Detection of precisely transcribed parts from inexact transcribed corpus

Abstract

Although large-scale spontaneous speech corpora are crucial resource for various domains of spoken language processing, they are usually limited due to their construction cost especially in transcribing precisely. On the other hand, inexact transcribed corpora like shorthand notes, meeting records and closed captions are widely available. Unfortunately, it is difficult to use them directly as speech corpora for learning acoustic models, because they contain two kinds of text, precisely transcribed parts and edited parts. In order to resolve this problem, this paper proposes an automatic detection method of precisely transcribed parts from inexact transcribed corpora. Our method consists of two steps: the first step is an automatic alignment between the inexact transcription and its corresponding utterance, and the second step is a support vector machine based detector of precisely transcribed parts using several features obtained by the first step. Experiments using the Japanese National Diet Record shows that automatic detection of precise parts is effective for lightly supervised speaker adaptation, and shows that it achieves reasonable performance to reduce the converting cost from inexact transcribed corpora into precisely transcribed ones.

Automatic Alignment Between Classroom Lecture Utterances and Slide Components

“…a speech signal and its transcription). The typical approach of the first category is to synchronize two contents with the dynamic programming (DP) algorithm based on acoustic similarity information [10]. The second category is that target contents are not parallel but are comparable at least (e.g.…”

Improving the Readability of ASR Results for Lectures Using Multiple Hypotheses and Sentence-Level Knowledge

“…For example, transcripts from the Japanese National Diet Record (JNDR) would be appropriate to train the language model since the transcripts are created by cleaning the raw spontaneous transcripts manually by professional stenographers. As an additional experiment, we combined the newspaper corpus (214 M words) and the JNDR (184 M words) [29], and trained a new language model using the combined cor- Table 4. With the new language model, our proposed method provided an almost same result for both corpora (53.5% → 53.4% and 39.1% → 39.0%).…”