Modular Audio Front End "AFEnext"

Natural speech input and optimal speech output in your applications and products

We talk to our smart speaker, our car or our robot vacuum cleaner. Voice control and voice documentation are even playing an increasingly important role in industry and hospitality. At the same time, devices, machines and applications give us direct audio feedback. The optimum conditions are not always available for this voice communication. In a video call, the distance to the microphone in the notebook or smartphone is large, the babble of voices in the restaurant makes every conversation a challenge and loud car interior sounds cause noise in the voice recording while driving. Customers also expect first-class sound from the smallest speaker for the best reproduction of speech and music.

With "AFEnext", we have developed an audio front end that uses intelligent algorithms and processes to adapt to a wide range of application scenarios and create the optimum basis for voice input. This also includes technology modules that enable customized audio output.

AI Denoising

"AFEnext" ensures the best possible recording of speech by intelligently filtering out unwanted background noise, other speakers, and removing reverberation. This only requires the recording of one microphone (instead of several, as is usually the case). Using machine learning methods, "AFEnext" offers a particularly reliable clean-up of the audio signal even under challenging conditions.

Speaker engine for professional sound

Big sound from small speakers: The "AFEnext" speaker engine improves the sound of music and speech. This means that even devices that don't focus on optimal speakers score with professional sound. Maximum transparency for Beethoven's Ninth? Or the best speech intelligibility for news broadcasts? We offer sound tuning from simple, switchable sound profiles to fully automatic, adaptive equalizer control via noise reference microphone. This means that the announcement in the shopping center remains intelligible even on a crowded Saturday afternoon.

Our speaker engine has the cone excursion under control at all times and makes speakers sound bigger than they actually are. Cabinet resonances are effectively suppressed. RMS, peak and thermal limiting ensure optimum protection against overloading.

Robust standard procedures

Automatic Gain Control (AGC)

The AGC, also known as automatic gain control, ensures that the microphone signal is always recorded at the same volume. If a speaker is standing further away in the room, the signal entering the microphones is significantly softer than if the person is standing directly in front of the microphone. The AGC automatically compensates for this difference in level without undesirably amplifying quiet background noise.

 

Beamforming

Beamforming is the technique of combining signals from several microphones into a single one, thereby achieving a desired directional effect towards a specific sound source. By focusing on a speech source in the room, lateral noise is attenuated, and speech intelligibility is increased. What humans can do intuitively (listen carefully) is not a matter of course for machines.

 

Echo Cancellation

Everyone knows it from phone calls: The person on the other end of the line switches the phone "to loudspeaker" − and suddenly you hear your own voice as an echo. That really disrupts the flow of conversation! The voice signal that comes out of the other person's loudspeaker goes back into the microphone. But a good echo canceler filters out the echo before it is sent over the line. This is no easy task, especially when both partners are speaking at the same time. Our echo cancellation can handle any scenario and is freely scalable, even on any number of loudspeakers and microphone channels simultaneously.

 

Dereverberation

The natural reverberation of a room can also interfere with phone calls. And the quality of automatic speech recognition also suffers greatly as a result. Blind reverberation suppression, also known as dereverberation, removes disturbing room components from speech and ensures a dry, easily understandable speech signal.

 

Comfort Noise Injection

"Hello, are you still there?" − many people are familiar with this situation from phone calls. If too much background noise is filtered out during pauses in speech, there is complete silence. This gives the impression that the line is "dead". A good comfort noise injection system fills the pauses in speech with an artificially generated noise that corresponds to the natural background noise in terms of timbre and level, so that the impression of a transparent, open communication channel is created despite the massive removal of background noise.

Combine "AFEnext" with other solutions

 

Personalized sound quality

Everyone has a personal sound preference and their own perception of loudness. This is why we develop processes to adapt the audio playback in smartphones, hearables, consumer electronics, notebooks or in-car infotainment systems, for example, to individual hearing requirements.

 

Analyze and optimize speech intelligibility

We develop signal processing methods that ensure better speech intelligibility and less listening effort in a wide range of applications.

 

SI-Live

Better communication with conference systems

A specific problem with video conferencing is that the speakers are often unaware that they are not being understood well at the other end of the line. 

"AFEnext" in different applications

 

Hearable for the smart industrial workplace

We are working on technological solutions for hearables that serve as hearing protection in noisy production environments and at the same time support worker communication.

Automatic voice monitoring for commands on the ship's bridge

An automatic voice monitoring system for the ship's bridge is intended to ensure greater safety. The Fraunhofer IDMT in Oldenburg is developing a helmet order monitoring system together with the start-up ELNAV, based in Split, Croatia.

The Hearing Car

Acoustic sensors are a key component, particularly in the development of autonomous vehicles. In combination with optical sensors, the car's sense of hearing ensures greater safety and reliability in road traffic.

 

»Help!«

Experts at the Fraunhofer IDMT in Oldenburg have developed sound and event-based detectors to make people in emergency situations heard and comply with current health and safety regulations.

Best sound in a small space

Fraunhofer IDMT develops audio technology for Deutsche Telekom's intelligent voice assistant

 

Intelligent acoustic sensor systems for your products and processes

We develop easily integrated sensor systems for recording and processing acoustic data in networks.

 

Audio System Technology

From initial idea to design and construction of the demonstrator to acoustic end-of-line inspection: We can develop audio system solutions for you according to your individual requirements and help you with hardware design, selection of components and algorithms as well as software applications.

 

Speech Recognition for Products and Processes

Speech detection systems allow operation of technical systems without visual or physical contact. The speech recognition technology of the Fraunhofer IDMT in Oldenburg functions robustly - with and without an internet connection.

Funding

The further development of the institute’s branch is funded by the Ministry of Science and Culture of the German State of Lower Saxony and the Volkswagen Foundation within the »Vorab« program.