Spontaneous Speech Corpus of Japanese

By K. Maekawa, H. Koiso, Sadaoki Furui, Hitoshi Isahara. International Conference on Language Resources and Evaluation, 2000

435 citations recorded by the source corpus.

Design issues of a spontaneous speech corpus is described. The corpus under compilation will contain 800-1000 hour spontaneously uttered Common Japanese speech and the morphologically annotated transcriptions. Also, segmental and intonation labeling will be provided for a subset of the corpus. The primary application domain of the corpus is speech recognition of spontaneous speech, but we plan to make it useful for natural language processing and phonetic/linguistic studies also.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/ac23e62a18186c9a51dfea16d3602f52ff8c3b3a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Improving Transformer-Based End-to-End Speech Recognition with Connectionist Temporal Classification and Language Model Integration (2019, methodology): “For the third task, we used the Corpus of Spontaneous Japanese (CSJ) [18], which contains 581 hours of Japanese lectures, because we are also interested in ASR for non-English languages.”
Domain adaptation of DNN acoustic models using knowledge distillation (2017, background): “Training set of the source model consisted of 1200 hours of Japanese utterances recorded in various acoustic environments, and includes real data of voice search application, call center recordings, Corpus of Spontaneous Japanese (academic presentations) [18], and Japanese Newspaper Article Sentence”
Hybrid CTC/Attention Architecture for End-to-End Speech Recognition (2017, methodology): “The second part of the experiments scaled up the size of the corpora by using the Corpus of Spontaneous Japanese (CSJ) [36] and HKUST Mandarin Chinese conversational telephone speech recognition (HKUST) [37].”
ESPnet: End-to-End Speech Processing Toolkit (2018, methodology): “The other experiments compare the performance of ESPnet with state-of-the-art ASR systems for the CSJ and HKUST tasks.”
Neural Dialogue Context Online End-of-Turn Detection (2018, methodology): “F0 represents 2 dimensional sequential features of F0 and ∆ F0; frame shift was set to 5 ms. SENONE represents 256-dimensional senone bottleneck features extracted from 3-layer senone LSTM-RNN with 256 units trained from a corpus of spontaneous Japanese speech (Maekawa et al., 2000).”
Joint CTC/attention decoding for end-to-end speech recognition (2017, methodology): “Although the processors were multicore CPUs and the computer had GPUs, we ran the decoding program as a Table 3 shows the relationships between the real-time factor (RTF) and the CER for the CSJ and HKUST tasks.”
Advances in Joint CTC-Attention Based End-to-End Speech Recognition with a Deep CNN Encoder and RNN-LM (2017, methodology): “The Japanese task is lecture speech recognition using the Corpus of Spontaneous Japanese (CSJ) [20].”
Language Independent End-to-End Architecture For Joint Language and Speech Recognition (methodology): “The experiments, which used speech databases composed of Wall Street Journal (English), Corpus of Spontaneous Japanese, HKUST Mandarin CTS, and Voxforge (German, Spanish, French, Italian, Dutch, Portuguese, Russian), demonstrate comparable/superior performance relative to language-dependent end-to-e”
CIAIR Simultaneous Interpretation Corpus (2004, background): “Introduction Recently, spoken language corpora have been constructed for the purpose of studying on speech processing in many research organizations (for example [1, 2]).”
A Comparative Study on Transformer vs RNN in Speech Applications (2019, background): “AISHELL [28] zh 170 read dev / test AURORA4 [29] (*) en 15 noisy read (dev 0330) A / B / C / D CSJ [30] ja 581 spontaneous eval1 / eval2 / eval3 CHiME4 [31] (*) en 108 noisy far-field multi-ch read dt05 simu / dt05 real / et05 simu / et05 real CHiME5 [32] en 40 noisy far-field multi-ch conversationa”