel audio over Bluetooth possible for the first time.

Actions Technology's ATS296X series has already added support for Bluetooth HDT, making it one of the first chip solutions in the industry to support the emerging standard.

 

• Modular Post-Processing Algorithms: Tailoring the Listening Experience to Every Scenario

Audio post-processing algorithms represent the final stage in shaping how sound is ultimately perceived by the listener. From equalization (EQ) and dynamic range control (DRC) to virtual surround sound and spatial audio rendering, different applications place very different demands on audio processing.

While many manufacturers rely on third-party algorithm IP and simply fine-tune parameters or enable specific features, Actions Technology takes a different approach. Its proprietary post-processing algorithms can be selectively customized and deeply optimized for specific use cases, allowing them to be flexibly adapted to both low-power, battery-operated devices and professional-grade, high-fidelity equipment.

This gives manufacturers the freedom to fine-tune the listening experience according to their own product positioning.

More importantly, unlike traditional DSP implementations that offer only a handful of fixed processing modes, Actions Technology's post-processing framework can dynamically select the most appropriate processing modules and parameter weights in real time based on changing acoustic conditions.

The result is a hybrid model that combines a manufacturer's predefined sound signature with the device's own intelligent optimization. The listening experience retains the brand's distinctive sonic identity while simultaneously adapting to the surrounding environment, delivering a more personalized and premium audio experience.

 

• High-Fidelity DAC: Preserving Every Detail Through to the Final Output

The audio signal's journey begins with digitization in the ADC and ends with reconstruction in the DAC. Ultimately, digital signals must be converted back into acoustic waves that the human ear can perceive, and the performance of the digital-to-analog converter (DAC) determines whether the gains achieved throughout the entire signal chain can be fully realized.

The DAC integrated into Actions Technology's SoCs delivers an SNR up to 122 dB, an ultra-low noise floor below 1.6 μVrms, and THD+N as low as -100 dB.

Achieving these performance levels within a fully integrated SoC allows the solution to compete directly with the discrete components offered by leading analog manufacturers such as TI, ADI, and AKM—even in high-end professional audio applications.

The high-performance DAC faithfully reproduces even the most subtle details in an audio recording without adding unnecessary coloration, bringing the entire end-to-end audio processing pipeline to a fitting conclusion at the chip level.

Actions Technology's audio SoC architecture

It's also worth noting that Actions Technology has been actively pursuing forward-looking intellectual property strategies across key global markets, including China, the United States, and Europe. Core technologies such as its ADC and DAC designs, ASRC, AI noise reduction, audio and video codecs, post-processing algorithms, and independent dual-RF wireless architecture have already been granted patents in China, with protection continuing to expand internationally. This growing patent portfolio provides the company with a strong technological moat as it works with global brands.

Ultimately, the fundamental difference between Actions Technology and conventional audio solutions lies in its system-level, end-to-end approach.

The industry's conventional approach is relatively straightforward: source an ADC/DAC from one vendor, a noise-reduction algorithm from another, a codec library from a third, and then integrate them all into a single SoC.

Individually, none of these components may be lacking. The problem is that they were never designed to work together as an optimized whole. Actions Technology takes a fundamentally different path.

By leveraging a highly integrated, single-chip architecture, the company designs both hardware and algorithms collaboratively from the earliest stages of development. Every component—from signal acquisition to final output—is engineered not as an isolated module, but as part of a tightly coordinated system optimized for seamless interaction with the stages that come before and after it.

The result is an audio pipeline that delivers signals from input to output with minimal overhead and maximum efficiency.

The single-chip approach also eliminates many of the problems associated with traditional external codec designs, including PCB routing interference and elevated noise floors. Beyond reducing system costs for manufacturers, it transforms abstract technical specifications into something users can actually hear and appreciate.
That is the true value of Actions Technology's end-to-end audio architecture. Like a perfectly synchronized orchestra, every component performs its own role while working in harmony with the others—coming together to create a richer, more compelling listening experience.

Quantifiable Metrics, Audible Excellence
There is a common perception in the audio industry: sound quality is either defined by cold, technical specifications or by the vague and often indescribable “black magic” of subjective listening.

At Actions Technology, these two perspectives are not mutually exclusive. Objective measurements provide the framework; subjective listening gives it life. Neither can stand alone. In an increasingly competitive market, the company has combined electroacoustic engineering with decades of accumulated listening expertise to develop a distinctive acoustic signature that is difficult to replicate.

 

• Measurable, Objective Performance

From chip architecture and development to debugging and mass production, high-quality audio within Actions Technology's ecosystem is treated as a scientific discipline—one that is quantifiable, reproducible, and controllable.

Key metrics, including SNR, THD+N, noise floor, frequency response, and dynamic range, are rigorously measured using professional test equipment. Outstanding objective performance has become one of the key reasons why leading global brands continue to place their trust in the company's products.

 

• Refined, Subjective Listening

Actions Technology was one of the earliest chip companies in the industry to establish a professional listening evaluation system. Its in-house “Golden Ear” team undergoes professional hearing assessments, systematic training, and rigorous evaluation before receiving certification from an accredited listening panel.
This enables team members to accurately evaluate frequency balance, sonic detail, soundstage positioning, transparency, and other critical aspects of the listening experience.

 

In 2023, Actions Technology received official listening panel certification.

The company also operates a professional acoustics laboratory certified by the National Institute of Metrology, China. The facility includes an anechoic chamber, a reverberation chamber, and dedicated listening rooms used for speaker testing, smartwatch and headphone call-quality evaluation, audio effects research, wireless microphone and headphone noise-cancellation testing, and speaker listening studies.

Equipped with advanced objective and subjective audio evaluation systems, the laboratory provides a professional acoustic environment for precision tuning and product validation.

Actions Technology's Acoustic Laboratory

 

Exceptional objective performance and refined listening expertise are not developed overnight. Since 2001, the Actions team has accumulated 25 years of production experience across multiple generations of audio technologies and chip platforms, along with an extensive database of acoustic measurements gathered from real-world applications. These advantages cannot be replicated through short-term investment or by simply licensing third-party IP, giving the company a natural technological lead within the industry.

Market Validation: From a Preferred Technology to the Preferred Choice of Customers

End-to-end audio optimization can minimize signal degradation at the chip level, but the final wireless audio experience is still influenced by several external factors, including source quality, the physical performance of speakers and amplifiers, and the bandwidth limitations of wireless communication protocols. Together, these factors determine how audio ultimately sounds. Yet regardless of changing external variables, the underlying audio SoC remains one of the most important factors influencing sound quality.

Actions Technology's end-to-end audio platform has already been deeply integrated into the supply chains of many leading global brands. Its technology has been deployed at scale across a broad range of products, including Bluetooth speakers, party speakers, wireless microphones, desktop microphones, wireless gaming headsets, wireless home theater systems, AI Note, AI/AR glasses, and desktop robots.

From a product perspective, intelligent wireless audio SoCs remain the company's core business and have repeatedly demonstrated their value in the marketplace.
According to publicly available product teardowns, Actions Technology's SoCs have achieved significant penetration across the flagship products of the world's three largest consumer audio brands. Claiming the top global market share in two vertical markets — branded Bluetooth speakers and wireless microphones, it has become the undisputed leader for smart wireless audio SoCs.

The reason the company's solutions continue to win the support of leading global brands lies in its successful transformation into an edge AI chip company.

Built on its established foundation of high-fidelity, end-to-end audio technology, Actions Technology has chosen audio as the first large-scale commercialization scenario for edge AI. This strategy has enabled the company to move beyond traditional audio signal processing toward AI-enhanced intelligent audio.

Looking ahead, it plans to expand beyond audio into broader multimodal perception while continuing to strengthen edge AI computing and wireless connectivity to support a growing ecosystem of AIoT applications.

 

Actions Technology’s Edge AI Audio Capabilities

Traditional von Neumann architectures face the formidable power wall and performance wall, creating a fundamental challenge for AI-powered audio: balancing low power consumption with high computational performance.

While much of the industry was still debating the future direction of edge AI, Actions Technology had already begun investing in the technology. In 2024, the company introduced its first family of edge AI chips based on MMSCIM (Mixed-Mode SRAM-based Computing-In-Memory) technology.

Featuring a heterogeneous CPU + DSP + NPU architecture combined with proprietary computing-in-memory technology, these chips deliver 100 GOPS of AI performance while achieving energy efficiency of 6.4 TOPS/W.

The architecture enables battery-powered devices to perform demanding tasks such as AI noise reduction and voice separation locally while maintaining low power consumption. At the same time, the company's highly integrated single-chip design preserves the acoustic performance required for high-fidelity audio applications.
In 2025, with the commercial launch of flagship wireless microphones powered by its edge AI chips, Actions Technology became the first SoC company in the industry to bring MMSCIM technology to consumer audio products at scale.

As of August 2026, the company's Gen1 MMSCIM edge AI portfolio—including the ATS3231, ATS362X, and ATW6095 families—has reached cumulative shipments in the millions, spanning scenarios from mass-market entertainment to professional content creation.

 

• - The ATS3231 series has been widely adopted by multiple leading international brands for wireless microphones, wireless gaming headsets, and professional recording and live-streaming equipment.
• - The ATS362X series have been incorporated into AI party speakers from internationally recognized brands, as well as professional sound cards and digital mixing consoles that demand exceptionally high audio quality.
• - The ATW6095 series is currently supporting the development of AI glasses, AI Note, desktop robots, and other emerging products, many of which are now entering the market.

Driven by leading‑edge technological innovation and commercial deployment capabilities, Actions Technology has earned wide recognition from authoritative industry bodies in recent years. Its multiple prestigious honors include: the ATS3231 series for ASPENCORE’s 2025 China IC Design Awards in the SoC/ASIC category, the ATS362X series for ASPENCORE’s 2026 China IC Design Awards in the SoC/ASIC category, and, at the corporate level, the World Electronics Achievement Awards – Promising AI Company of the Year 2025, among other industry accolades.

Key Features of Actions Technology's First-Generation Edge AI Chips

Audio is the primary gateway to seamless, always-on AI interaction. Without edge AI audio capabilities, truly intelligent auditory experiences simply cannot exist.

By combining its end-to-end high-fidelity audio platform with a proven ability to commercialize edge AI technologies, Actions Technology has successfully navigated demanding customer qualification processes, complex global supply chains, and an increasingly competitive market.

Step by step, the company has evolved from being a preferred technology provider into becoming the preferred partner for customers around the world.