Automatic phrasing in French

Philippe Martin

Whether we read aloud or silently a text in lexically stressed languages such as Italian or English, we segment speech not in words, but in sequences containing a content word (noun, adverb, verb or adjective) together with its associated grammatical word(s). These sequences are called accent phrases, and contain a single pitch accent whose location is defined in the lexicon. The simple fact that we can restore lexical stress in silent reading suggests that we don’t really need the actual presence of specific acoustic features of speech, such as vowel duration, fundamental frequency change or intensity modulation, often mentioned in the literature as parameters of stress. In French, the actual phrasing, i.e. the segmentation into accent phrases, depends strongly on the speech rate adopted by the speaker or the reader. Using a slow speech rate, all words could be stressed on their last syllable, whereas a fast speech rate could merge up to 10 syllables together in a single accent phrase containing more than one content word. Based on various characteristics of accent phrases in French, an algorithm operating in a top-down fashion for automatic identification of stressed syllables is described and applied on examples of read and spontaneous speech.

 DOI: 10.21437/SpeechProsody.2018-123

Cite as: Martin, P. (2018) Automatic phrasing in French. Proc. 9th International Conference on Speech Prosody 2018, 607-611, DOI: 10.21437/SpeechProsody.2018-123.

  author={Philippe Martin},
  title={Automatic phrasing in French},
  booktitle={Proc. 9th International Conference on Speech Prosody 2018},