![]() ![]() In addition to providing high-quality visualizations of speech patterns, Fon provides users with a powerful speech analysis capability. Users can also select the kind of relationship they are interested in (e.g., absolute or relative) and browse through the spectrogram to determine the probability distribution of the word's shape, location in the vocabulary, and shape of its occurrence in the phrase. ![]() Fon allows users to specify the number of time points at which a word occurs, and automatically creates a spectrum with the associated label. This ability has given linguists unprecedented access to the structure of languages. These tools have revolutionized how language researchers analyze languages by allowing them to examine the relationship between words, sounds, and meanings. Fon provides users with the opportunity to create and store custom spectrogram visualizations or generate new custom visualizations based on a large number of input fields. ![]() This approach results in an automatic, fine-grained and readable annotation, which is language-independent, speaker-independent and does not depend upon a particular phonological model of prosody.Fon is an award-winning analysis software that has been used by tens of thousands of language experts around the world to identify patterns and relationships among languages. Unlike commonly used supervised learning techniques the system does not require a labelled training corpus. The labelling system combines several processing steps: segmentation into syllabic nuclei, pause detection, pitch stylization, pitch range estimation, pitch movement classification, and pitch level assignment. Compound movements consist of a concatenation of simple ones. Five elementary pitch movements of individual syllables are distinguished on the basis of direction (rise, fall, level) and size (large and small melodic intervals, adjusted to the speaker’s pitch range). Five pitch levels are defined: Bottom and Top of the speaker’s pitch range, as well as Low, Mid, and High, which are determined on the basis of pitch changes in the local context. We describe a system for the automatic labelling of pitch levels and pitch movements in speech corpora. Provided additional modules for the detection of prominence and prosodic boundaries, the resulting annotation may serve as an input for a phonological annotation. For pitch levels low, mid and high an F-measure between 0.946 and 0.815 is obtained and for pitch movements a value between 0.708 and 1. ![]() The results, expressed in terms of standard measures of precision, recall, accuracy and F-measure are encouraging. The paper also includes a preliminary evaluation of the annotation system, for a reference corpus of nearly 14 minutes of spontaneous speech in French and Dutch, in order to quantify the annotation errors. It uses a dedicated and rule-based procedure, which unlike commonly used supervised learning techniques does not require a labeled corpus for training the model. The automatic tonal annotation system combines several processing steps: segmentation into syllable peaks, pause detection, pitch stylization, pitch range estimation, classification of the intra-syllabic pitch contour, and pitch level assignment. For pitch movements, both simple and compound, the transcription indicates direction (rise, fall, level) and size, using size categories (pitch intervals) adjusted relative to the speaker’s pitch range. Of the five pitch levels, three (low, mid, high) are defined on the basis of pitch changes in the local context and two (bottom, top) are defined relative to the boundaries of the speaker’s global pitch range. This fine-grained transcription provides labels indicating pitch level and pitch movement of individual syllables. This paper first proposes a labeling scheme for tonal aspects of speech and then describes an automatic annotation system using this transcription. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |