Speech Transcription Task Research Articles

Intelligibility, recall, and voice evaluation across accents

Speech elicits variable responses from listeners. Voices can vary in their intelligibility, how well listeners can recall messages produced by the voice, and what kind of social evaluation it explicitly or implicitly evokes. These different responses may be due to individual-specific attributes within a voice or the accent it carries. For example, familiar or more standard language varieties may be more intelligible producing more easily recalled, and eliciting more positive social evaluations from listeners. The current study uses 35 English-speaking voices from 7 different language backgrounds that vary in familiarity and prestige to the listener (n = 430) population, which is a representative heterogeneous sampling from the local university community. Specific voices were chosen from a larger data set based on their acoustic similarity. Listeners either completed a speech transcription task (quantifying intelligibility) or a cloze task (quantifying recall) and all listeners provided an evaluation of the voices’ likability and perceived comprehensibility. Bayesian data analysis is used to quantify and characterize the relationship between a voice’s intelligibility and how well it is recalled, and whether this relationship is predicted by social evaluation and listener experience. These results have implications for theories of speech recognition and how listeners process accents.

Habilidades de processamento auditivo central e exame pericial de comparação de locutor

ABSTRACT Purpose to describe which abilities of central auditory processing are more frequently related for the group of specialists to the performance of the speaker comparison test (CL), traditionally performed by forensic experts. Methods a prospective, descriptive study with quantitative and qualitative analysis and data were obtained through a consensus of experts. Five speech therapists participated in the meeting, two specialists in audiology (EA), two specialists in voice (VS), and an expert speech therapist (FP). The meeting was held virtually and synchronously, lasting 1 hour and 30 minutes. The tasks performed during the Speaker Comparison (LC) exam were considered from a protocol available in the literature. The AEs received explanations about each of the tasks and were asked to discuss which auditory processing skills (ACP) would be involved in the performance of each of them. Results seven PAC skills were considered in the experts' meeting as essential for the tasks performed in the CL exam. Temporal ordering was the most cited skill, being present in six tasks, and the speech material transcription task is the one that requires more skills from the PAC. Conclusion Seven PAC skills were considered in the experts' meeting as essential for the tasks performed in the CL exam. Temporal ordering was the most cited skill, being present in six tasks, and the speech material transcription task is the one that requires more skills from the PAC.

Filled pause refinement based on the pronunciation probability for lecture speech.

Nowadays, although automatic speech recognition has become quite proficient in recognizing or transcribing well-prepared fluent speech, the transcription of speech that contains many disfluencies remains problematic, such as spontaneous conversational and lecture speech. Filled pauses (FPs) are the most frequently occurring disfluencies in this type of speech. Most recent studies have shown that FPs are widely believed to increase the error rates for state-of-the-art speech transcription, primarily because most FPs are not well annotated or provided in training data transcriptions and because of the similarities in acoustic characteristics between FPs and some common non-content words. To enhance the speech transcription system, we propose a new automatic refinement approach to detect FPs in British English lecture speech transcription. This approach combines the pronunciation probabilities for each word in the dictionary and acoustic language model scores for FP refinement through a modified speech recognition forced-alignment framework. We evaluate the proposed approach on the Reith Lectures speech transcription task, in which only imperfect training transcriptions are available. Successful results are achieved for both the development and evaluation datasets. Acoustic models trained on different styles of speech genres have been investigated with respect to FP refinement. To further validate the effectiveness of the proposed approach, speech transcription performance has also been examined using systems built on training data transcriptions with and without FP refinement.

Use of contexts in language model interpolation and adaptation

Language models (LMs) are often constructed by building multiple individual component models that are combined using context independent interpolation weights. By tuning these weights, using either perplexity or discriminative approaches, it is possible to adapt LMs to a particular task. This paper investigates the use of context dependent weighting in both interpolation and test-time adaptation of language models. Depending on the previous word contexts, a discrete history weighting function is used to adjust the contribution from each component model. As this dramatically increases the number of parameters to estimate, robust weight estimation schemes are required. Several approaches are described in this paper. The first approach is based on MAP estimation where interpolation weights of lower order contexts are used as smoothing priors. The second approach uses training data to ensure robust estimation of LM interpolation weights. This can also serve as a smoothing prior for MAP adaptation. A normalized perplexity metric is proposed to handle the bias of the standard perplexity criterion to corpus size. A range of schemes to combine weight information obtained from training data and test data hypotheses are also proposed to improve robustness during context dependent LM adaptation. In addition, a minimum Bayes’ risk (MBR) based discriminative training scheme is also proposed. An efficient weighted finite state transducer (WFST) decoding algorithm for context dependent interpolation is also presented. The proposed technique was evaluated using a state-of-the-art Mandarin Chinese broadcast speech transcription task. Character error rate (CER) reductions up to 7.3% relative were obtained as well as consistent perplexity improvements.

Training data selection for improving discriminative training of acoustic models

This paper considers training data selection for discriminative training of acoustic models for large vocabulary continuous speech recognition (LVCSR). Three novel data selection approaches are proposed. First, the average phone accuracy over all hypothesized word sequences in the word lattice of a training utterance is utilized for utterance-level data selection. Second, phone-level data selection based on the difference between the expected accuracy of a phone arc and the average phone accuracy of the word lattice is investigated. Finally, frame-level data selection based on the normalized frame-level entropy of Gaussian posterior probabilities obtained from the word lattice is explored. The underlying characteristics of the presented approaches are extensively investigated and their performance is verified by comparison with standard discriminative training approaches. Experiments conducted on a broadcast news speech transcription task show that with the aid of phone- and frame-level data selection we can reduce more than half of the turnaround time for acoustic model training and simultaneously obtain a comparably good set of discriminative acoustic models.

The use of fundamental frequency information by cochlear implant patients with residual low‐frequency hearing

Recent studies have demonstrated significant benefits of complementing the electric hearing provided by a cochlear implant (CI) with residual low‐frequency acoustic hearing. It has been suggested that the benefit of low‐frequency hearing (<250 Hz) is due to improved representation of fundamental frequency (F0). It was hypothesized that removing F0 information would negatively impact speech perception in bimodal CI patients (EAS) more so when they are using both modes of perception. It was further hypothesized that these patients' more effective use of F0 information would result in higher intelligibility scores and less unpredicted lexical boundary errors (LBEs) in a speech transcription task. Six EAS patients transcribed normal and F0‐flattened phrases. Results indicate EAS patients demonstrated a significant benefit from the residual low‐frequency acoustic hearing and perform better in this mode. Five of the six EAS patients were able to transcribe the normal phrases with greater accuracy than the flat F0 phrases in both conditions. These patients correctly parsed lexical boundaries and transcribed phonemic targets with greater accuracy in the bimodal condition. These results demonstrate the importance of pitch and low frequency spectral information in the perception of speech.

Efficient WFST-Based One-Pass Decoding With On-The-Fly Hypothesis Rescoring in Extremely Large Vocabulary Continuous Speech Recognition

This paper proposes a novel one-pass search algorithm with on-the-fly composition of weighted finite-state transducers (WFSTs) for large-vocabulary continuous-speech recognition. In the standard search method with on-the-fly composition, two or more WFSTs are composed during decoding, and a Viterbi search is performed based on the composed search space. With this new method, a Viterbi search is performed based on the first of the two WFSTs. The second WFST is only used to rescore the hypotheses generated during the search. Since this rescoring is very efficient, the total amount of computation required by the new method is almost the same as when using only the first WFST. In a 65k-word vocabulary spontaneous lecture speech transcription task, our proposed method significantly outperformed the standard search method. Furthermore, our method was faster than decoding with a single fully composed and optimized WFST, where our method used only 38% of the memory required for decoding with the single WFST. Finally, we have achieved high-accuracy one-pass real-time speech recognition with an extremely large vocabulary of 1.8 million words

Discriminative Training for Large-Vocabulary Speech Recognition Using Minimum Classification Error

The minimum classification error (MCE) framework for discriminative training is a simple and general formalism for directly optimizing recognition accuracy in pattern recognition problems. The framework applies directly to the optimization of hidden Markov models (HMMs) used for speech recognition problems. However, few if any studies have reported results for the application of MCE training to large-vocabulary, continuous-speech recognition tasks. This article reports significant gains in recognition performance and model compactness as a result of discriminative training based on MCE training applied to HMMs, in the context of three challenging large-vocabulary (up to 100 k word) speech recognition tasks: the Corpus of Spontaneous Japanese lecture speech transcription task, a telephone-based name recognition task, and the MIT Jupiter telephone-based conversational weather information task. On these tasks, starting from maximum likelihood (ML) baselines, MCE training yielded relative reductions in word error ranging from 7% to 20%. Furthermore, this paper evaluates the use of different methods for optimizing the MCE criterion function, as well as the use of precomputed recognition lattices to speed up training. An overview of the MCE framework is given, with an emphasis on practical implementation issues

Large‐scale continuous speech recognition system design using discriminative training

Discriminative training is difficult to implement but essential to attaining state‐of‐the‐art performance in automatic speech recognition systems today. Most of the discriminative training results for large scale recognition tasks (with vocabularies well over 10<th>000 words) so far use the maximum mutual information (MMI) framework, but recent results for the minimum classification error (MCE) framework suggest that MCE too yields significant improvements in recognition accuracy and system compactness on large‐scale tasks. MCE embodies rather well the general intuition that recognition system design should attempt to improve performance (i.e., recognition accuracy) directly, by optimizing a criterion function that is closely related to performance, rather than indirectly, by optimizing a criterion such as overall log likelihood that does not reflect performance. This presentation provides an overview of the MCE framework, and describes recent MCE speech recognition results obtained with both the MIT Galaxy system and the NTT Communication Science Labs SOLON system. The tasks examined include a 33<th>000‐word vocabulary telephone‐based spontaneous speech weather information task, a 22<th>000‐word telephone‐based name recognition task, and a 100<th>000‐word Japanese lecture speech transcription task.

Vocal tract normalization in speech recognition: Compensating for systematic speaker variability

The performance of speech recognition systems is often improved by accounting explicitly for sources of variability in the data. In the SWITCHBOARD corpus, studied during the 1994 CAIP workshop [Frontiers in Speech Processing Workshop II, CAIP (August 1994)], an attempt was made to compensate for the systematic variability due to different vocal tract lengths of various speakers. The method found a maximum probability parameter for each speaker which mapped an acoustic model to the mean of the models taken from a homogeneous speaker population. The underlying acoustic model was that of a straight tube, and the parameter estimation was accomplished by warping the spectrum of each speaker linearly over a 20% range (actually accomplished by digitally resampling the data), and finding the maximum a posteriori probability of the data given the warp. The technique produces statistically significant improvements in accuracy on a speech transcription task using each of four different speech recognition systems. The best parametrizations were later found to correlate well with vocal tract estimates computed manually from spectrograms.