Studio AI voice prompter method and system

Abstract

The invention provides a studio AI voice prompter method and system, and relates to the technical field of studio control and intelligent prompting. According to the method and the device, the playing content of the broadcast list is obtained, the prompt manuscript is generated, the recognition text fragment output by real-time voice recognition is matched with the manuscript content, and whether the scrolling is started or not is judged; when scrolling is allowed, the pixel scrolling speed is calculated according to the speech speed statistical result, the page position is updated, and synchronization of the scrolling progress and the oral broadcast progress is achieved; and automatically resetting the matching reference and stopping scrolling when the manuscript is updated or broadcast reading is finished, and supporting manual start and stop. According to the invention, the scrolling accuracy and stability of the prompter and the broadcast control efficiency of the studio are improved.

Description

Technical Field

[0001] This invention relates to the field of studio control and intelligent teleprompter technology, and in particular to a studio AI voice teleprompter method and system. Background Technology

[0002] With the increasing informatization of studios, the application of digital studios is becoming more and more widespread. The use of teleprompters can be traced back to the traditional news and program production process. Traditionally, hosts or actors needed to read their lines from paper manuscripts held in their hands or placed on a table. However, this method has limitations and inconveniences, such as the need for frequent page turning, which can lead to omissions or misreadings. Therefore, there was a desire for breakthroughs in teleprompter control to reduce the burden on announcers and improve broadcast safety. This led to the development of studio artificial intelligence (AI) voice teleprompters. These teleprompters utilize AI models and audio processing technology to calculate the scrolling speed in real time based on the host's or actor's speaking speed and the content being spoken, matching the teleprompter text to accurately track and read the lines. This reduces some of the limitations and inconveniences associated with traditional paper manuscripts.

[0003] While studio AI teleprompters offer many advantages in improving efficiency and accuracy, they also present some potential drawbacks and challenges. For example: (1) AI language recognition has certain environmental requirements. The environment needs to be free of obvious noise so that the speech can be clearly transmitted to the teleprompter service for content and speech rate judgment. (2) The teleprompter cannot roll back according to the voice and cannot be controlled simultaneously with the hand and foot pedals. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a studio AI voice teleprompter method and system. By driving the scrolling screen through real-time speech recognition and content matching, the scrolling progress of the teleprompter is synchronized with the host's speaking progress, thereby effectively improving the accuracy and stability of the studio broadcast control process.

[0005] To achieve the above objectives, the present invention provides the following solution: A studio AI teleprompter method includes: The spoken content is obtained from the broadcast schedule of the broadcast control system, and the spoken content is spliced ​​into a teleprompter script for display on the teleprompter page. Algorithm parameters such as matching threshold, speed coefficient, speech rate statistics time window and terminal termination delay, as well as font size, line spacing and number of characters per line are loaded. When the broadcast control is turned on, the audio from the host's microphone is received, the audio is recognized in real time, the recognized text segments are output according to time, and the most recently recognized text segment is used to form the read text window. Within the text prompt display area, multiple consecutive texts of the same length as the already read text window are truncated according to a preset truncation step size to form a candidate text window set. The already read text window is compared with each candidate text window in the candidate text window set to obtain the corresponding matching degree. The maximum matching degree is taken as the matching degree for this matching calculation. The matching degree is compared with the matching threshold. When the matching degree is not lower than the matching threshold, scrolling is allowed; otherwise, the page remains still. When scrolling is allowed, the effective number of characters in the recognized text segment is accumulated within the speech rate statistics time window to obtain the total number of characters and calculate the spoken character speed. The page line height and the average number of characters that can be displayed per line are determined based on the font size, the line spacing and the number of characters per line. The pixel scrolling speed is calculated based on the spoken character speed, the page line height, and the average number of characters. The vertical offset is calculated based on the pixel scrolling speed and the time interval between the scrolling calculation cycle. The vertical scrolling position of the page is updated according to the vertical offset to synchronize the scrolling progress with the spoken progress. When the broadcast control system updates the spoken content, it reassembles the prompt text according to the current font size, line spacing and number of characters per line and refreshes the display, resets the content matching reference, and stops scrolling when the current news content is finished being read and the matching degree is continuously lower than the matching threshold within the terminal termination delay, allowing manual start and stop operations.

[0006] Preferably, the matching threshold, the speech rate statistics time window, the speed coefficient, and the terminal termination delay constitute algorithm parameters, wherein the matching threshold is used to control the minimum similarity requirement between the recognized text segment and the text in the prompting text, and the value ranges from 0% to 100%; the speech rate statistics time window is used to determine the statistical duration when calculating the speed of spoken characters; the speed coefficient is used to adjust the smoothness and speed of scrolling; the terminal termination delay is used to determine when to automatically stop scrolling when no valid spoken input is detected; the font size, the line spacing, and the number of characters per line constitute performance parameters, which are loaded separately at the user level to allow different users to personalize the configuration according to their reading habits and display devices.

[0007] Preferably, calculating the matching degree between the read-through text window and the candidate text window includes: The identified text fragments within the already read text window are used as reference texts, and the text with the same length as the already read text window is used as the candidate text window within the current display area of ​​the prompting text. The similarity score is obtained by comprehensively calculating the similarity between the reference text and the candidate text window based on character overlap, edit distance and continuous segment matching degree. The similarity score is normalized to a range of 0% to 100% to obtain the matching degree.

[0008] Preferably, real-time speech recognition of the audio includes: The received audio is divided into audio segments; Each audio segment is stream-decoded, and a recognition text segment is output after each audio segment is decoded. A timestamp is added to each recognition text segment. The timestamp is used to align the recognition text segment with the actual timeline of the broadcast control process in time, so as to improve the accuracy of content matching and support the recording and playback of the broadcast control process.

[0009] Preferably, receiving audio from the host's microphone includes: Connect the microphone to the host's lavalier microphone, headset microphone, or microphone output from the mixing console via the sound card interface; The electrical signal output from the lavalier microphone, the headset microphone, or the microphone output from the mixing console is used as the host's microphone audio. The audio is digitally acquired in a mono format with a sampling rate of 16,000 times per second and a quantization bit depth of 16 bits to obtain the digital acquisition result. The digitized acquisition results are encoded using signed linear pulse code modulation to meet the accuracy and latency requirements of real-time speech recognition.

[0010] Preferably, at the end of each speech rate statistics time window, the effective character widths corresponding to the recognized text segments within the speech rate statistics time window are summed to obtain the total character width within the speech rate statistics time window. The ratio of the total character width to the average character width that can be displayed per line is taken as the equivalent number of lines to be read within the speech rate statistics time window. The pixel scrolling speed is recalculated based on the equivalent number of lines to be read, the page line height, and the speed coefficient. The updated pixel scrolling speed is multiplied by the time interval of the scrolling calculation cycle to obtain the vertical offset within the current scrolling calculation cycle.

[0011] Preferably, the broadcast control system updates the spoken content by: After receiving the update data sent by the broadcast control system, the update type is determined according to the program entry number and the corresponding text content. If the program entry number increases or a new program entry identifier appears, it is determined to be adding new news. If the program entry number remains unchanged but the text content changes, it is determined to be replacing the current news text. After generating the updated teleprompter text and refreshing the teleprompter page, the content matching reference is reset to the read text window and the candidate text window reconstructed based on the current display area of ​​the updated teleprompter text. The matching degree is then recalculated based on the reconstructed read text window and the candidate text window to ensure that the scrolling process driven by speech recognition is continuous and without skipping, misalignment, or delay when switching texts.

[0012] Preferably, when the broadcast control system updates the verbally broadcast content, it further includes: While generating the updated teleprompter script, the updated teleprompter script content, along with the program title, program entry identifier, program entry number, and update time information, are stored in the data storage space for subsequent broadcast control records and version management. During the broadcast control process, the prompt script can be manually edited through the user interface. After the manual editing is completed, the modified prompt script will be synchronously updated to the corresponding record in the data storage space to ensure that the prompt script display is consistent with the broadcast control record.

[0013] Preferably, it further includes: The text range covered by the candidate text window is determined based on the window spacing coefficient. When the window spacing coefficient increases, the text range covered by the candidate text window increases, thereby increasing the tolerance of the matching degree to the playback position offset. When the window spacing coefficient decreases, the text range covered by the candidate text window decreases, thereby increasing the sensitivity of the matching degree to the playback position offset. During the operation of the teleprompter, the manual start and stop operation includes manually triggering a scrolling start command when the matching degree has not reached the matching threshold and manually triggering a scrolling stop command when the matching degree is not lower than the matching threshold, so that the scrolling behavior can be temporarily taken over by a human in special broadcast control scenarios.

[0014] A studio AI voice teleprompter system includes: The text access device is used to obtain spoken content from the broadcast control system's broadcast schedule, splice the spoken content into a teleprompter text, display it on the teleprompter page, and load algorithm parameters such as matching threshold, speech rate statistics time window, and terminal termination delay, as well as font size, line spacing, and number of characters per line. The speech recognition device is used to receive the host's microphone audio when the broadcast control is turned on, perform real-time speech recognition on the audio, output the recognized text segments according to time, and form the read text window with the most recent recognized text segment. The content matching device is used to extract multiple consecutive texts of the same length as the already read text window within the text display area according to a preset extraction step size, and to form a candidate text window set. The already read text window is compared with each candidate text window in the candidate text window set to obtain the corresponding matching degree. The maximum matching degree is taken as the matching degree calculated in this matching. The matching degree is compared with the matching threshold. When the matching degree is not lower than the matching threshold, scrolling is allowed; otherwise, the page remains still. A character counting device is used to accumulate the effective number of characters in the recognized text segment within the speech rate counting time window when scrolling is allowed, to obtain the total number of characters and calculate the spoken character speed, and to determine the page line height and the average number of characters that can be displayed per line based on the font size, the line spacing and the number of characters per line. The scrolling control device is used to calculate the pixel scrolling speed based on the speed of the spoken characters, the page line height and the average number of characters, calculate the vertical offset based on the time interval between the pixel scrolling speed and the scrolling calculation cycle, and update the vertical scrolling position of the page according to the vertical offset, so that the scrolling progress is synchronized with the spoken progress. The text update and rollback device is used to reassemble the prompt text according to the current font size, line spacing and number of characters per line and refresh the display when the broadcast control system updates the oral content, reset the content matching reference, stop scrolling when the current news content is finished being read and the matching degree is continuously lower than the matching threshold within the terminal termination delay, and allow manual start and stop operations.

[0015] The present invention discloses the following technical effects: This invention achieves real-time perception of the host's speaking progress by converting the host's microphone audio into recognized text fragments in real time, using the most recently recognized text fragment to form a read-already text window, and then matching it with the candidate text window in the currently displayed area of ​​the prompt script. Through a dynamic comparison mechanism of matching degree and matching threshold, it can instantly determine whether the host is reading the current area of ​​content, thereby triggering or preventing scrolling. Unlike the "scrolling ahead" or "scrolling behind" problems caused by the fixed-speed scrolling method commonly used in background technologies, this invention synchronizes the scrolling process with the speaking process through content-based real-time calibration, significantly improving the accuracy and controllability of the broadcast control process.

[0016] This invention accumulates the effective character widths corresponding to the identified text segments within a speech rate statistics time window to obtain the total character width. It then combines this with the page line height and the average displayable character width per line to calculate the equivalent number of lines to be read aloud. Based on this, the pixel scrolling speed is calculated, forming a scrolling drive quantity directly corresponding to the spoken speed. This method fundamentally overcomes the jitter, abrupt changes, and instability problems caused by traditional systems that rely on fixed character lengths, fixed durations, or empirical speed models for scrolling speed estimation. This invention achieves a continuous quantization chain from character width to line count to pixel speed, making the scrolling effect smoother, more stable, and naturally integrated with the spoken rhythm.

[0017] This invention, when updating spoken content in the broadcast control system, reassembles the teleprompter text according to the current layout parameters and refreshes the display. Simultaneously, it resets the content matching reference, allowing the new read text window and the candidate text window of the new text to re-establish a matching relationship, thus ensuring the correct scrolling start position after content switching. Traditional teleprompter systems, failing to rebuild the matching basis during content updates, are prone to content skipping or misreading. This invention, through a complete text reconstruction and re-matching mechanism, ensures the correct restoration of the scrolling reference point, enabling smooth transitions when switching news items, columns, or making temporary changes.

[0018] This invention achieves content-level playback position recognition by comparing candidate text windows with already played text windows of the same length, combined with matching degree calculation. It also allows for configuring window spacing coefficients to adjust the coverage of the matching window. When the host skips, misses, temporarily adds text, or changes in speaking speed, this mechanism automatically absorbs playback offsets, enabling the system to tolerate minor playback errors without causing scrolling errors. This overcomes the problem of scrolling judgment failure caused by speech recognition errors in the background technology. Overall, it improves the system's fault tolerance and playback control stability.

[0019] This invention not only automatically stops scrolling when the matching degree between the identified text fragment and the teleprompter content falls below a threshold, but also incorporates mechanisms such as terminal termination delay, manual start / stop operations, and default position restoration. This allows for manual intervention even during automated operation, meeting the practical needs of various broadcast control scenarios, including live streaming, recorded broadcasts, and sudden interruptions. Compared to existing devices that can only fix scrolling or rely on manual button control, this invention offers flexible switching between "automatic," "semi-automatic," and "manual" control modes, greatly enhancing the system's practical value and the studio user experience. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the operation of the device provided in this embodiment of the invention; Figure 3 This is a schematic diagram illustrating the method of using the device provided in an embodiment of the present invention.

[0022] Explanation of reference numerals in the attached figures: 1. Third-party automation system; 2. Document access module; 3. Document display module; 4. Document saving module; 5. Data server; 6. Speech recognition module; 7. Content matching module; 8. Scrolling control module. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The purpose of this invention is to provide a studio AI voice teleprompter method and system, which adjusts the scrolling based on the dynamic matching results of the identified text fragments and the content of the teleprompter script, so that the teleprompter's scrolling behavior can adapt to the actual reading rhythm of the host, significantly improving the smoothness of broadcast control and the reliability of reading.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, the present invention provides a studio AI voice teleprompter method, including: Step 100: Obtain the spoken content from the broadcast control system's broadcast schedule, splice the spoken content into a teleprompter script, display it on the teleprompter page, and load the matching threshold, speed coefficient, speech rate statistics time window, and terminal termination delay algorithm parameters, as well as font size, line spacing, and number of characters per line. Step 200: When the broadcast control is turned on, receive the host's microphone audio, perform real-time speech recognition on the audio, output the recognized text segments according to time, and form the read text window with the most recent recognized text segment. Step 300: Within the text prompt display area, extract multiple consecutive texts of the same length as the already read text window according to the preset extraction step size as a candidate text window set. Compare the already read text window with each candidate text window in the candidate text window set to obtain the corresponding matching degree, and take the maximum matching degree as the matching degree calculated in this matching. Step 400: When scrolling is allowed, accumulate the effective number of characters in the recognized text segment within the speech rate statistics time window to obtain the total number of characters and calculate the spoken character speed. Determine the page line height and the average number of characters that can be displayed per line based on the font size, line spacing and number of characters per line. Step 500: Calculate the pixel scrolling speed based on the spoken character speed, page line height, and average number of characters. Calculate the vertical offset based on the pixel scrolling speed and the time interval between the scrolling calculation cycle. Update the vertical scrolling position of the page according to the vertical offset to synchronize the scrolling progress with the spoken progress. Step 600: When the broadcast control system updates the spoken content, reassemble the prompt text according to the current font size, line spacing and number of characters per line and refresh the display, reset the content matching reference, and stop scrolling when the current news content is finished being read and the matching degree continues to be lower than the matching threshold within the terminal termination delay, allowing manual start and stop operations.

[0027] In step 300 of this embodiment, the read-already text window represents the input text to be matched, which consists of the most recently identified text segment, and the current display area of ​​the teleprompter script represents all the text content visible on the current teleprompter page. To determine the position of the host's currently read content within the current display area of ​​the teleprompter script, this embodiment does not simply compare the read-already text window with a single candidate text, but instead sequentially extracts multiple consecutive text segments of the same length as the read-already text window from the current display area of ​​the teleprompter script, using each segment as a candidate text window.

[0028] In the specific calculation process, this embodiment uses the already read text window as the reference text, and sequentially calculates the matching degree between it and each of the candidate text windows, obtaining a corresponding similarity score for each candidate text window. Subsequently, all similarity scores obtained in the current display area of ​​the prompting text are compared, and the maximum similarity score is selected as the matching degree for this round of matching calculation, used to determine whether the host is reading the corresponding content in the current display area of ​​the prompting text.

[0029] By matching position by position and taking the maximum matching degree as described above, this embodiment can still accurately identify the current reading position even when the host has normal speech rate fluctuations, slight skipping, or recognition errors, thereby improving the robustness of the matching results and providing a stable and reliable input basis for subsequent scrolling trigger judgment and scrolling speed calculation.

[0030] This embodiment enables the teleprompter to automatically scroll while the host is reading, and to remain stationary when no valid speech is detected, thus preventing accidental scrolling. The technical approach of this embodiment includes the following steps: ① Obtain the spoken content from the broadcast control system's playlist via HTTP; ② The spoken content is concatenated in HTML format and displayed on the teleprompter page; ③ After enabling broadcast control, the real-time speech recognition module is activated to continuously output the spoken text; ④ Match the recognized text with the text currently displayed on the page. When the matching degree reaches the threshold, start scrolling and adjust the scrolling offset according to the speech rate; when the matching degree is lower than the threshold, keep the current screen still. ⑤ When the broadcast control system updates the spoken content during the broadcast control process, the teleprompter refreshes the page display in real time after receiving the update; ⑥ During the operation of the teleprompter, manual start and stop operations are permitted; ⑦ When the current news content has been read and subsequent content cannot be effectively matched with the spoken content, the scrolling will stop naturally and the process will end.

[0031] The automatic scrolling algorithm in step ④ is as follows: I. Within a fixed statistical period T (i.e., the speech rate statistical time window), the effective number of characters output by the speech recognition module is accumulated to obtain the total number of characters charCount within the statistical period, and the speaking character rate charRate of the host per unit time is calculated, where: charRate=charCount / T.

[0032] II. Calculate the page line height and the average number of characters that can be displayed per line based on the current document's layout parameters, where: The line height (lineHeight) is calculated using the following formula: ; Where fontSize is the font size of the current document, factor is the line height multiple (1.25 by default when not set), and ⌊·⌋ indicates rounding down; The total pixel height of the current document after rendering is totalPixels. The total number of lines is calculated by converting the line height: lines=totalPixels / lineHeight; After removing tags and special characters from the document content, the total number of valid characters (totalSize) is obtained. Therefore, the average number of characters that can be displayed per line (fontCount) is: fontCount = totalSize / lines; III. Calculate the pixel scrolling speed (i.e., pixel scrolling speed) per second based on the spoken character speed (charRate), line height (lineHeight), and the average number of characters that can be displayed per line (fontCount). speed=(charRate×lineHeight) / fontCount IV. Based on the time interval Δt between the pixel speed (speed) and the scrolling calculation cycle, calculate the vertical offset (Offset) required for this scroll, where: Offset = speed × Δt; The system obtains the current vertical scroll position of the page and calls the ScrollTo function to continuously scroll according to the calculated Offset, ensuring that the scrolling behavior is synchronized with the host's speaking speed. In this exemplary embodiment, the ScrollTo function is used to perform vertical page scrolling. Essentially, it's a page position update instruction. After the system calculates the vertical offset Offset corresponding to the current refresh cycle, the ScrollTo function moves the teleprompter page content to the new display position based on that Offset. The ScrollTo function is not limited to a specific programming language or library interface; its function is to receive the vertical offset and set the page's scroll position to the specified coordinates, ensuring a smooth visual effect for the scrolling action within consecutive refresh cycles. In other words, in this embodiment, ScrollTo is an instruction call to control the page's presentation position. Its execution result is that the teleprompter text scrolls by the number of pixels corresponding to the Offset, thus keeping the teleprompter's scrolling progress consistent with the host's speaking progress.

[0033] This invention utilizes real-time speech recognition and content matching to enable the teleprompter to automatically initiate scrolling based on the content being read by the presenter. By adjusting the scrolling speed, the progress of the scrolling is synchronized with the actual reading, significantly improving the teleprompter's real-time tracking ability of the spoken content. This mechanism effectively reduces manual intervention, achieves automatic scrolling control based on real-time spoken content, and improves the system's efficiency in broadcasting scenarios.

[0034] This invention utilizes a matching degree threshold to control the scrolling logic. When the matching degree between the identified text and the screen content does not reach the threshold, the screen remains still, preventing accidental scrolling caused by the host interrupting, pausing, chatting, or interrupting the broadcast. This method eliminates the need for frequent manual operation to complete the scrolling process, reducing the operational burden on studio staff and significantly improving the stability and accuracy of the scrolling process.

[0035] This invention adjusts the scrolling offset (i.e., vertical offset) based on the output rhythm of the real-time recognized text, allowing the scrolling speed to adapt to changes in the presenter's speaking speed, thus ensuring consistency between the teleprompter content and the spoken rhythm. This method improves the presenter's reading comfort and enhances the fluency of the reading, better meeting the practical needs of frequent changes in speaking speed in a studio environment.

[0036] This invention allows users to customize AI algorithm parameters such as matching threshold, speed calculation delay (represented by a speech rate statistics time window), speed coefficient, window spacing coefficient, and terminal termination delay, enabling the system to adapt to different presenter speech rates, program broadcast control rhythms, and studio environments. The aforementioned algorithm parameters are loaded uniformly at the system level to ensure the stability of the core logic. Simultaneously, the presentation parameters of this invention (including font size, line spacing, and number of characters per line) can be loaded independently at the user level to meet the differentiated needs of different reading habits, display devices, and program types. By combining the global consistency of algorithm parameters with the user-differentiated configuration of presentation parameters, this invention enhances the teleprompter's adaptability and practical controllability across various studio scenarios, and achieves high execution efficiency based on real-time speech recognition, meeting the real-time requirements of the studio.

[0037] See Figure 2 The device in this embodiment includes: a third-party automation system 1, a document access module 2, a document display module 3, a document saving module 4, a data server 5, an automatic speech recognition (ASR) module 6, a content matching module 7, and a scrolling control module 8.

[0038] The third-party automation system 1 consists of automation systems from different manufacturers. The document access module 2 uses an independent interface. Each time the interface is called, it reads the spoken content from the playlist in the automation system. The document display module 3 typically uses two LCD screens with a resolution of 1920×1080. One of the LCD screens is connected to the teleprompter service host via HDMI, and then displayed via an extension. After the teleprompter service starts, it displays the content on the main screen and also projects it onto the extended screen. The spoken content is spliced ​​in HTML format and displayed on the page using pre-set parameters such as line spacing, font size, and font color. The document saving module 4 is responsible for storing the accessed spoken text and document information into the data server 5 for subsequent querying, playback records, and multi-system synchronization. The data server 5 is used to store all document versions and playback control records, providing a persistent data source for the teleprompter system. The speech recognition module 6 continuously receives the audio binary stream from the microphone and parses the audio stream into text segments in real time. The content matching module 7 matches the text fragments output by the ASR with the text currently displayed in the document display module 3, calculates the matching degree, and compares it with a threshold. The scrolling control module 8 determines whether to scroll based on the output of the content matching module and calculates the scrolling speed in real time based on the speech rate output by the speech recognition module 6.

[0039] As an example, the audio acquisition and data interaction method of the speech recognition module 6 in this embodiment further includes the following: 1) Specific methods and audio parameters for audio acquisition In this embodiment, the speech recognition module 6 receives the microphone audio signal through the local sound card interface of the studio teleprompter service host. The microphone can be a lavalier microphone or a headset microphone worn by the host, or a microphone signal connected to the mixing console, which is sent to the line input of the teleprompter host via the mixing console.

[0040] Voice acquisition uses a unified voice format for digital processing, with the following parameters being preferred: I. Sampling rate is 16kHz; II. The quantization depth is 16 bits; III. The number of audio channels is mono; IV. Signed linear PCM encoding is used, and the data is stored in little-endian byte order.

[0041] The system continuously reads audio data from the microphone according to a fixed-size audio buffer and divides it into audio blocks of approximately 50–80 ms in length for transmission. In this embodiment, the duration of each audio block is approximately 60 ms to balance real-time performance and recognition stability.

[0042] 2) Communication methods and data windows of the speech recognition interface In this embodiment, the speech recognition module and the backend recognition service communicate via WebSocket streaming. After establishing a WebSocket connection, the system sends a fixed-format initialization control message to negotiate the operating parameters of the recognition service. This control message includes the following fields: I. Recognition Mode (fixed to two-pass decoding mode) First, a first-pass decoding with high real-time performance is performed, and then a second-pass decoding is used to enhance the recognition results when necessary, in order to obtain higher text quality.

[0043] II. Audio segment size configuration (fixed to "5,10,5") The three sections have the following meanings: 1. Initial fragment size: 5; 2. Mid-stage stable phase fragment size: 10; 3. Segment size for the final stage: 5.

[0044] The recognition service uses this three-segment structure to determine the decoding granularity of the audio stream during the startup, stabilization, and termination phases.

[0045] III. Fragmentation transmission interval (fixed at 10ms) This interval is used to control the time step of audio blocks sent to the recognition service, so that the effective duration of each audio block meets the requirements of stable streaming recognition.

[0046] IV. Audio format identifier field (fixed to "pcm"); The recognition service decodes the audio stream in linear PCM format accordingly.

[0047] After initialization, the system continuously sends the real-time collected audio data to the recognition service according to the preset segment size. Each data block corresponds to a speech duration of about tens of milliseconds. After each data block is sent, there is a short wait according to the segment interval, so that the overall audio transmission remains basically consistent with the actual speaking rhythm, thereby achieving near real-time streaming recognition.

[0048] When the system detects that the host has finished speaking or receives a stop command, the teleprompter sends an end flag to the recognition service, notifying the service to complete the decoding process of the current session.

[0049] 3) Timestamps and Granularity of ASR Text Output The recognition service returns the recognition result to the speech recognition module 6 in the form of structured data. The returned content includes at least the recognized text fragment at the current moment, and may carry timestamp information and an identifier indicating whether it is the final recognition result.

[0050] Under the two-pass decoding mechanism employed in this embodiment, the recognition service continuously outputs incremental recognition results throughout the session. Specifically, after each audio segment is decoded, the recognition service promptly returns the newly added text content for the current stage. The system extracts the newly added audio segments by comparing the returned text with the previously received text, and uses this information to drive the content matching module and scrolling control module to update in real time. This incremental output strategy enables the system to continuously obtain the host's spoken content within tens of milliseconds, achieving highly real-time scrolling control.

[0051] When the recognition service determines that all audio data for the current session has been received and decoded, it will output a complete final recognition result, along with a clear final identifier. The system uses this identifier to determine whether to continue accumulating text or use the final recognition result as the session output for subsequent processing.

[0052] When the recognition result contains timestamp information, the system can use this time information to align the recognized text with the actual timeline of the broadcast control process, thereby improving the matching accuracy and supporting subsequent analysis or recording playback functions.

[0053] Optionally, the method of using the above-described device in this embodiment is as follows: Figure 3 As shown, the specific steps include: ①The spoken content in the broadcast schedule of the third-party automated system is obtained via HTTP and processed by the text access module 2; ② The acquired spoken content is concatenated and displayed in HTML format and rendered in the document display module 3; ③Activate the speech recognition module (ASR) 6 to receive the microphone audio binary stream in real time and parse it into text segments; ④ Match the identified text fragments with the text content in the current display area, calculate the matching degree and compare it with the threshold; (1) When the matching degree is greater than or equal to the threshold, it is considered that the host is reading the content of the current area; (2) When the matching degree is less than the threshold, it is considered that the current state is not in the reading state, and the teleprompter does not perform scrolling.

[0054] As an example, the calculation of matching degree can be based on the following basic idea: Use the most recently identified text segment as the "read-through text window"; Extract text of equal (or similar) length from the currently displayed area of ​​the document and use it as a "candidate text window"; Similarity scores are obtained by methods such as character overlap, edit distance, and degree of matching of consecutive segments. The final matching degree is expressed in the range of 0% to 100%.

[0055] As an example, in this embodiment, when calculating the similarity between the identified text segment and the candidate text window of the teleprompter, the system first obtains the reference text in the read-out text window and the content of the candidate text window with the same length as the window, and records the two text segments as the reference text sequence and the candidate text sequence, respectively. This embodiment calculates three indicators sequentially: First, character overlap, which measures the degree of overlap between the two text segments at the character level by counting the occurrences of the same characters in the two text segments and normalizing them according to the text length; Second, edit distance, which is obtained by calculating the minimum number of operations (insertion, deletion, and replacement) performed on the reference text sequence and the candidate text sequence, and normalizing this distance according to the text length, to characterize the degree of difference between the two text segments in the overall character sequence; Third, continuous segment matching degree, which is obtained by searching for the longest continuous identical substring in the reference text sequence and the candidate text sequence, recording the length of the substring, and normalizing it according to the total text length, to reflect the consistency between the two text segments in continuous playback segments. This embodiment calculates the above three indicators in parallel within each matching calculation cycle, providing basic data for subsequent comprehensive similarity calculation.

[0056] The matching algorithm in this embodiment is used to establish a stable alignment relationship between the real-time speech recognition text stream and the reference text, and outputs a matching degree result to control the teleprompter's scrolling or pausing. This matching degree result serves as the core judgment basis for the scrolling control logic, determining whether the current speech recognition input corresponds to the content currently displayed in the reference text area, thereby triggering scrolling forward or paused. During algorithm operation, it continuously receives real-time output text segments from the speech recognition module and outputs a matching degree score in each round of calculation. The matching degree score is compared with a preset threshold to drive subsequent scrolling control and alignment correction processes.

[0057] In this embodiment, the system continuously retrieves text segments returned in real time from the speech recognition module. These text segments are word, phrase, or short sentence-level results output by the speech recognition model in real-time decoding mode, and are entered into the speech recognition input queue in the order of arrival. The matching processing thread sequentially retrieves text segments from the head of the input queue, performs character preprocessing on the text segments, including removing punctuation marks, non-alphanumeric characters, and non-Chinese characters, and concatenates multiple text segments into a continuous input text to be matched according to a predetermined window length rule. Simultaneously, the system maintains a segment source list to record which speech recognition segments are sequentially concatenated to form the input text to be matched, supporting subsequent input window forwarding and deletion of the oldest segment.

[0058] To uniformly handle mixed Chinese and English reading scenarios and reduce the impact of homophones, this embodiment converts both the input text to be matched and the text in the currently displayed area of ​​the reference text into Pinyin form before performing matching calculations. This results in an input Pinyin string and a reference Pinyin string, thus mapping different character systems to the same comparable representation. Using the length of the input Pinyin string as a baseline length, the system slides and extracts continuous substrings of the same length from the start to the end of the reference Pinyin string. Each extracted substring serves as a candidate text window for similarity calculation with the input Pinyin string.

[0059] In this embodiment, similarity calculation adopts an edit distance-based approach. For each candidate text window, the system calculates the minimum number of edit operations required to convert the input pinyin string into the candidate substring. These edit operations include insertion, deletion, and replacement. The number of edit operations is normalized according to the string length to obtain a percentage-based similarity score, where a smaller number of edit operations indicates greater similarity between the two text segments, resulting in a higher similarity score. After completing the similarity calculation for all sliding positions, the system obtains a set of similarity scores and takes the highest similarity score as the final matching degree for this round of matching calculation. This score represents the degree of matching of the input text to be matched to the most likely corresponding position in the currently displayed area of ​​the reference document.

[0060] After obtaining the final matching degree, this embodiment compares it with a preset matching threshold. When the matching degree reaches or exceeds the matching threshold, the system determines that the current input content is successfully aligned with the reference text and triggers the teleprompter to scroll forward. At the same time, the input text to be matched is trimmed by the corresponding length from the beginning to move the input window forward. When the matching degree is lower than the matching threshold, the system determines that the matching has failed, triggers a pause in scrolling, and enters the alignment correction logic. The alignment correction logic includes, but is not limited to, discarding the earliest speech recognition segment in the input window, adjusting the input starting position, and recording the number of failures so that a stable alignment relationship can be re-attempted when subsequent speech recognition inputs arrive.

[0061] In this embodiment, when the length of the input pinyin string to be matched exceeds the length of the pinyin string in the reference window, making it impossible to complete the sliding truncation, this embodiment can directly determine the matching degree of this round as low and enter the failure handling branch, or adopt a degradation strategy to perform partial matching of the input content to ensure that the algorithm can continue to execute under boundary conditions. The entire matching and scrolling control process runs continuously in a loop. In each loop, input window construction, similarity calculation, threshold judgment, and scrolling control operations are performed, and a state machine-like control logic is formed by combining the failure count, thereby maintaining the stability and controllability of the teleprompter's scrolling behavior in continuous reading, skipping reading, or recognition fluctuation scenarios.

[0062] As another optional implementation, this embodiment combines the three indicators mentioned above—character overlap, normalized edit distance, and continuous segment matching degree—to obtain a similarity score used to drive the scrolling logic. The calculation formula is as follows:

[0063] In this embodiment, the parameters in the formula are defined as follows: where, The similarity score calculated in this embodiment ranges from 0 to 1. It can be multiplied by 100 to obtain a percentage for comparison with the matching threshold. Character overlap is the normalized overlap ratio obtained by multiplying the number of identical characters in two texts by 2 and then dividing by the sum of the lengths of the reference text and the candidate text. Normalized edit distance is the normalized difference obtained by dividing the minimum number of edit operations between two text segments by the larger of the lengths of the two text segments. The degree of continuous segment matching is the normalized continuous matching ratio obtained by dividing the length of the longest consecutive identical substring in two text segments by the larger of the lengths of the two text segments; when and When both are 0, this embodiment directly sets To avoid a denominator of 0, this embodiment uses the above form of square and weighted summation followed by division. On the one hand, this keeps the calculation process independent of manual adjustment of weight parameters. On the other hand, it gives a higher contribution to the larger value among the three indicators, making the similarity score more sensitive to key matching features.

[0064] In this embodiment, the calculation and usage process of the continuous segment matching degree is as follows: Within each matching calculation cycle, the system searches for the longest consecutive identical character sequence from the reference text sequence and candidate text sequences and records it as the longest continuous matching segment. Let the length of this segment be... The reference text length is The candidate text length is Then there is For example, when the reference text length is 20 characters, the candidate text length is 20 characters, and the longest continuous matching segment length is 10 characters, the result is... If the character overlap is at this time 0.9, Normalized edit distance The similarity score, calculated using the above formula, is approximately... ,correspond The matching degree; this embodiment compares the similarity score with a preset matching threshold. When the matching degree of consecutive segments is high and the edit distance is small, and the character overlap is high, the matching degree is determined. The value will increase significantly, making it easier to trigger the scrolling logic; conversely, the screen will remain still, thereby improving the accuracy of recognizing the actual reading position and enhancing the system's robustness to interruptions, pauses, and minor misreadings.

[0065] In typical broadcasting scenarios, the matching threshold can be set in the range of 60% to 90%.

[0066] This embodiment argues that a higher threshold is not always better, for the following reasons: Threshold too high: Scrolling may not be triggered due to accent, pauses, or minor slips of the tongue; Threshold too low: Scrolling may be accidentally triggered by brief noise or irrelevant sound.

[0067] Therefore, this invention allows for the adjustment of the threshold based on the host's habits, the noise environment of the scene, and the type of program to obtain optimal stability.

[0068] ⑤ Once the matching degree reaches the threshold, a scrolling algorithm based on speech rate is executed, specifically including: I. Accumulate the text output by speech recognition within the configured speech rate statistics time window, and accumulate the effective display width (i.e. effective character count) of the recognized text within each speech rate statistics time window to obtain the total character width of the time window, which is used to estimate the actual reading speed of the host within the time window. II. Calculate the page line height and the average character width per line based on the display parameters of the current prompt text. The line height is obtained by multiplying the font size by a line height coefficient. The total pixel height of the rendered text is divided by the line height to obtain the total number of lines. Then, the total effective character width of the text is divided by the total number of lines to obtain the average character width per line under the current layout settings. In this embodiment, the line height coefficient is used to represent the proportional relationship between the font height and the actual displayed line spacing when the page line height is set, reflecting the visual space occupation characteristics of different fonts at the same font size. The page line height is obtained by multiplying the font size by the line height coefficient. This line height serves as the basis for subsequent calculations of the average number of characters per line and pixel scrolling speed. It is not a limited algorithm parameter but rather an internal calculation variable used for typesetting in this embodiment.

[0069] As an example, in this embodiment, the "total effective character width" used for speech rate statistics is the cumulative display width value corresponding to the identified text segment, used to reflect the actual display space occupied by the text within that time window. Since there is a fixed mapping relationship between character display width and character count under fixed font size and font rendering method, in this embodiment, "effective character count," "total character count (charCount)," and "total effective character width" all represent the same cumulative amount of read content, differing only in their units of measurement; they are different calculation expressions of the same statistical object. To maintain terminology consistency, "total effective character width" corresponds to "total character count."

[0070] In this embodiment, during the actual calculation process, various statistical quantities characterize the scale of the broadcast content within the same time window using different dimensions. "Effective character count" and "total character count (charCount)" are measured in units of characters, with the dimension "characters," describing the number of characters in the identified text segment within the speech rate statistics time window. "Total effective character width" is measured in units of the rendered character display width, with the dimension "pixels (px)," measuring the visual width occupied by the same group of characters when displayed on the page. Since there is a definite mapping relationship between character width and character count under fixed font size and style, this embodiment can use the statistical results of character count and pixel count dimensions equivalently for estimating the broadcast rhythm when calculating the spoken character speed and pixel scrolling speed.

[0071] In actual calculations, after the recognized text segment is rendered, its character width value is converted into the corresponding number of characters or total number of characters for subsequent speech rate statistics; similarly, when the total number of characters (charCount) is obtained, it can be converted into the display width by combining the font size and the average character width, thus obtaining the effective total character width. This is achieved through the conversion relationships between the above units.

[0072] III. The ratio of the total character width within the statistical time window to the average character width of each line is converted to obtain the equivalent number of lines read by the host within the time window. Then, the corresponding scrolling speed (i.e., pixel scrolling speed) is calculated by combining the line height and the preset speed coefficient, so that the scrolling speed changes with the actual speaking speed of the host. IV. The scrolling control module 8 calculates the vertical offset required in the current refresh cycle based on the calculated scrolling speed, and updates the scrolling position accordingly, so that the scrolling progress of the teleprompter content is basically consistent with the reading progress of the host. V. In this embodiment, the speech rate statistics and scrolling speed are updated repeatedly according to the statistical time window period. No additional moving average or exponential smoothing method is used, and no independent minimum or maximum limit value is set for the scrolling speed. The scrolling smoothness and sensitivity are adjusted by the speed coefficient and the time window length.

[0073] ⑥ During the broadcast control process, the third-party automation system can modify the broadcast content at any time. After receiving the updated data, the text access module 2 will reassemble the updated text to generate the corresponding HTML content according to the layout parameters such as font, line spacing, and number of characters per line preset in the user interface, and send it to the text display module 3 for display refresh in real time.

[0074] When an update event occurs, this system determines the update type based on the changes in the program entry number and text content: (1) If the entry number is incremented or a new news entry identifier appears, it is considered as adding a new news item; (2) If the entry number is the same but the text changes, it is considered as replacing the current news article.

[0075] Whether it's adding or replacing content, this system will automatically reset the matching window of the content matching module after the content on the displayed page is updated, so that it can immediately rebuild the matching reference based on the latest displayed content. This ensures that the scrolling process driven by voice recognition remains continuous when switching documents, without any skipping, misalignment, or delayed scrolling.

[0076] Meanwhile, the document saving module 4 will write the updated document content and structured information to the data server 5, including the program title, item identifier, item number, document content, update timestamp and other necessary broadcast control information, for subsequent version management, broadcast control records and system synchronization.

[0077] ⑦ During the broadcast control process, the teleprompter content can also be manually edited, and the changes will be synchronized to the data server 5 by the script saving module 4; ⑧ During the operation of the teleprompter, manual start and stop operations can be performed to adapt to the studio broadcast control process; When the current news content reaches its last item and the identified content can no longer form a valid match, the scrolling control module 8 stops scrolling, and the process ends.

[0078] In addition, this embodiment also provides a studio AI voice teleprompter system, including: The text access device is used to obtain spoken content from the broadcast control system's broadcast schedule, splice the spoken content into a teleprompter text, display it on the teleprompter page, and load algorithm parameters such as matching threshold, speech rate statistics time window, and terminal termination delay, as well as font size, line spacing, and number of characters per line. The speech recognition device is used to receive the host's microphone audio when the broadcast control is turned on, perform real-time speech recognition on the audio, output the recognized text segments according to time, and form the read text window with the most recent recognized text segment. The content matching device is used to extract multiple consecutive texts of the same length as the already read text window within the text display area according to a preset extraction step size, and to form a candidate text window set. The already read text window is compared with each candidate text window in the candidate text window set to obtain the corresponding matching degree. The maximum matching degree is taken as the matching degree calculated in this matching. The matching degree is compared with the matching threshold. When the matching degree is not lower than the matching threshold, scrolling is allowed; otherwise, the page remains still. A character counting device is used to accumulate the effective number of characters in the recognized text segment within the speech rate counting time window when scrolling is allowed, to obtain the total number of characters and calculate the spoken character speed, and to determine the page line height and the average number of characters that can be displayed per line based on the font size, the line spacing and the number of characters per line. The scrolling control device is used to calculate the pixel scrolling speed based on the speed of the spoken characters, the page line height and the average number of characters, calculate the vertical offset based on the time interval between the pixel scrolling speed and the scrolling calculation cycle, and update the vertical scrolling position of the page according to the vertical offset, so that the scrolling progress is synchronized with the spoken progress. The text update and rollback device is used to reassemble the prompt text according to the current font size, line spacing and number of characters per line and refresh the display when the broadcast control system updates the oral content, reset the content matching reference, stop scrolling when the current news content is finished being read and the matching degree is continuously lower than the matching threshold within the terminal termination delay, and allow manual start and stop operations.

[0079] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0080] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A studio AI voice teleprompter method, characterized in that, include: The spoken content is obtained from the broadcast schedule of the broadcast control system, and the spoken content is spliced ​​into a teleprompter script for display on the teleprompter page. Algorithm parameters such as matching threshold, speed coefficient, speech rate statistics time window and terminal termination delay, as well as font size, line spacing and number of characters per line are loaded. When the broadcast control is turned on, the audio from the host's microphone is received, the audio is recognized in real time, the recognized text segments are output according to time, and the most recently recognized text segment is used to form the read text window. Within the text prompt display area, multiple consecutive texts of the same length as the already read text window are truncated according to a preset truncation step size to form a candidate text window set. The already read text window is compared with each candidate text window in the candidate text window set to obtain the corresponding matching degree. The maximum matching degree is taken as the matching degree for this matching calculation. The matching degree is compared with the matching threshold. When the matching degree is not lower than the matching threshold, scrolling is allowed; otherwise, the page remains still. When scrolling is allowed, the effective number of characters in the recognized text segment is accumulated within the speech rate statistics time window to obtain the total number of characters and calculate the spoken character speed. The page line height and the average number of characters that can be displayed per line are determined based on the font size, the line spacing and the number of characters per line. The pixel scrolling speed is calculated based on the spoken character speed, the page line height, and the average number of characters. The vertical offset is calculated based on the pixel scrolling speed and the time interval between the scrolling calculation cycle. The vertical scrolling position of the page is updated according to the vertical offset to synchronize the scrolling progress with the spoken progress. When the broadcast control system updates the spoken content, it reassembles the prompt text according to the current font size, line spacing and number of characters per line and refreshes the display, resets the content matching reference, and stops scrolling when the current news content is finished being read and the matching degree is continuously lower than the matching threshold within the terminal termination delay, allowing manual start and stop operations.

2. The studio AI teleprompter method according to claim 1, characterized in that, The matching threshold, the speech rate statistics time window, the speed coefficient, and the terminal termination delay constitute the algorithm parameters. The matching threshold controls the minimum similarity requirement between the recognized text segment and the text in the prompting document, with a value ranging from 0% to 100%. The speech rate statistics time window determines the statistical duration for calculating the speed of spoken characters. The speed coefficient adjusts the smoothness and speed of scrolling. The terminal termination delay determines when to automatically stop scrolling if no valid spoken input is detected. The font size, line spacing, and number of characters per line constitute the performance parameters, loaded separately at the user level to allow different users to personalize their configurations based on reading habits and display devices.

3. The studio AI teleprompter method according to claim 2, characterized in that, Calculating the matching degree between the read text window and the candidate text window includes: The identified text fragments within the already read text window are used as reference texts, and the text with the same length as the already read text window is used as the candidate text window within the current display area of ​​the prompting text. The similarity score is obtained by comprehensively calculating the similarity between the reference text and the candidate text window based on character overlap, edit distance and continuous segment matching degree. The similarity score is normalized to a range of 0% to 100% to obtain the matching degree.

4. The studio AI teleprompter method according to claim 1, characterized in that, Real-time speech recognition of audio, including: The received audio is divided into audio segments; Each audio segment is stream-decoded, and a recognition text segment is output after each audio segment is decoded. A timestamp is added to each recognition text segment. The timestamp is used to align the recognition text segment with the actual timeline of the broadcast control process in time, so as to improve the accuracy of content matching and support the recording and playback of the broadcast control process.

5. The studio AI teleprompter method according to claim 1, characterized in that, Receive audio from the host's microphone, including: Connect the microphone to the host's lavalier microphone, headset microphone, or microphone output from the mixing console via the sound card interface; The electrical signal output from the lavalier microphone, the headset microphone, or the microphone output from the mixing console is used as the host's microphone audio. The audio is digitally acquired in a mono format with a sampling rate of 16,000 times per second and a quantization bit depth of 16 bits to obtain the digital acquisition result. The digitized acquisition results are encoded using signed linear pulse code modulation to meet the accuracy and latency requirements of real-time speech recognition.

6. The studio AI teleprompter method according to claim 2, characterized in that, At the end of each speech rate statistics time window, the effective character widths corresponding to the recognized text segments within the speech rate statistics time window are summed to obtain the total character width within the speech rate statistics time window. The ratio of the total character width to the average character width that can be displayed per line is taken as the equivalent number of lines to be read within the speech rate statistics time window. The pixel scrolling speed is recalculated based on the equivalent number of lines to be read, the page line height, and the speed coefficient. The updated pixel scrolling speed is multiplied by the time interval of the scrolling calculation cycle to obtain the vertical offset within the current scrolling calculation cycle.

7. The studio AI teleprompter method according to claim 1, characterized in that, The broadcast control system updates the spoken content, including: After receiving the update data sent by the broadcast control system, the update type is determined according to the program entry number and the corresponding text content. If the program entry number increases or a new program entry identifier appears, it is determined to be adding new news. If the program entry number remains unchanged but the text content changes, it is determined to be replacing the current news text. After generating the updated teleprompter text and refreshing the teleprompter page, the content matching reference is reset to the read text window and the candidate text window reconstructed based on the current display area of ​​the updated teleprompter text. The matching degree is then recalculated based on the reconstructed read text window and the candidate text window to ensure that the scrolling process driven by speech recognition is continuous and without skipping, misalignment, or delay when switching texts.

8. The studio AI teleprompter method according to claim 7, characterized in that, When the broadcast control system updates the verbal content, it also includes: While generating the updated teleprompter script, the updated teleprompter script content, along with the program title, program entry identifier, program entry number, and update time information, are stored in the data storage space for subsequent broadcast control records and version management. During the broadcast control process, the prompt script can be manually edited through the user interface. After the manual editing is completed, the modified prompt script will be synchronously updated to the corresponding record in the data storage space to ensure that the prompt script display is consistent with the broadcast control record.

9. The studio AI teleprompter method according to claim 1, characterized in that, Also includes: The text range covered by the candidate text window is determined based on the window spacing coefficient. When the window spacing coefficient increases, the text range covered by the candidate text window increases, thereby increasing the tolerance of the matching degree to the playback position offset. When the window spacing coefficient decreases, the text range covered by the candidate text window decreases, thereby increasing the sensitivity of the matching degree to the playback position offset. During the operation of the teleprompter, the manual start and stop operation includes manually triggering a scrolling start command when the matching degree has not reached the matching threshold and manually triggering a scrolling stop command when the matching degree is not lower than the matching threshold, so that the scrolling behavior can be temporarily taken over by a human in special broadcast control scenarios.

10. A studio AI voice teleprompter system, characterized in that, include: The text access device is used to obtain spoken content from the broadcast control system's broadcast schedule, splice the spoken content into a teleprompter text, display it on the teleprompter page, and load algorithm parameters such as matching threshold, speech rate statistics time window, and terminal termination delay, as well as font size, line spacing, and number of characters per line. The speech recognition device is used to receive the host's microphone audio when the broadcast control is turned on, perform real-time speech recognition on the audio, output the recognized text segments according to time, and form the read text window with the most recent recognized text segment. The content matching device is used to extract multiple consecutive texts of the same length as the already read text window within the text display area according to a preset extraction step size, and to form a candidate text window set. The already read text window is compared with each candidate text window in the candidate text window set to obtain the corresponding matching degree. The maximum matching degree is taken as the matching degree calculated in this matching. The matching degree is compared with the matching threshold. When the matching degree is not lower than the matching threshold, scrolling is allowed; otherwise, the page remains still. A character counting device is used to accumulate the effective number of characters in the recognized text segment within the speech rate counting time window when scrolling is allowed, to obtain the total number of characters and calculate the spoken character speed, and to determine the page line height and the average number of characters that can be displayed per line based on the font size, the line spacing and the number of characters per line. The scrolling control device is used to calculate the pixel scrolling speed based on the speed of the spoken characters, the page line height and the average number of characters, calculate the vertical offset based on the time interval between the pixel scrolling speed and the scrolling calculation cycle, and update the vertical scrolling position of the page according to the vertical offset, so that the scrolling progress is synchronized with the spoken progress. The text update and rollback device is used to reassemble the prompt text according to the current font size, line spacing and number of characters per line and refresh the display when the broadcast control system updates the oral content, reset the content matching reference, stop scrolling when the current news content is finished being read and the matching degree is continuously lower than the matching threshold within the terminal termination delay, and allow manual start and stop operations.

Images