Ciro Martins

Broadcast news subtitling system in Portuguese

The subtitling of broadcast news programs are starting to become a very interesting application due to the technological advances in automatic speech recognition and associated technologies. However, to build this kind of systems, several advances are necessary both in terms of the technological components and on main blocks integration. In this paper, we are presenting the overall architecture of a subtitling system running daily at RTP (the Portuguese public broadcast company). The goal is to integrate our components in a system for the subtitling of RTP programs. The global system includes the subtitling of recorded and direct programs.

Dynamic language modeling for a daily broadcast news transcription system

When transcribing Broadcast News data in highly inflected languages, the vocabulary growth leads to high out-of-vocabulary rates. To address this problem, we propose a daily and unsupervised adaptation approach which dynamically adapts the active vocabulary and LM to the topic of the current news segment during a multi-pass speech recognition process. Based on texts daily available on the Web, a story-based vocabulary is selected using a morpho-syntatic technique. Using an Information Retrieval engine, relevant documents are extracted from a large corpus to generate a story-based LM. Experiments were carried out for a European Portuguese BN transcription system. Preliminary results yield a relative reduction of 65.2% in OOV and 6.6% in WER.

Situação epidemiológica da brucelose bovina no Estado de Santa Catarina

A study to characterize the brucellosis epidemiological situation in the State of Santa Catarina was carried out. The State was divided into five regions. Three hundred herds were randomly sampled in each region and a pre-established number of animals were sampled in each of these herds. A total of 7,801 serum samples from 1,586 herds were collected. In each herd, it was applied an epidemiological questionnaire regarding herd features and also husbandry and sanitary practices that could be associated with risk of infection. The serum samples were screened for antibodies to Brucella spp. by the Rose-Bengal Test (RBT), and all RBT-positive sera re-tested by the 2-mercaptoethanol test (2-ME). The herd was considered positive if at least one animal was positive on both RBT and 2-ME tests. The prevalences of infected herds and animals in Santa Catarina State were, respectively: 0.32% [0.10-0.69%] and 0.06% [0.0-0.17%]. The prevalence of infected herds in the regions were: region 1, 0.33% [0.0-0.99%]; region 2, 0.33% [0.0-1.0%]; region 3, 0.25% [0.0-0.75%]; region 4, 0.66% [0.08-1.84%]; and region 5, 0.33% [0.0-1.00%].

Speaker-adaptation in a hybrid HMM-MLP recognizer

Presently the most important systems for large vocabulary, continuous speech recognition are speaker-independent. These systems deal with the inter-speaker variability through a large pool of speakers. However, this approach has several drawbacks due to its inability to cope with the individual speaker characteristics. The problem is more extreme for the cases of fast or non-native speakers. In this paper we present a technique for speaker-adaptation in the context of a hybrid HMM-MLP system for large vocabulary, speaker-independent, continuous speech recognition. This technique is implemented both in supervised and unsupervised modes. In the unsupervised case both static and incremental approaches are explored. The results show that speaker-adaptation within the hybrid HMM-MLP framework can substantially improve system performance. In the incremental unsupervised mode, the improvement is obtained without any extra demands on the speaker, i.e. without an enrolment phase.

Dynamic Vocabulary Adaptation for a daily and real-time Broadcast News Transcription System

The daily and real-time transcription of broadcast news (BN) is a challenging task both in acoustic and in language modeling. To achieve optimal performance, several problems have to be overcome. Particularly, when transcribing BN data in highly inflected languages, the vocabulary growth leads to high OOV word rates. To address this problem, we propose a daily vocabulary and LM adaptation framework which directly extracts new words based on contemporary written news available on the Internet and some linguistic knowledge about the words found on those news. Experiments have been carried out for a European Portuguese BN transcription system. Preliminary results computed on 7 shows, yields a relative reduction of 61% in OOV and 2.1% in WER.

An incremental speaker-adaptation technique for hybrid HMM-MLP recognizer

One of the problems of speaker-independent continuous speech recognition systems is their inability to cope with the inter-speaker variability. When we find test speakers with different characteristics from the ones presented in the training pool we observe a large degradation on the system performance. To overcome this problem speaker-adaptation techniques may be used to provide near speaker-dependent accuracy. In this work we present a speaker-adaptation technique applied to a hybrid HMM-MLP system for large vocabulary, continuous speech recognition. This technique is based on an architecture that employs a trainable linear input network (LIN) to map the speaker specific features input vectors to the speaker-independent system. This speaker-adaptation technique is evaluated in an incremental speaker-adaptation task using a Wall Street Journal (WSJ) database. Both supervised and unsupervised modes are evaluated. The results show that speaker-adaptation within the hybrid framework can substantially improve system performance.

Dynamic language modeling for European Portuguese

This paper reports on the work done on vocabulary and language model daily adaptation for a European Portuguese broadcast news transcription system. The proposed adaptation framework takes into consideration European Portuguese language characteristics, such as its high level of inflection and complex verbal system. A multi-pass speech recognition framework using contemporary written texts available daily on the Web is proposed. It uses morpho-syntactic knowledge (part-of-speech information) about an in-domain training corpus for daily selection of an optimal vocabulary. Using an information retrieval engine and the ASR hypotheses as query material, relevant documents are extracted from a dynamic and large-size dataset to generate a story-based language model. When applied to a daily and live closed-captioning system of live TV broadcasts, it was shown to be effective, with a relative reduction of out-of-vocabulary word rate (69%) and WER (12.0%) when compared to the results obtained by the baseline system with the same vocabulary size.

Dynamic Language Modeling for the European Portuguese

Up-to-date language modeling is recognized to be a critical aspect of maintaining the level of performance for a speech recognizer over time for most applications. In particular for applications such as transcription of broadcast news and conversations where the occurrence of new words is very frequent, especially for highly inflected languages like the European Portuguese. An unsupervised adaptation approach, which dynamically adapts the active vocabulary and language model during a multi-pass speech recognition process, is presented. Experimental results confirmed the adequacy of the proposed approaches. Experiments were carried out for a European Portuguese Broadcast News transcription system with the best preliminary results yielding a relative reduction of 65.2% in OOV word rate and 6.6% in WER.