Search

23 Jun 2003: Language Model. • Sources of data (using LDC character-to-word segmentor)– Acoustic training data (modifier Kneser-Ney)– News corpora: TDT[2,3,4], China Radio, People’s Daily, Xinhua (Good-. ... Acoustic 206.6 190.8AcousticNews Corpora 199.6 179.8

1 Mar 2000: Experiments in Broadcast News Transcription. P.C. Woodland, T. Hain, S.E. Johnson, T. ... FX. all other speech (e.g. spontaneous non-native). Table 1: Broadcast news focus conditions.

29 Apr 2024: 19] uFACT: Unfaithful alien-corpora training for semantically consistent data-to-text generation. ... We propose uFACT (Un-Faithful Alien Corpora Training), a training corpus construction method for data-to-text (d2t) generation models.

22 Nov 2006: The two acoustictraining data sources, and each of the news corpora, were kept asdistinct sources for language model (LM) generation. ... The total contributionfrom all the news corpora was about 0.12, with the majority fromPeople’s Daily (0.09).

20 Feb 2018: in-domainutterance pairs, and up to 91.4% when adding the out-of-domainbilingual corpora detailed in Section 2.2. ... 11] J. Tiedemann, “News from OPUS - A collection of multi-lingual parallel corpora with tools and interfaces,” in Re-cent Advances

15 Jun 2018: Experimentswere conducted on the Penn Tree Bank and BBC Multi-GenreBroadcast News (MGB) corpora, where the proposed approachsignificantly outperforms standard forms of recurrent models inperplexity. ... PTB consists mainly oftext related to finance,

27 May 2004: The rest of the paper is organised as follows. In Section 2, wedescribe the English broadcast news corpora that used in this work.Then, our lightly supervised discriminative training approach ispresented ... Rich Tran-scription Workshop, 2003. [4] D.

3 Nov 2023: 13] uFACT: Unfaithful alien-corpora training for semantically consistent data-to-text generation. ... We propose uFACT (Un-Faithful Alien Corpora Training), a training corpus construction method for data-to-text (d2t) generation models.

23 Dec 2004: The total contribution fromall the news corpora was about 0.12, with the majority from Peo-ple’s Daily (0.09). ... All experiments use the interpolated language modelwith the news corpora. Language Model System (S3) CER (%)dev04.

1 Mar 2000: 41-48 (Lansdowne,VA, Feb. 1998). The 1997 HTK BROADCAST NEWS TRANSCRIPTION SYSTEM. ... using the broadcast news training texts, the acoustic training data and 1995 Marketplace transcriptions.