The Broadcast News Transcription System

It creates closed-captioning from the announcer's speech

Akio ANDO, Associate Director,
Human Science

Realization of "human-friendly broadcasting" is important in order to make available broadcasting services for the elderly or hearing impaired. Expanding closed-captioned broadcasts is one of the means to accomplish this. To meet the demands for this news program closed-captioning service, our Laboratories have been developing a broadcast news transcription system which can recognize an announcer's speech and automatically create a caption with which the broadcast may be subtitled.

The broadcast news transcription system recognizes the announcer's speech and creates a caption with natural Japanese sentences using the appropriate Kanji and Kana characters. The system's characteristics are as follows:

  • The system automatically learns frequently used words and expressions in a news manuscript database. This self-learning function is performed by the systems utilization of the topics and proper nouns that appear often in recent news programs.
  • Characteristics of an announcer's voice are automatically studied using an announcer speech database.
The performance of the present system is as follows:
  • Recognition accuracy: Maximum recognition accuracy for studio newscasters' speech is 95%. This rate decreases depending on news topic categories and the average accuracy is 85%.
  • Running time: The time delay between the announcers' actual words and the appearance of the closed captioning on the screen averages about 10 seconds.
Manually reviewing the material is a necessary part of the final process to assure the accuracy of the caption. We have confirmed that a mistake that happens every 6 seconds can be recognized and corrected on live broadcasts. Therefore, this means that a recognition accuracy averaging 95% has to be realized.
Moreover, a time delay of no more than 5 seconds between the announcer and the closed-captioning is desirable. The reviewing and correction process takes about 3 seconds, which leaves a delay of less than 2 seconds available for recognition.

Field Emitter Image Sensor with HARP target

Aiming at ultrahigh sensitivity and high definition compact camera

Norifumi EGAMI, Senior Researcher,
Advanced Imaging Devices

Ultrahigh sensitivity and high definition compact cameras have been required to produce high quality TV images even in poor lighting conditions. To meet this requirement, we have been studying a new flat vacuum image sensor consisting of a matrix field emitter array and an ultrahigh sensitivity HARP (High-gain Avalanche Rushing amorphous Photoconductor) target.
Researches on devices with field emitter arrays have so far concentrated on flat panel displays. Image sensors with field emitter arrays, however, also have attractive potentials as next generation TV image pick-up devices. We have proposed a new field emitter image sensor with a HARP target, and investigated fundamental pick-up characteristics of an experimental image sensor.
The new image sensor with a field emitter array offers the following excellent performance.

  • Ultrahigh sensitivity
    By applying a new super-HARP target, sensitivity about 100 times as high as that of CCD can easily be achieved.
  • Ultrahigh resolution
    By using a matrix field emitter array with very small pixels, ultrahigh resolution can be obtained.
  • Wide dynamic-rang
    Dynamic-rang is determined on the amount of emission current from each pixel of the field emitter array. Wide dynamic-rang can be attained because the emission current density of the field emitter array is higher than that of the thermal cathode for the conventional camera tube.
  • Large-area image sensing
    Large-area image sensing suitable for medical X-ray pick-up is possible by using large size field emitter arrays.
  • Low power consumption

Field emitter array :A field emitter array is a two-dimensional matrix electron source consisting of many small pixels. Each pixel includes a one tip or many tips micro-miniature field emitter.
The field emitter is fabricated by IC technology, and electrons are emitted from each tip by applying voltage (electric-field). The field emitter has many advantages, such as low power consumption and high emission current density, compared to the conventional thermal cathode.