BeClear Speech Enhancement Technologies

Philips BeClear Speech Enhancement technologies deliver crisp and clear premium voice quality through a flexible suite of algorithms tailored to communication and voice control use cases.BeClear is the result of Philips’ over 3 decades long history in audio signal processing resulting in a patented set of innovative, but at the same time robust and manufacturable core Speech Enhancement algorithms proven in the marketplace. It is for this wealth of experience that Philips has been able to resolve orthogonal requirements of large distance and high noise in one solution. AI components are incorporated only where these provide a clear and understood benefit over traditional signal processing.

BeClear is Amazon Alexa far-field certified for ASR applications and TEAMS, ZOOM and Amazon Voice certified (depending on acoustic formfactor) for application in communication use cases. BeClear combined with a proper acoustic design formfactor achieves superior performance with a high margin in certification, making BeClear the preferred choice also for more cost-effective designed formfactors.

Philips BeClear is provided as optimized flexible configurable libraries for commercially established MCU and DSP platforms that are available in the market. The underlying BeClear source code offers scalability options to balance MIPS and memory consumption, enabling porting to a wide range of alternative MCU and DSP platforms.

When applicable, the dedicated Philips team provides support during integration, tuning of the algorithm to a specific device formfactor and certification process.

Our mission

Speech is everywhere around us and a vital component in our social interaction with others.

Communicating using speech is low-threshold and natural as everybody carries his speech with him. Apart from the information, speech also conveys emotion making a conversation a personalized experience. Whether you're in the car, in a conference room or in your living room - if you're using technology to communicate, you would like a natural experience without the need to shout or repeat.
Many people experience though that communication by means of speech can become a challenge and often tiresome if there is too much environmental interference such as reflections, background noise sources and other people talking. If in addition, as is the case in many conferencing systems, only one person can talk at the time (half duplex), effective communication becomes close to impossible. Next to communication, speech is increasingly being used for user control interfaces. For example: ‘hands-free’ control of your television or in home control to switch lights or select music.

Our mission is to provide the consumer with a clean channel for communication and/ or voice control usage under all circumstances.

BeClear Target Markets

With highly configurable libraries and smart scalability that optimizes both MIPS and memory, BeClear delivers the perfect blend of performance and efficiency—making it the go-to solution across diverse markets.

System on Chip solutions

System on Chip (SoC) suppliers have the opportunity to offer BeClear to their end customers as an integral part of their SoC propositions. In addition to adapting the BeClear algorithms for the SoC, the dedicated Philips BeClear team is there to assist. They provide comprehensive training for fine-tuning the algorithms and offer reliable second-line support to ensure a smooth experience.

Conferencing

Our social work interactions have transformed into hybrid settings, where individuals and groups connect from various locations. To ensure effective communication, it is crucial that speech feels natural and intelligibility is prioritized.

Using BeClear SuperHandsFree each participant feels equally immersed in the conference call. The BeClear Satellite processing extension seamlessly extends this experience towards larger conferencing rooms.

Mobile and wearable devices

Our mobile devices enable us to stay connected, wherever we are. But using our phones or wearables in noisy environments can be a challenge.

We also increasingly use our mobile phones in speaker mode, held at arm’s length, for video calls or to let others join in on the call. And this will only increase with wearable devices such as smartwatches.

Due to their size, wearable devices are also driving a growing demand for voice input and control. The devices need to be able to recognize commands reliably to be able to execute them without the user needed to repeat them several times.

Whether using voice control or making a call, the BeClear Speech Enhancement algorithm ensures clear, relaxed and effortless communication.

Smart home

In the smart home, there is also a trend towards increased voice interaction and control of our devices. We will no longer need to search for the remote control, we can just tell the television to switch channels or play the latest episode of our favorite show.

Because of their centralized position in the Smart Home, this means that these devices need to be able to capture our voices at distances larger than 5m.

Using voice control, the BeClear Speech Enhancement algorithm ensures clear, relaxed and effortless interactions.

Automotive

We still spend a lot of time on the road. Hands-free calling enables us to stay in touch while in the car. There is also is a growing demand for using voice commands to operate the smart features in our car, from programming the GPS to making a call or changing the radio station. Voice control is much safer than fiddling with controls or trying to type in a phone number while driving.

Current voice recognition engines in in-car entertainment system work quite well when the car is not moving. But the noise in and around the car is severe and this makes it hard to recognize speech. With BeClear Speech Enhancement, voice recognition and calling in the noisy car environment are like a face to face conversation, relaxed and effortless.

BeClear offering

BeClear comprises of a set of extensions that are built around the core algorithm BeClear SuperHandsFree (SHF) configured for wideband (WB) operation (Figure 1).

BeClear SHF can flexibly and seamlessly be combined with one or more of its extensions to address virtually any speech enhancement use case as exemplified in Table 1.

Extension	Example use cases
Bandwidth Extension	High quality conferencing
Multichannel AEC	Soundbars
Satellite processing	Larger conference rooms
Direction of Arrival	Light-ring or camera control based on active speaker
Co-watching	Communication during shared content viewing (Gaming or sports)

Table 1 – Example use cases for BeClear SHF in combination with BeClear extensions

Figure 1 – BeClear offers several extensions build around the core BeClear SHF algorithm

BeClear SuperHandsFree Core Algorithm
BeClear SuperHandsFree Core Algorithm
BeClear SHF is designed to provide enhanced speech under reverberant and noisy conditions at distances beyond 5m.

As illustrated in Figure 2, built around non-traditional adaptive beamforming technology, a fast tracking free-running beam is deployed that quickly adapts towards the strongest coherent source captured by the microphone array. The free running beam detects new coherent sources and allocates one or more focused beams capturing these sources such as a desired speaker or a point noise source such as a radio or TV. Incoherent or diffuse noise sources, such as an air conditioner or nearby traffic will be attenuated. Due to their slower adaptation, compared to the free-running beam, an improved Signal to Noise ratio can be achieved for sources in the focused beams.

The application of dynamic diffuse noise suppression makes SHF suitable for use in adverse acoustic environments.

Figure 2 – The BeClear SHF fast tracking Free running beam (shown in orange) allocates a number of Focused beams (1..N shown in light blue).

The SHF algorithm adaptively optimizes its outputs regardless of the number of microphones or their arrangement, not limited to circular and linear microphone arrays.

The algorithm provides separate optimized outputs for communication as well as the speech control/recognition use cases.

The 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 a voice trigger (WakeWord)

Multi-user → Tracking of multiple users (e.g., for conferencing)
BeClear SHF 2.0
BeClear SHF 2.0
Philips recently released the next generation SHF 2.0, building upon the successful BeClear SHF.

Figure 3 – BeClear SHF 2.0 focus on desired sources in the presence of a strong interferer (shown as the red beam)

BeClear SHF 2.0 outperforms BeClear SHF version in multiple ways. SHF 2.0.

uses AI technology to determine whether a source contains desired speech content or (undesired) non-speech content.

comprises a new building block that adaptively untangles the diffuse tail and a single undesired interferer (viz., the noise beam in Figure 3); allowing the adaptive beamformer to focus on a weaker desired source (Figure 4). For example, SHF 2.0 picks up the desired speaker from a distance with a radio playing right next to the microphone array.

Figure 4 – Left Pane: With SHF, the Free Running (FR) beam may be locked onto a strong interfering source (N, shown in red). Right pane: Using SHF 2.0, the Free Running beam is not distracted anymore by the strong interfering source (n, shown in red) and can proceed allocating focussed beams to sources (Sx)

not only decorrelates point noise sources but also the diffuse field for low frequencies. As a result, the beamformer will not bias anymore to the diffuse background noise and is able to focus on those sources for low frequencies, thereby better handle smaller microphone arrays.

employs a new echo suppression architecture that results in better suppression of echoes during far-end only, while at the same time significantly improving double talk performance. As a positive side-effect of this new approach of echo suppression, the tuning process is greatly simplified.

allows for better de-reverberation at low frequencies, which results in better speech intelligibility in more reverberant environments.

de-reverberation is de-coupled from the non-stationary noise-estimate, allowing for more aggressive non-stationary noise suppression without significantly affecting the near-end speech.

within a single algorithm provides simultaneous optimized output for communication and voice recognition/ control use cases.
Multi-channel AEC

Multi-channel AEC

BeClear Multi-channel AEC provides state-of-the-art full duplex AEC beyond mono and stereo AEC. The architecture is scalable to any number of microphone channels and more than 2 speakers as used in e.g., 7.1 channel soundbars. BeClear Multi-channel AEC provides for several flexible resource optimizations for reduced bandwidth (e.g., subwoofer low frequency extension (LFE) channels as in a 7.1 setup) and correlated loudspeaker channels (e.g., downmixed loudspeaker channels or virtualizers). These optimizations are particularly relevant for AEC configurations required for Dolby ATMOS content (e.g., 5.1.2 or 7.1.4 configurations), using a higher number of loudspeaker channels (e.g., 6 to 12 or beyond).
Bandwidth Extension

Bandwidth Extension

SHF and SHF 2.0 are available for WideBand (WB, 16kHz sampling rate) and SuperWideBand (SWB, 32kHz sampling rate). SWB delivers significantly improved voice quality compared to WB by extending the frequency range, resulting in a more natural and detailed audio experience. This expanded range captures more nuances of speech and music, leading to enhanced clarity, reduced listening fatigue, and improved intelligibility, particularly for challenging sounds like sibilants (s, f).
Direction of Arrival

Direction of Arrival

The BeClear Direction of Arrival (DoA) extension provides azimuth angles for each of the allocated beams including the free running beam. This information may be used by the application to enhance the user experience. For example, to direct a camera or light ring. In order to employ DoA, the microphone geometry must be known so that the microphone formfactor can be mapped to world-coordinates.
Satellite processing
Satellite processing
Larger rooms benefit from adding one or more satellite units. This not only improves clarity for speakers outside the critical (reverberation) distance, but it also improves Double Talk due to larger loudspeaker microphone distances.

The BeClear Satellite extension comprises of the BeClear Satellite Combiner and (optional) BeClear Satellite Processing modules allowing for robust satellite configurations.

The BeClear Satellite Combiner module combines outputs of multiple units into a single output (see Figure 5). These units can be based on BeClear SHF, a single microphone and even microphone arrays, such as ceiling arrays. Automatic Gain Control (AGC) is applied on the combined signal.

Figure 5 - The BeClear Satellite Combiner combines outputs of multiple BeClear SHF units (shown by the blue boxes) into a single output

for satellite units where no BeClear SHF processing is applied (e.g., single mic), the BeClear Satellite Processing module is required (see Figure 6). In that case additional echo and stationary noise suppression is applied.

Figure 6 - For satellite units in which no BeClear SHF processing is applied (e.g., single mic), the BeClear Satellite Processing module is required

The location of the libraries for the satellite combiner and processing modules, as well as the number of satellite units into BeClear Satellite Combiner is scalable and depending on the use case and available resources.
Co-watching

Co-watching

Consider the use case where two users are communicating while at the same time playing audio content. For example: with reference to Figure 7, User-A and User-B both watching the same football match (or playing the same game) on TV-A and TV-B and discussing the match (or game) over a (video)-chat. Since the loudspeakers on both ends are continuously active playing the match or game sound, the AEC is applied in continuous double-talk. This will not only cancel the far-end signal played over the loudspeaker, but it will also distort the desired near-end speech resulting in poor intelligibility on the other side (and vice-versa).

Figure 7 – Using BeClear Co-watching, users can enjoy high quality communication in the presence of match or game sound.

Using BeClear co-watching, perceptual masking is employed to suppress the far-end signal to an extend that the quality of the desired near-end speech is still retained with high intelligibility. As a result, users can enjoy a clear communication channel in the presence of the match or game sound. In Beclear co-watching, the far-end signal and match or game sound are separately available as inputs.

Hear the difference

Experience how BeClear improves the speech for your use case.

Noise Suppression
In the example below you can experience how Philips noise suppression is able to almost completely remove the background noise. In the examples the (unprocessed) microphone signal is provided as well as the processed output. Suppression is set to 30dB. Different types and levels of noise are added.

Babble noise

Noise level	Microphone input	After processing
50 dB	(586.0KB)	(586.0KB)
60 dB	(596.0KB)	(596.0KB)

Stationary Noise

Noise level	Microphone input	After processing
50 dB	(547.0KB)	(547.0KB)
60 dB	(546.0KB)	(546.0KB)

De-reverberation
In the example below you can experience how Philips de-reverberation suppresses the reverberant echoes. In the example the (unprocessed) microphone signal is provided as well as the processed output. The user is at a 4 meter distance in a room with a reverberation time T60 = 800ms.

Microphone input	After processing
(198.0KB)	(198.0KB)

AEC Double talk performance
In the example below you can experience how the Philips Acoustic Echo Canceller is able to maintain full duplex. In the example the (unprocessed) microphone signal is provided as well as the processed output. You will hear that the echo resulting from the far-end (female) voice is suppressed in the processed signal whereas the near-end (male) voice remains intelligible without in double talk. The users are at a 4 meter distance in a room with a reverberation time T60 = 800ms. The far end loudspeaker is at a (center-center) distance of 60cm from the microphone array. Note that AEC results are very tightly connected to the microphone and speaker configuration in the product design.

Microphone input	After processing
(198.0 KB)	(198.0KB)

Point-noise suppression

In the example below you can experience how Philips point-noise suppression is able to almost completely remove an interfering source. In the examples the (unprocessed) microphone signal is provided as well as the processed output, both for communication usage as well as input to a WakeWord or speech recognition engine (ASR).

Mic1 input (Unprocessed)	Output (Processed)

A1 Dishwasher	(2.6 MB)	A2 Communication	(2.6 MB)
		A3 ASR	(2.6 MB)
B1 Motor Noise	(1.92 MB)	B2 Communication	(1.92 MB)

De-reverberation

De-reverberation

In the example below you can experience how Philips de-reverberation suppresses the reverberant echoes. In the example the (unprocessed) microphone signal is provided as well as the processed output. The user is at 1.5 meter distance from the microphone array in a highly reverberant room with a reverberation time T60 = 800-900ms.

Mic1 input (Unprocessed)

Output (Processed)

A1 Reverberated

(2.68 MB)

A2 De-reverberated

(2.68 MB)

Echo suppression

In the example below you can experience how the Philips Acoustic Echo Canceller is able to maintain full duplex in a soundbar formfactor. In the example the (unprocessed) microphone signal is provided as well as the processed output. For better comparison, the processed output is provided for the case when the far-end is active and when the far-end is not active. You will hear that the echo resulting from the far-end voice is suppressed in the processed signal whereas the near-end voice remains intelligible without in double talk. The near-end user is at 4.5 meter distance in a room with a reverberation time T60 = 400ms. Note that AEC results are very tightly connected to the microphone and speaker configuration in the product design.

Mic1 input (Unprocessed)	Output (Processed)

A1 Far-end + Near-end desired Speech	(1.9 MB)	A2 Far-end Active	(1.9 MB)
		A3 No Far-end Active	(1.9 MB)

Linear Multi-channel Echo Cancellation

Linear Multi-channel Echo Cancellation

BeClear linear AEC compensates for the echo by estimating the acoustic paths between the loudspeakers and microphones. Further echo suppression to completely remove the residual echo is obtained by BeClear echo suppression shown in the Echo suppression demo.

The example below provides the unprocessed microphone signal containing both the near-end and far-end signals and the processed output signal removing the linear AEC component. The example demonstrates a significant reduction of the far-end signal in the processed output signal as a result of the linear AEC.

Mic1 input (Unprocessed)

Output (Processed)

A1 Far-end Music + Near-end desired Speech

(5.87 MB)

A2 Linear 7.1 channel AEC cancellation

(5.87 MB)

Amazon Partner page

BeClear SuperHandsFree (SHF) provides high quality, clear speech for speech recognition and communication at larger distances, even in noisy environments.

Visit our Amazon Partner page

Contact us

* This field is mandatory

Country

State or Province

*

First name *

*

Last name *

Company

*

Role *

*

Email address *

Telephone number

*

Company address *

*

Website *

*

Current active territories *

*

Company annual sales *

*

Main products and or services *

*

Current brands and / or licenses if applicable *

*

Distribution channels and key customers *

*

Company overview / Key information relevant to your company *

*

Interested Categories for licensing *

*

Key information relevant to your enquiry *

Philips values and respects your privacy. Please read our privacy notice for more information.