Chapter 2
Orthographic and phonetic transcriptions of Rhapsodie recording
In this chapter, we present the principles that we used for the orthographic and phonological transcriptions in Rhapsodie, as well as the process of automatic segmentation. We opted for three main principles in the orthographic transcription: (1) no adaptation of the standard spelling using tricks such as i-z-ont or pasque; (2) no punctuation; (3) phenomena that are peculiar to speech are duly represented: filled pauses, word repetitions, self-repairs, word fragments, interjections, onomatopoeias, and discourse markers. Then a phonetic transcription is obtained using an automatic grapheme-to-phoneme (g2p) conversion tool followed by manual verification; lastly, on the basis of the sound recording and the phonetic transcription, we provide a multi-layer alignment (or segmentation) at the phonetic, syllabic, and lexical levels, thanks to an automatic approach based on a speech recognition engine.
Article outline
- 1.Introduction
- 2.From sound to orthographic transcription
- 2.1The myth of the copyist
- 2.2.“Transcription as theory”
- 2.3What to transcribe?
- 3.The three leading principles of orthographic transcription in Rhapsodie
- 3.1Standard spelling
- 3.2Punctuation and silent pauses
- 3.3Handling of features that are specific to speech
- 3.4Transcription conventions
- 3.5Availability of orthographic transcription in two formats
- 4.Phonetic transcription and segmentation
- 4.1Macro-segmentation at utterance level
- 4.2Grapheme-to-phoneme conversion
- 4.3Phone segmentation
- 4.4Creation of a syllabic tier
- 5.Specific processing of multi-speaker recordings
- 6.Quality and productivity
- 7.Conclusion