← Back to Technology

Technology topic / Audio

Adaptive audio purification

Clearer sound, built into every listening experience

An audio purification solution for audio apps, Bluetooth headphones and Bluetooth speakers. Multi-level noise suppression, low-frequency interference filtering and smooth switching between processing modes help reduce persistent noise in the source audio while striking the right balance between clarity and content detail.

Available as an online processing service, with local offline and embedded adaptation according to project needs.

Concept illustration: a luminous audio waveform passes from scattered granular noise through a translucent processing plane into a smooth, coherent wave. AI-generated image

Core features

Technical animation: noisy and processed waveforms side by side, with a live spectrum showing suppression of the target band
Noisy and processed waveforms compared; a live spectrum shows suppression of the target band.Animations use synthetic signals to illustrate the principle and do not represent measured results.

Adaptive noise suppression

Analyzes the spectral characteristics of audio to suppress persistent background noise. Processing strength adjusts to the material and the application, balancing purification with natural sound.

Three purification levels

Balanced mode suits light background hiss; Enhanced mode targets more noticeable persistent interference; Deep mode provides stronger suppression. You can switch back to the original audio at any time to choose by ear.

Targeted low-frequency hum filtering

Filters low-frequency hum from sources such as power-line interference at target frequencies, including 50 Hz and 100 Hz. Parameters can be configured to suit the material and product.

Continuous stream processing

Continuously receives, processes and outputs audio streams, with buffer management and status monitoring for uninterrupted playback in applications.

Smooth mode switching

Purification modes can be changed during playback. A short crossfade smooths the transition so listeners are not disturbed by abrupt changes.

Original vs. level-matched comparison

Compare the original audio, the processed audio and a level-matched version. Reducing loudness differences helps listeners focus on changes in noise and detail.

Saved results and task recovery

Supports file processing, result retrieval and recovery of saved tasks, with file integrity checks for easy integration into media processing and business workflows.

Flexible deployment

Concept illustration: a server, a laptop and an embedded circuit board linked by cyan audio light strands, representing flexible deployment. AI-generated image

Online serviceCentralized processing, flexible access

Processes audio files and continuous audio streams through an API. Suited to websites, mobile apps and business platforms, with centralized maintenance, unified configuration and room to scale.

Optimization focusCoordinating transfer and playback buffers, monitoring processing status and switching modes within a session.

Local offlineLess dependence on the network

Can be adapted for local computation according to project needs, purifying audio on the target device without uploading audio to a remote service or downloading results.

Optimization focusLocal file I/O and matching processing parameters to available compute, for restricted networks and on-device data retention.

Embedded deploymentAdapted to device resources

Can be integrated into smart audio devices and dedicated hardware, with processing configurations chosen for the processor architecture, memory and power budget.

Optimization focusBlock size, buffer length, memory use and compute cost, balancing quality, latency and resource use on the target hardware.

Bluetooth headphones

Source-side purification as part of wireless listening

Audio can be purified on the source, such as a phone or player, and then sent to headphones over Bluetooth. On-device integration can also be assessed based on chip capability and partner firmware.

Typical uses

  • Reducing persistent background noise in recordings, podcasts and other audio content.
  • Choosing purification strength per content and comparing with the original.
  • Adjusting processing modes during continuous playback.

Source-side processing can work without modifying headphone firmware, depending on how much control the application has over the playback chain. On-device deployment must be validated against the chip, firmware interfaces and resource budget. This is purification of playback content, not active noise cancellation; it does not claim to reduce ambient noise heard by the wearer.

Concept illustration: unbranded over-ear wireless headphones with a cyan audio ribbon flowing toward the ear cup, representing playback processing. AI-generated image
Technical animation: source decoding, audio purification, Bluetooth transmission and headphone playback, with moving packets, processing blocks and before/after waveforms
Source decoding → audio purification → Bluetooth transmission → headphone playback.Animations use synthetic signals to illustrate the principle and do not represent measured results.

Bluetooth speakers

Adjustable purification for wireless playback

Audio can be pre-processed on a phone, media device or gateway before it reaches a Bluetooth speaker. For products with suitable integration conditions, deployment inside the speaker's processing chain can also be assessed.

Typical uses

  • Reducing persistent noise in speech material and background audio.
  • Targeted filtering of specific low-frequency hum.
  • Choosing purification strength flexibly for different material.

For music, comparison with the original and level-matched listening help assess how well bass instruments, reverb and quiet details are preserved. Final results depend on the source, Bluetooth codec, speaker drivers and acoustic environment.

Concept illustration: an unbranded portable Bluetooth speaker beside a phone, with cyan sound waves radiating outward. AI-generated image
Technical animation: complementary weighting of modes A and B, crossfade mixing and continuous output, with a Bluetooth speaker playing sound
Modes A and B shift in complementary weights and are crossfaded into a continuous output.Animations use synthetic signals to illustrate the principle and do not represent measured results.

Tuning for different sound conditions

Concept illustration: cyan light ribbons flowing continuously through three translucent rings, representing continuous processing and smooth transitions. AI-generated image
Light background hiss
Start with balanced processing to reduce interference while keeping detail.
Persistent background interference
Increase processing strength and choose the right level by listening to the material.
Power-line hum
Filter the target frequencies while preserving useful low-frequency content.
Complex music
Use multiple levels and comparison with the original to balance purification against timbre, decay and quiet passages.
Transient noise, wind or overlapping voices
Assess with dedicated samples and confirm suitability for the project.

On silicon

DSP + GX8002: Embedded Real-Time Audio Processing

We ported the audio purification algorithm to the NationalChip GX8002 low-power voice chip. Sample-by-sample processing runs on the CK804 MCU's DSP instructions and floating-point unit, with 48 kHz mono PCM16 input over I²S and 10 ms frames.

The algorithm keeps state across frames, allocates no heap in its core, and needs only 108 bytes of state per stream. The full firmware cross-compiles, links and packages successfully.

Eight-stage processing chain

  1. Input conditioning

    Impulse suppression and prediction.

  2. Filtering and detection

    Pre-emphasis, DC blocking and envelope detection.

  3. Noise tracking

    Minimum tracking with adaptive smoothing.

  4. Gated attenuation

    Adjusted by scene and noise level.

  5. Dynamic compression

    Soft knee and level control.

  6. Clarity enhancement

    Mixing of enhancement components.

  7. Continuity

    Activity smoothing and transient handling.

  8. Output protection

    Amplitude and slew-rate limiting.

Chip and engineering highlights

Processing core
32-bit CK804 MCU up to 50 MHz with DSP instructions and FPU; the on-chip NPU is not used in this design.
On-chip memory
208 KB SRAM and SPI NOR with XIP; SIP Flash size depends on the variant.
Audio path
48 kHz PCM16 over I²S, processed serially in the SDK audio task with ring buffers and a pass-through bypass.
Scene presets
Five presets—call, voice message, music, video and listening enhancement—with noise, clarity and gain controls from 0 to 100.
Footprint
108 B of state per stream; about 4.8 KB of core and adapter code; 50,300 B for the complete NOR firmware including the SDK.

Consistency checks

18 setsComparison tests across 5 scenes, extreme settings and edge-case signals

432,000PCM samples compared

≤ 1 LSBMaximum sample difference from the reference implementation, with RMSE no higher than 0.025 LSB

Cross-compilation and host-side consistency checks are complete. Board audio output, per-frame time, end-to-end latency, power and real-world noise reduction still need on-board validation. Consistency results show the implementation matches the reference algorithm; they do not measure noise reduction or SNR gain. “Listening enhancement” is a scene name, not a medical function. Chip specifications follow the manufacturer's documents for the exact variant.

From requirements to integration

  1. Scenario assessment

    Identify noise types, source characteristics and target devices.

  2. Sample validation

    Provide before/after and level-matched listening to choose the right mode.

  3. Platform adaptation

    Decide where processing runs—online, local or embedded—and define interfaces.

  4. Integration and acceptance

    Verify audio quality, latency, resource use and stability in the target environment.

FAQ

Can it work without a network connection?

Offline deployment can be adapted according to project needs. Available features and performance of an offline version are determined by delivery and acceptance on the target platform.

Do I need to replace my Bluetooth headphones or speakers?

With source-side processing where the playback chain can be controlled, the receiving device usually needs no changes. Running inside the device requires an assessment of the chip, firmware and licensing conditions.

Is stronger processing always better?

Choose according to the material. Stronger processing can affect high-frequency detail, reverb and low-frequency content; we recommend confirming parameters with level-matched comparison.

Is this the same as active noise cancellation?

This technology purifies the audio content itself. Headphone active noise cancellation, microphone call noise reduction and echo cancellation are different capabilities that require separate assessment.

Contact and partnership

Find the right purification approach for your audio product

Share audio samples, your target platform and integration requirements, and we will assess processing results and deployment options together.