Glossary

Acoustic monitoring processes/methods can be used for sensor-based process monitoring or to determine/evaluate product properties. Applications can be found, for example, in quality assurance or predictive maintenance, where acoustic signatures (and/or changes to them) are analysed for the condition and error monitoring of machines, plant and products. Increased/higher product quality and its assurance, as well as less/plannable/minimised downtime can generate added value. 

Artificial intelligence (AI) is a form of machine learning that imitates human abilities, such as problem-solving strategies, text comprehension, audio enhancement, acoustic pattern recognition or image analysis. Typically, deep artificial neural networks are used, which model the human nervous system and transpose it into machine-readable form. These DNNs can be trained or adopted with the help of a large amount of data. 

An artificial neural network is mimicking some structures of the human brain. Both consist of a large number of individual neurons connected to each other. These neurons receive, weight, and summarise incoming signals and forward the modified result. In artificial neural networks, several groups of neurons arranged on top of each other form what is called a layer. Many successive layers are aggregated to form a deep artificial neural network. The parameters that determine how the individual inputs of a single neuron are weighted are variable and optimised during training so that the task in question is optimally solved. In this way, phonemes, sounds, speech, and other items can be recognised. 

Audio signal enhancement describes methods used to process acoustic signals in such a way that the highest possible quality is achieved at the point of processing. Processing can be done by a person or a machine. Apart from the selection of optimal sound transducer components with the associated digitalisation and specification of their number, audio signal enhancement also includes processing using digital signal processing algorithms. By means of AI-based background noise reduction, voices can be extracted from a mixture with such noise, for example, or a single voice with which the AI has previously been trained can be extracted from a babble of voices. However, the target signal can also be something other than a voice, e.g. people snoring at the same time can be split into individual audio tracks.

Augmentative and Alternative Communication (AAC) encompasses various no-tech and low-tech options for people of all ages who have severe speech or language problems. In addition to gestures and facial expressions, symbol cards or communication boards, speech-generating devices, tablets, and other technical methods are used to find other ways to communicate besides talking.

Speech recognition is the process that enables computers to automatically transcribe spoken language into written words and phrases. This can be used, for example, to control devices with voice commands or to transcribe free text automatically.

What is understood by beamforming is a signal processing strategy in which several microphones are offset in such a way that sound is no longer received from all directions simultaneously but from a defined direction – basically a kind of audio focus. Intelligent algorithms or the user can select and process specific directions from which the microphone array captures the audio signals so that, for example, the microphone is directed in telephone conferences to where it is needed at the time – for instance towards the speaker. 

Binaural hearing, also known as directional hearing, refers to spatial hearing using both ears. It involves processing two incoming acoustic signals to create an auditory impression. This enables us to localize the direction of a sound source and distinguish between different sound sources. Binaural hearing allows us to focus on a specific voice even in challenging listening environments, especially when it originates from a different direction than the background noise.

Biomarkers for medicine or biology are measurable parameters of biological processes that provide prognostic or diagnostic information. These can be measurable changes, e. g. in blood count or in continuous data series such as speech or EEG. Our “Assistive Speech and Language Analysis Group”, for example, uses features of the speech signal as biomarkers that can help in the early detection and prevention of diseases. Biomarkers in this context can be the voice, the speed of speech or the words a person uses in everyday life.

Computer linguistics develops and studies models that can be used to understand and interpret human language in both oral and written form based on algorithms. It forms the interface between linguistics and computer science and is often referred to as computational linguistics or natural language processing in English-language literature.

The electrical activity of the brain is measured with the help of electroencephalography; the data obtained in this way are called electroencephalograms. The abbreviation for both is EEG. For an EEG, sensors are attached to various parts of the head to measure the natural changes in voltage caused by brain activity. To visualize the tiny spikes, an amplifier is required. If certain neurological disorders or brain damage are suspected, as well as for monitoring during operations or in the sleep laboratory, an accompanying EEG is useful.

Electromyography (EMG) is the use of electrodes to measure the electrical activity of muscles. This neurological examination makes it possible to draw conclusions about muscle health and the respective nerves.

In an end-of-line test (EOL), a quality check is carried out at the end of the production line, i.e. the end of the line. Another test system is in-line testing.

Hearables are intelligent earphones, including headphones, headsets, or hearing aids, that offer additional functionalities. Our technologies for smart hearables are focused on industrial workplaces, and e. g. can include voice control for production machines and provide acoustic monitoring capabilities.

Industry 4.0 (i4.0) is the term used to describe the digitalisation of the entire value chain in industrial production. Networked systems enable the digital transformation of machines and processes to make the factory of tomorrow more efficient and flexible.

Industry 5.0 (i5.0) is the further development of i4.0. Industry 5.0 focuses on the interface between humans and machines. The aim is to use artificial intelligence (AI) and innovative technologies to make work processes more user-centric and intuitive by humans and machines working alongside each other. 

Alongside end-of-line testing, in-line testing is another test procedure, in which a series of quality checks are already conducted during production.

Language disorders are language cognition impairments and can affect all language modalities. Both speech comprehension and speech production can be affected at the level of sound structure, vocabulary, meaning and word and sentence grammar. Language disorders can be caused by developmental factors or neurogenic diseases.

Machine learning is a general description for algorithms designed to teach a machine (involving neural networks) to solve a specific problem. To archive this, sufficiently large amounts of known and annotated data (images, audio, metadata, etc.) are presented to the network both on the input data (source) and on the output side (expected result). With various optimization strategies, the weights inside the neural network are adjusted until - for a given input data - the calculated output at the last layer of the neural network matches the expected result.

A microphone array describes an arrangement of several microphones. By adding adaptive algorithms to individual microphone signals, acoustic events can be spatially localised and signal recording optimised specifically for one direction, e.g. a speaker. Two- or three-dimensional arrangements of two up to any number of microphones are possible.

Neurotechnology sees the brain as the interface between humans and information/communication systems. This can be the measurement of brain activity to obtain information about cognitive or biological processes or to control devices, as well as the electrical stimulation of brain structures as a treatment method. Neurotechnology encompasses hardware in the form of sensors and amplifiers and the underlying information technology for signal processing.

The purpose of proactive machine maintenance processes is to avoid downtime in production. This is also known as predictive maintenance.

Radar is the term used to describe localization and distance measurement by means of radio waves. Radar is an acronym for “radio detection and ranging”. Radar-based measuring devices can also be used to record vital signs such as breathing or pulse, for example. The technology is also used in other areas, such as aviation and seafaring.

Sensor networks consist of different and often spatially distributed sensors (nodes) that perform a specific task. The result data from the individual sensors are often sent to a parent instance. This parent instance makes individual raw sensor data available to the user or fuses and aggregates them into higher-value information for the user.

Signal processing is a general description for the analysis and processing of signals. Signals in this context can take different forms: audio signals, EEG, images, videos, etc. Nowadays, digital signal processing prevails in everyday use. Machine learning, but also sound adjustment methods, such as the use of equalisers or dynamic compressors, are examples of signal processing methods.

Sound quality is a subjective perception that varies from listener to listener. Several factors, including room acoustics, recording or amplifier quality, and individual preferences, influence how sound quality is evaluated. Context also plays a significant role in determining what is considered desirable sound. Psychoacoustics can help assess how product features impact sound quality and, conversely, how the sound of a product influences its perceived quality.

The term “speech disorder” describes an impairment in motor and/or articulatory abilities. Symptoms range from incorrect production of individual speech sounds, changes in speech tempo and dynamics to rhythm or fluency disorders. Speech disorders can be caused by developmental factors, structural changes to the vocal organs or neurogenic diseases.

Speech intelligibility refers to the extent to which spoken language can be understood and comprehended by a listener. It can be determined through listening experiments or computational models. Listening experiments involve participants rating or transcribing speech samples to assess intelligibility. Computational models use algorithms to objectively estimate intelligibility based on acoustic and linguistic features. Both approaches contribute to understanding and improving speech intelligibility.

Automatic speech processing methods in a medical context are algorithmic and computer-based techniques for the objective evaluation of the acoustic speech signal. These methods can be used in diagnostic or therapeutic analysis to detect speech, language or voice disorders. Various speech processing technologies can be used, including digital signal processing or automatic speech recognition.

Voice disorders are impairments of vocal function and therefore primarily characterised by limited vocal performance and changes in vocal sound. The causes of such an impairment can be functional (e. g. voice use) or organic (e. g. paralysis, tumours). Changes in voice tone range from hoarse, breathy and quiet to rough, strained or raspy. In addition to mental factors and breathing, the use of the laryngeal muscles plays an important role in phonation.

Voice filter technology is used to extract certain speakers from an audio recording in order to improve speech intelligibility.