METHOD FOR CONTROLLING A TELEPROMPTER AND TELEPROMPTER FOR DYNAMIC DISPLAY OF A PREDEFINED TEXT

Description

[0001] The present invention relates to a method for controlling a teleprompter for the dynamic display of a predefined text on a screen and to a corresponding teleprompter for the dynamic display of a text.

[0002] Teleprompters are used for the dynamic display of text and are generally well known. These teleprompters are often used by newsreaders, presenters, or even politicians to dynamically display a predefined text on a screen so that this text can be read out by the speaker or used to support a presentation. The text displayed on the screen can either be read directly by the speaker or, alternatively, projected towards the speaker via a mirror. In some applications, semi-transparent mirrors are used, which are arranged in such a way that the speaker can look towards an audience or a camera while reading the text during their speech or presentation.

[0003] Typically, these teleprompters are controlled by an operator. The operator follows the speaker's speech and compares it with the predefined text. Depending on the speaker's speech rate, the operator adjusts, for example, the speed of the text displayed on the screen. This process is relatively cumbersome and requires additional personnel, who ideally should be trained and, in particular, must also be fluent in the speaker's language.

[0004] As an alternative to using an operator, the speaker can be provided with a control device with which they can control the displayed text. In this context, a control device in the form of a foot pedal is proposed in WO 2016 / 009194 A1. By using the described foot pedal, it is no longer necessary to have an additional person operate the teleprompter. This can reduce the costs incurred, for example, when recording a news broadcast. A disadvantage of this solution, however, is that additional hardware is required. Another disadvantage is that the speaker must concentrate on his or her presentation on the one hand and actively control the text display on the other.

[0005] Furthermore, GB 2389220 A describes a teleprompter that allows a user to display parts of a lecture that are to be delivered in the future. The described teleprompter uses a speech recognition algorithm that evaluates the recorded speech signals.

[0006] In Anonymous: "Apple to Add Teleprompter Tools to iMovie and Final Cut Pro - Patently Apple", February 10, 2012 (2012-02-10), XP055562257, Found on the Internet: URL:https / / www.patentlyapple.com / patently-apple / 2012 / 02 / apple-to-add-teleprompter-tools-to-imovie-and-final-cut-pro.html [found on 2019-02-26] a teleprompter tool is described in which a user can manually adjust his speaking rate via a graphical user interface.

[0007] US 2018 / 070026 A1 describes another teleprompter in which a script comprising text is identified, which script is displayed on a user interface for use during the rendering of a video (e.g., while the user is recording and / or streaming the video).

[0008] In principle, it is also possible to detect the speaker's spoken text and compare it with the predefined text. This would make it possible to control the display of the predefined text depending on the spoken words. This has the advantage that no additional hardware is required. However, this solution is not without its drawbacks. In particular, there is the problem that such a solution is prone to the speaker making a mistake during their speech.

[0009] Based on the problem described above, the object of the present invention is to provide a method for controlling a teleprompter and a corresponding teleprompter which allow the most convenient control of the displayed text, whereby no additional hardware is required and a simple and robust system is provided.

[0010] To achieve the above-mentioned object, the present invention proposes a method for controlling a teleprompter for the dynamic display of a predefined text on a screen of the teleprompter, the method comprising the following steps: Recording at least one identity-specific input of a speaker using a recording device and generating a corresponding electrical signal, recognizing the speaker by evaluating the generated electrical signal, determining at least one display parameter depending on the recognized speaker, dynamically displaying the text on the screen of the teleprompter using the at least one display parameter.

[0011] By recognizing the current speaker and subsequently determining a display parameter based on the recognized speaker, the present invention advantageously enables automatic adaptation of the display parameters, in particular the display speed of the text, to the current speaker. In this way, the optimal display parameters for a specific speaker can be set without the need for an additional operator or additional hardware. This can significantly reduce the production costs for a news broadcast, for example. Furthermore, the present invention offers the advantage that, particularly at international events with multilingual speakers, no teleprompter operator is required who is fluent in all the languages ​​used at the event.

[0012] The teleprompter according to the invention can be configured so that the predefined text moves vertically or horizontally on the screen. Alternatively, the predefined text can also be dynamically displayed on the screen, for example, word by word or line by line. While a conventional teleprompter merely outputs the text, the teleprompter according to the invention is configured so that the teleprompter reacts to the speaker's voice and automatically adjusts at least one display parameter depending on the recognized speaker. The recognition of the speaker by evaluating a generated electrical signal (in particular an audio signal or an image signal) is explained in more detail below.

[0013] According to one embodiment of the teleprompter according to the invention, at least one display parameter determined depending on the recognized speaker can relate to the display speed of the dynamically displayed text on the screen. In particular, a display parameter can be the scrolling speed of the text. For example, a corresponding display parameter can define that the specified text is displayed on the screen at a scrolling speed of, for example, ten lines per minute or 60 words per minute. Alternatively, the display parameter can define that, for example, ten lines of the specified text are displayed per minute or three words of the specified text are displayed per second.

[0014] According to a further embodiment of the method according to the invention, it can be provided that at least one display parameter determined as a function of the known speaker relates to at least a font color, a background color or a font size of the displayed text or a screen configuration parameter. In this way, it can be taken into account which display parameters are perceived as optimal for a specific user. Thus, according to the method according to the invention, it can be provided that upon recognition of a first user, the predetermined text is displayed in black on a white background, while upon recognition of a second user, the display parameters are adjusted such that the predetermined text is displayed in light gray on a black background. Likewise, it can be considered particularly pleasant for a user if the displayed text is displayed in a particularly large font.The screen configuration parameters, such as the brightness of the display or the display contrast, can also be stored for a specific user, so that the preferred screen configuration parameters for a specific speaker are loaded as soon as this speaker is recognized by the method according to the invention.

[0015] According to a preferred embodiment of the invention, it can be provided that the recording of the speaker's identity-specific input includes the recording of a speech input from the speaker; the recording device comprises a microphone; and the electrical signal comprises an audio signal.

[0016] Alternatively or additionally, the method according to the invention can also provide, for example, that the recording of the speaker's identity-specific input includes recording an image of the speaker; the recording device comprises a camera; and the electrical signal comprises an image signal.

[0017] According to a further exemplary embodiment of the method according to the invention, it can be provided that the recognition of the user is based on a detection of the speaker's fundamental frequency. The speaker's fundamental frequency can be determined by evaluating the audio signal. The determination can be carried out in particular in the time domain, in the spectral domain or in the cepstral domain. In the time domain, the audio signal can be evaluated in such a way that periodic regions of the recorded audio signal (in particular recorded voiced tones) are evaluated and the period duration T 0 of the audio signal is determined, with the speaker's fundamental frequency then being calculated according to f 0 = 1 / T 0 . According to one embodiment of the method according to the invention, it can be provided that the speaker's fundamental frequency is repeatedly determined over a predefined period of time and then the average speaker's fundamental frequency is determined.This has the advantage that a varying speaker fundamental frequency, which can be caused, for example, by a particularly pronounced intonation of the speaker, has less influence on the determined and speaker-specific speaker fundamental frequency.

[0018] According to a further embodiment of the method according to the invention, it can be provided that the user recognition is based on a spectral analysis of the audio signal. In particular, it can be provided that the speaker's fundamental frequency is determined in the spectral domain. Determining the speaker's fundamental frequency in the spectral domain can be particularly advantageous because any noise can strongly influence the audio signal in the time domain, so that determining the speaker's fundamental frequency can be difficult, especially in the case of a noisy audio signal in the time domain. In particular, a short-term spectral analysis of the audio signal can be performed to determine the speaker's fundamental frequency in the spectral domain. In this case, individual blocks of the time-dependent audio signal are subjected to a Fourier transformation, in which case an FFT (Fast Fourier Transformation) can be used in particular.The Fourier transformation converts the audio signal into the frequency range where simplified detection of the speaker's fundamental frequency is possible. As explained below in connection with the figures, the speaker's fundamental frequency can be determined as the distance between individual peaks in the audio signal spectrum.

[0019] According to the method according to the invention, the speaker's recognition can also be based on a cepstral analysis of the audio signal and a detection of the speaker's fundamental frequency in the cepstral domain. Cepstral analysis is particularly suitable for determining the speaker's fundamental frequency, as it enables robust and precise recognition and is less susceptible to noise.

[0020] Cepstral analysis essentially involves determining and evaluating the spectrum of the audio signal. The analysis focuses on a given cepstral region where a peak corresponding to the speaker's fundamental frequency is expected. The position of the peak can be used to determine the speaker's fundamental frequency.

[0021] According to one embodiment of the method according to the invention, the speaker's recognition can be based on determining the speaker's fundamental frequency in the spectral or cepstral range, whereby only audio signals that can be assigned to a voiced sound are evaluated for speaker recognition. This has the advantage that the recognition of the speaker's fundamental frequency is carried out with greater precision when evaluating voiced sounds than is possible with unvoiced sounds.

[0022] According to a particularly preferred embodiment of the invention, it can also be provided that a voiced / unvoiced detection is used to determine the audio signals that can be assigned to a voiced sound. Furthermore, it can be provided that the first cepstral coefficient is evaluated for voiced / unvoiced detection. A voiced sound can be inferred if the first cepstral coefficient is greater than zero. In this way, a simple voiced / unvoiced detection in the cepstral range can be implemented, which can significantly increase the precision of speaker recognition.

[0023] Furthermore, according to one embodiment of the method according to the invention, it can be provided that the determination of a display parameter takes place by selecting a display parameter from a provided database in which at least one display parameter is assigned to each of several speakers. The provided database can contain, in particular, speaker fundamental frequencies and display parameters assigned to the speaker fundamental frequencies. It can be provided that the database is stored in a memory of the teleprompter. Alternatively, it can be provided that the database is stored in a separate system accessible via a network connection.

[0024] According to a further embodiment of the invention, it can be provided that the speaker is recognized repeatedly, in particular at predefined and preferably fixed time intervals. This has the particular advantage that a change of speaker can be recognized, so that with the method according to the invention, several speakers can use one and the same teleprompter, while, for example, the display speed of the text is automatically adapted to the currently active speaker. It can also be provided that a quasi-continuous detection of the speaker's fundamental frequency takes place, so that, according to the method according to the invention, a timely adjustment of the display parameters is possible. Furthermore, it can be provided, for example, that the display parameter is always redetermined as soon as a change in the detected speaker's fundamental frequency is detected.

[0025] Furthermore, according to one embodiment of the method according to the invention, the method can comprise speech pause detection and repeated speaker recognition, wherein the interval between two consecutive speaker recognition processes occurs dynamically and as a function of the detected speech pauses. This advantageously ensures that a speaker change and a corresponding adjustment of the display parameters takes place reliably as soon as the first speaker has finished their presentation and the second speaker has begun their presentation. This also advantageously ensures that the resources of the teleprompter are used particularly efficiently, since permanent speaker recognition is not required, as the number of speaker recognition processes is reduced by speaker recognition only taking place after each individual speech pause.

[0026] Furthermore, according to a further embodiment of the method according to the invention, the recording of the speaker's speech input can be carried out using a microphone array and a beamforming method can be used to increase the accuracy of speaker recognition. The use of the beamforming method has the advantage that the position of the current speaker can be detected and the directional characteristic of the recording device can be focused on the current speaker. In this case, the microphone array can amplify the speech signal received from the speaker while suppressing background noise (for example, external noise or any conversations of other people in the room). In this way, particularly precise recognition of the speaker can be ensured.

[0027] It can also be provided that the speaker's recognition is based on facial recognition. Using a facial recognition algorithm can ensure particularly accurate speaker recognition. It can also be provided that both the speaker's voice input and a recorded image of the speaker are evaluated for speaker recognition. By evaluating the additional information, the accuracy of speaker recognition can be further increased.

[0028] Furthermore, it can be provided, in particular, that the speaker's recognition is based on facial recognition. Facial recognition can be used either alone or in combination with the voice recognition described above.

[0029] According to a preferred embodiment, the method for recognizing the speaker may include extracting several features from the image signal and subsequently evaluating the features. For example, it may be provided that features that are characteristic of a person's face are evaluated. These features may also be referred to herein as facial features.

[0030] Within the scope of the present invention, numerous different implementation variants can be used for the extraction of the features as well as for the classification of the features.

[0031] In particular, according to the present invention, it can be provided that a method for face recognition according to Paul Viola and Michael Jones is implemented. The aim of this method is to find a specific face and discard the irrelevant background. This is based on a window with a predefined size, which is moved over the entire image to be examined. Each individual image section is then examined for features. In particular, Haar features (named after the Haar wavelets developed by Alfred Haar) can be used for this purpose. The Haar features calculate brightness differences within the respective image section, with the position and scaling being freely selectable. Large-area features filter out low-frequency features (differences in brightness), whereas narrow features filter out high-frequency features (e.g. corners and edges).The calculation is performed by summing pixels in the respective block regions. With a standard window size of 24 x 24 pixels, the Viola and Jones algorithm results in a total of between 160,000 and 180,000 such differently scaled features.

[0032] To ensure reliable facial recognition, individual Haar features can be combined into classifiers, which jointly decide by majority vote whether the object being viewed represents the face of a specific person or not. The classifiers can be trained using a training dataset. The classifier learns from numerous test images of different people. The more test images provided, the higher the expected accuracy of the subsequent facial recognition. The classifier can be based on various methods and models. Various procedures known in the field of machine learning can be used. In particular, it can be planned to use an artificial neural network that is trained using the test images and then used as a classifier for facial recognition.

[0033] Furthermore, according to the present invention, the feature recognition used for speaker recognition can be based on the SURF / SIFT algorithm and a histogram comparison. SURF (Speeded Up Robust Features) is an accelerated algorithm for the robust detection of prominent features in images. It is based on mean filters and integral images. It was preceded by the "Scale-invariant Feature Transform" (SIFT), which, in contrast, works with Gaussian filters. After an image has been smoothed using the appropriate filter, the prominent points are determined. They are prominent when they deviate significantly from their background. For this purpose, their gradients (e.g., at corners or edges) are assigned to all points and stored locally as a histogram. The most prominent gradients are then retained as features.The major advantage of these features is that they are insensitive to changes in illumination and perspective distortions (rotation, translation, scaling). Grayscale images are typically used for these methods due to their widespread availability. However, more and better robust features can be found in color images with their three-dimensional feature space, which is expanded by two dimensions. The use of color images for the SURF or SIFT algorithm is therefore recommended and can help ensure a particularly high recognition rate. Using the RANSAC10 algorithm (not described in detail here), matches can then be searched for in two images with prominent, homologous points and thus compared.

[0034] Furthermore, it may be intended that the body or stature of a speaker be evaluated to uniquely identify them. The features used in this context can also be referred to as body features. The body features may be used alone or in combination with other features, particularly facial features. By combining different types of features, the accuracy of speaker recognition can be significantly increased.

[0035] Furthermore, according to the present invention, it can be provided that the extracted features or feature combinations are compared with features or feature combinations that are stored in a database, wherein features or feature combinations as well as speakers assigned to the features or feature combinations are stored in the database.

[0036] Furthermore, to solve the above-mentioned problem, a teleprompter for dynamically displaying a predefined text is proposed, with a screen for displaying the text, a recording device for recording an identity-specific input, an evaluation unit for converting the identity-specific input into an electrical signal, for evaluating the electrical signal and for recognizing the speaker, and a control unit for generating a control signal for controlling the screen for dynamically displaying the text using the at least one display parameter and depending on the recognized speaker.

[0037] According to one embodiment of the teleprompter according to the invention, the teleprompter can be provided with a microphone or a microphone array. The microphone can record the speaker's voice. The microphone array offers the advantage of localizing the active speaker, allowing for the suppression of distracting noises using a beamforming technique.

[0038] According to the present invention, the teleprompter may also comprise a camera or a camera array.

[0039] Furthermore, the teleprompter according to the invention can be provided with the recording device having a storage unit in which a database is stored containing data about a plurality of speakers as well as the display parameters assigned to the speakers. The database can either be statically predetermined or can be expanded dynamically over time. For this purpose, it can be provided, for example, that a machine learning method is used, through which the teleprompter according to the invention learns over time which display parameters are preferable for a specific speaker. Thus, the device according to the invention can learn over time that a first display speed should be selected for a first speaker, while a faster display speed is appropriate for a second speaker.In particular, the learning process can evaluate data from the past in which a teleprompter operator manually set the display parameters for a specific speaker.

[0040] According to a further embodiment of the device according to the invention, the teleprompter can be provided with a network module for connecting to a cloud storage unit. This allows all data, including information about speaker characteristics (e.g., the speaker's fundamental frequency), and display parameters associated with the speakers to be stored in a separate database, so that the teleprompter does not require a large amount of memory.

[0041] The present invention is explained below with reference to the figures. The figures show various embodiments of the invention, although it is understood that the present invention is not limited to the embodiments shown in the figures. In the following figures, the Fig. 1 shows an embodiment of the teleprompter according to the invention, Fig. 2 shows an overview of an embodiment of the method according to the invention, Fig. 3 shows a detailed representation of the individual method steps of speaker recognition, Fig. 4 shows a further embodiment of speaker recognition, Fig. 5 shows an audio signal in the time domain, Fig. 6 shows an audio signal in the frequency domain, Fig. 7 shows an audio signal in the cepstral range and Fig. 8 shows an example of the data stored in a database for different speakers.

[0042] In the Fig. 1 An embodiment of the teleprompter 10 according to the invention is shown. The teleprompter 10 has, in particular, a screen 12 on which a predetermined text is dynamically displayed. In particular, the text can scroll vertically along the screen 12 at a predetermined speed. The displayed text can, for example, be a news text that a newsreader can read from the screen 12. Furthermore, the teleprompter 10 has a recording device 14 for recording a voice input from a speaker. In the Fig. 1 In the illustrated embodiment, the recording device 14 has a microphone array 16, which is composed in particular of eight microphones 18. The recording device 14 is therefore designed, on the one hand, to record the speaker's voice and, on the other hand, to implement a beamforming method. This allows the speaker to be localized and the directional characteristic of the microphone array 16 to be aligned to the current speaker. The speaker's voice input recorded by the recording device 14 can be converted into an audio signal and subsequently evaluated. During the evaluation, the recorded audio signal can, in particular, be compared with a database stored in a storage unit 20 in order to identify the current speaker.The display parameters assigned to the specific speaker can then be adjusted to enable the most convenient playback of the displayed text. In the exemplary embodiment shown here, the storage unit 20 is stored in a separate device, enabling data exchange with the storage unit 20 via a network module 21.

[0043] In the Fig. 2 An embodiment of the method S100 according to the invention is shown, comprising method steps S110 to S140. In a first step S110, an acoustic input from the speaker is recorded. The voice input is converted into an audio signal. Subsequently, the speaker is recognized (S120), and the recorded audio signal is evaluated. Subsequently, a display parameter assigned to the speaker is determined depending on the recognized speaker (S130). Finally, the predefined text is displayed on the screen using the display parameter assigned to the specific speaker (S140).

[0044] In the Fig. 3 The individual steps of a first embodiment of speaker recognition (S120A) are shown. First, the recorded audio signal is converted to the frequency domain (S121A).

[0045] Afterwards, in the Fig. 3 In the exemplary embodiment shown, voiced / unvoiced detection is performed. The purpose of voiced / unvoiced detection is to ensure that preferably only those audio frames that represent voiced sounds are used for speaker recognition. This offers the advantage that the fundamental frequency determination provides greater accuracy for voiced sounds than for unvoiced sounds. However, it should be emphasized that voiced / unvoiced detection (S122A) is to be regarded as optional within the scope of the present invention. If a voiced sound has been recognized, several peaks are determined from the audio signal in the frequency domain (S123A). Subsequently, the distance between two peaks of the audio signal in the frequency domain is determined (S124A). For example, the two peaks with the lowest frequencies can be considered and compared with each other.Alternatively, the distance between more than two peaks can be determined and the average value of the individual measured distances can then be calculated. This allows an average speaker fundamental frequency to be determined, at which any irregularities or noise in the audio signal have a lesser influence on the determined speaker fundamental frequency. This can increase the precision of detecting the speaker fundamental frequency. The speaker fundamental frequency can be used to identify the speaker.

[0046] In the Fig. 4 the individual steps of a further embodiment of speaker recognition are shown (S120B). First, the audio signal is converted to the frequency domain. For this purpose, the audio signal is preferably subjected to an FFT (Fast Fourier Transformation). The audio signal is then transformed into the cepstral domain (S122B). For this purpose, the spectrum of the audio signal is determined in the spectral domain. Here, too, an FFT of the already transformed audio signal is preferably carried out. Finally, a peak is determined in a predetermined range (S123B). The position of the peak of the audio signal in the cepstral range can then be used to determine the speaker's fundamental frequency. Alternatively, the position of the peak can be used directly to determine the speaker. For this purpose, for example, the position of the peak can be compared with a database in which further peak positions and speakers assigned to these values ​​are stored.

[0047] In the Fig. 5 An example of an audio signal in the time domain 22 is shown. As can be seen in this figure, the audio signal has a periodic shape, especially in the case of voiced sounds. If the speaker is to be identified by evaluating his or her fundamental frequency, the Fig. 5 The period duration of the represented audio signal can be determined. The fundamental frequency can then be calculated from the period duration, as explained above. While this method yields good results for speech signals with a high SNR (signal-to-noise ratio), reduced accuracy in detecting the speaker's fundamental frequency can be expected at lower SNRs. In these cases, determining the speaker's fundamental frequency in the spectral domain is preferable.

[0048] In the Fig. 6 Therefore, an audio signal in the frequency range 22A is shown. The audio signal 22A can be determined, for example, by subjecting the audio signal in the time domain 22 to a short-time Fourier transform (STFT). In this case, for example, a short time window of the audio signal in the time domain 22 is extracted and subjected to an FFT. The time window can, for example, comprise 20 to 40 ms of the audio signal in the time domain 22. After performing the STFT, the Fig. 6 The audio signal shown is obtained in the frequency range 22A. As shown in the Fig. 6 As can be seen, the audio signal in the frequency range 22A has several peaks 22B. In addition, Fig. 6 an envelope 22C of the audio signal is shown. Fig. 6 it can be seen that the individual peaks 22B are the same distance from one another. This is particularly characteristic of voiced sounds. By determining the distance between two peaks, the fundamental frequency of the speaker can be determined. To do this, for example, the two peaks 22B with the lowest frequencies can be selected and the difference in the position of the two peaks determined. Alternatively, for example, the five peaks 22B or the ten peaks 22B with the lowest frequencies can be selected, the distance between these peaks calculated, and then the mean value of these distances determined. Another alternative for determining the speaker's fundamental frequency is to consider the first peak 22B and, for example, the first and the (1+n)th peak 22B and to divide the difference in the position of these peaks by n. By taking multiple peaks orof peaks that are further apart from each other, the accuracy of the fundamental frequency determination can be increased.

[0049] In the Fig. 7 the audio signal is shown in the cepstral range 22D. The cepstrum of the audio signal can be determined by an FFT of the audio signal in the frequency range 22A. It is characteristic of a cepstrum that a peak 22E can be seen in a central range 24, the position of which is characteristic of the speaker's fundamental frequency. To determine the speaker's fundamental frequency in the cepstral range, it is therefore necessary to define a predetermined cepstral range 24 in which the peak 22E characteristic of the fundamental frequency is to be expected. The predetermined cepstral range 24 has a lower limit 24A and an upper limit 24B. These arise from the fact that the speaker's fundamental frequency is to be expected in a certain range. For example, the speaker's fundamental frequency for a male speaker can be expected in a range of approximately 90 to 160 Hertz, whereas the speaker's fundamental frequency for female speakers is to be expected in a range of approximately180 to 260 Hertz is to be expected. In contrast, the speaker's fundamental frequency in children can be in a range above 300 Hertz. After identifying the position of peak 22E, the speaker's fundamental frequency can be determined, and the determined value can be compared with a database. Alternatively, the position of peak 22E can be used directly to identify a specific speaker. For this purpose, the position of peak 22E can be stored in a database instead of the speaker's fundamental frequency, with each peak position being assigned to a speaker.

[0050] In the Fig. 8 By way of example, a database 26 is shown in which four different speaker fundamental frequencies (120 Hertz, 140 Hertz, 220 Hertz and 235 Hertz) are stored, which are assigned to four different speakers (Hans, Peter, Jana and Sarah). If, for example, a speaker fundamental frequency of 120 Hertz is detected according to the method according to the invention, the display parameters (AP1, AP2, AP3 and AP4) stored for the speaker Hans can be automatically selected. After the Fig. 8 In the example shown, the text shown on the teleprompter screen can be shown at a display speed of six lines per minute (AP1), while the displayed text is shown on the screen in a font size of 20 (AP2). Furthermore, in the example shown, the displayed text can be shown in black and on a white background (AP3 and AP4). If, on the other hand, a speaker fundamental frequency of 220 Hertz is detected, a display speed of 12 lines per minute (AP1) stored for the speaker Jana is automatically selected, with the displayed text being shown on the screen in a font size of 25 (AP2). The database 26 also notes that the text to be shown to the speaker Jana should be shown in white and on a black background (AP3 and AP4).If the speaker changes, the system can adapt the display parameters to the current speaker within a very short time.

[0051] For illustrative purposes, the present invention has been explained primarily in connection with speaker recognition based on the recognition of the speaker's fundamental frequency. However, it will be apparent to those skilled in the art that the invention is not limited to this. Rather, other characteristic properties suitable for distinguishing between different speakers can also be used. In particular, within the scope of the present invention, it is also possible to extract features from the recorded audio signal that represent the intonation characteristic of a specific speaker. Furthermore, it is possible to extract features from the recorded audio signal that describe the speaker's pronunciation or dialect. BEZUGSZEICHENLISTE

[0052] 10Teleprompter 12Screen 14Recording device 16Microphone array 18Microphone 20Storage unit 21Network module 22Audio signal in the time domain 22AAudio signal in the frequency domain 22BPeak of the audio signal in the frequency domain 22CEnvelope of the audio signal in the frequency domain 22DAudio signal in the cepstral range 22EPeak of the audio signal in the cepstral range 24Predefined cepstral range 24ALower limit of the predefined cepstral range 24BUpper limit of the predefined cepstral range 26Database S100Inventive method S110First step of the inventive method S120Second step of the inventive method S130Third step of the inventive method S140Fourth step of the inventive method S120AFirst embodiment of speaker recognition S120BSecond embodiment of speaker recognition S121AFirst step of the first embodiment of the Speaker recognition S122Asecond step of the first embodiment of speaker recognition S123Athird step ofFirst embodiment of speaker recognition S124Fourth step of the first embodiment of speaker recognition S121First step of the second embodiment of speaker recognition S122BSecond step of the second embodiment of speaker recognition S123BThird step of the second embodiment of speaker recognition

Claims

1. A method (S100) for controlling a teleprompter (10) for the dynamic display of a predefined text on a screen (12) of the teleprompter (10), the method (S100) comprising the following steps: - recording (S110) at least one identity-specific input of a speaker by using a recording device, and generating a corresponding electrical signal, - recognizing (S120) the speaker by evaluating the generated electrical signal, - determining (S130) at least one display parameter depending on the recognized speaker, - dynamically displaying (S140) the text on the screen (12) of the teleprompter (10) by using the at least one display parameter.

2. The method (S100) according to claim 1, characterized in that - recording (S110) the at least one identity-specific input of the speaker includes recording a voice input of the speaker; - the receiving device comprises a microphone (18); and - the electrical signal comprises an audio signal (22).

3. The method (S100) according to any one of claims 1 or 2, characterized in that - recording (S110) the identity-specific input of the speaker includes recording an image recording of the speaker; - the receiving device comprises a camera; and - the electrical signal comprises an image signal.

4. The method (S100) according to any one of the preceding claims, characterized in that at least one display parameter determined depending on the recognized speaker is the display speed of the dynamically displayed text on the screen (12).

5. The method (S100) according to any one of the preceding claims, characterized in that at least one display parameter determined depending on the recognized speaker relates to at least one font color, a background color or a font type of the displayed text or a screen configuration parameter.

6. The method (S100) according to any one of claims 2 to 5, characterized in that the speaker recognition is based on a detection of the speaker fundamental frequency.

7. The method (S100) according to any one of claims 2 to 6, characterized in that the speaker recognition is based on a spectral analysis of the audio signal (22).

8. The method (S100) according to any one of claims 2 to 7, characterized in that the speaker recognition is based on a cepstral analysis of the audio signal (22) and a detection of the speaker fundamental frequency in the cepstral domain.

9. The method (S100) according to any one of claims 2 to 8, characterized in that the speaker recognition is based on determining the speaker fundamental frequency in the spectral range or in the cepstral range, wherein only audio signals (22) to be associated with a voiced sound are evaluated for the speaker recognition, wherein preferably a voiced / unvoiced analysis is used for determining the audio signals (22) to be associated with a voiced sound, wherein the first cepstral coefficient of the audio signal (22) is evaluated for the voiced / unvoiced analysis of the audio signal (22), and wherein a voiced sound is inferred if the first cepstral coefficient is greater than zero.

10. The method (S100) according to any one of claims 2 to 9, characterized in that the method (S100) comprises a speech pause detection and a repeated speaker recognition, wherein the interval between two consecutive speaker recognition processes is dynamic and depends on the speech pause detected.

11. The method (S100) according to any one of the preceding claims, characterized in that the voice input of the speaker is recorded by using a microphone array (16), and that a beamforming method is used to increase the accuracy of the speaker recognition.

12. The method (S100) according to any one of claims 3 to 11, characterized in that the speaker recognition is based on a face recognition.

13. The method (S100) according to any one of claims 3 to 12, characterized in that the method for recognizing the speaker comprises the extraction of several features from the image signal and the subsequent evaluation of the features.

14. The method (S100) according to claim 13, characterized in that the extracted features or feature combinations are compared to features or feature combinations stored in a database (26), wherein features or feature combination and speakers associated with the features or the feature combinations are stored in the database (26).

15. A teleprompter (10) for the dynamic display of a predefined text, comprising - a screen (12) for displaying the text, - a recording device (14) for recording an identity-specific input, - an evaluation unit for converting the identity-specific input into an electrical signal for evaluating the electrical signal and for recognizing the speaker, and - a control unit for generating a control signal for controlling the screen (12) for the dynamic display of the text by using the at least one display parameter and depending on the recognized speaker.