Context dependent initial/final acoustic modeling for continuous Chinese speech recognition

Acoustic modeling is important for continuous Chinese speech recognition. The extended initial/final (XIF) set is the basic speech recognition unit set to analyze the Chinese language characteristics outperformed the standard IF set. Decision tree-based state tying technology was used to construct the content dependent initial/final acoustic model (Tri-XIF model), with an appropriate question set design based on Chinese linguistic knowledge. The methods were developed to optimize the Tri-XIF modeling, including transcription refinement, question set extension, and model size reduction. Tests show that the Tri-XIF modeling is much better than either Tri-phone modeling or syllable modeling, the syllable error rate reduced by 24.53% comparing to the Tri-phone model and 41.65% to syllable model. The model size was reduced by above 20% with little performance deterioration using the methods in the Tri-XIF model.