Detailed information BeClear SHF
Market Focus and Applicability
BeClear SuperHandsFree (SHF) provides high quality, clear speech for the use cases of speech recognition and communication at larger distances under reverberant and noisy conditions. BeClear SHF deals with both diffuse and point-noise sources.
The figure below schematically illustrates the BeClear SHF algorithm. By means of a fast tracking, free-running beam (the beam in orange), the SHF algorithm generates one or more focused beams (the beams in blue) capturing the coherent sources. The illustration is schematically only, because through the BeClear building blocks, BeClear does not make beams in the traditional sense.
- A fast-tracking free running beam continuously adapts to the source with the highest coherent energy component
- The BeClear SHF algorithm allocates focused beams to coherent sources, for example to a desired speaker or a point noise source such as a radio or TV
- When a focused beam has properly converged to a source, signal content outside the focused beam will be suppressed, thereby increasing the SNR in each of the focused beams
- Each focused beam is able to autonomously track slow (e.g. head) movements of a coherent source
- Focused beams are automatically re-allocated to new sources based on internal parameters
- Incoherent or diffuse noise sources such as an air conditioner or nearby traffic will be suppressed
This makes BeClear SHF ideal for voice control of smart TVs and smart speakers that are located in noise living rooms with multiple point noise sources.
By design, the BeClear SHF algorithm will automatically adapt to the number of microphones and the geometry they are placed in. Thus, BeClear SHF provides maximum freedom in designing the form factor of your device BeClear SHF can however also be combined with BeClear Multichannel AEC for more than 2 AEC-reference signals.
As illustrated in the below figure, the SHF algorithm provides for a customizable application layer for constructing the desired output signal based on the available beams. E.g.
- Communication → Select the beam with the strongest coherent source
- Voice recognition → Select the beam with the voice trigger
- Multi-user → Tracking of multiple users
BeClear SHF comes bundled with a set of pre-defined application layers that are built based on Philips’ extensive domain knowledge.
Differentiators of the BeClear technologies are
- Algorithm automatically adapts to the environment (reverberation, noise conditions) and microphone geometry
- Natural sounding speech
- Highly configurable
- Flexibility in SW adaptability
- Platform (SoC) agnostic
- Multi-channel Acoustic Echo cancellation
• No modification of loudspeaker signals needed
• No need for explicit double talk detector
• Limited divergence during double talk
• Especially beneficial for Stereo AEC and higher
• Reduced complexity mode for bandwidth limited (LFE) signal
• Non-traditional beamforming in near-field and far-field
• High clarity index
• Noise suppression
• Fast adaptation for stationary and non-stationary noise sources
• Robust performance in challenging environments
• Support for point noise source suppression
Target Market Segments and key value propositions
• Segment 1 – Communication: Natural speech. Especially under reverberated conditions, BeClear is able to maintain a high clarity index
• Segment 2 – Voice recognition: Wakeword and voice control/recognition under adverse noise conditions
• Segment 3 – Gaming: distinction of multiple users
BeClear is deployed in the market on ARM Cortex (e.g A53, A35) based SoCs from several chipset vendors.
BeClear does not include wake-word detection software. Philips partners with Sensory and Amazon for the wake-word detection.
Future Markets, and roadmap plans to address target future markets
While BeClear SHF already provides robust point noise source suppression, the BeClear roadmap will continue to provide better performance for even more negative SNR for far field and near field.
The overall description and block diagram of the functional architecture is illustrated in the figure below
The key SW and processing algorithm blocks are:
- Multichannel AEC (scalable to any number of microphone and far-end channels. Currently tested up to 8 (7.1) far-end channels)
- Beamforming (de-reverberation)
- Post-processing (Dynamic echo and noise suppression; Automatic Gain Control (AGC); Direction of Arrival (DoA))
The functional characteristics include:
- Noise reduction methodology
- Microphone Array support: Minimum of 3 (recommended 4) microphones. Agnostic to microphone geometry, both circular and linear arrays are supported
- Direction of Arrival in all directions
- Optional far-end signal processing (in the case of strong non-linearities)
Software and Tools that Philips provides are:
- SDK for the BeClear SHF library (24, 32 and 64 bit) for Windows, Linux and ARM, (other platforms on request)
- Tuning guide, tools and support