Objective evaluation of grapheme to phoneme conversion for text-to-speech synthesis in French

Abstract

This paper reports on a cooperative international evaluation of grapheme-to-phoneme (GP) conversion for text-to-speech synthesis in French. Test methodology and test corpora are described. The results for eight systems are provided and analysed in some detail. The contribution of this paper is twofold: on the one hand, it gives an accurate picture of the state-of-the-art in the domain of GP conversion for French, and points out the problems still to be solved. On the other hand, much room is devoted to a discussion of methodological issues for this task. We hope this could help future evaluations of similar systems in other languages.

Predicting the phonetic realizations of word-final consonants in context – A challenge for French grapheme-to-phoneme converters

“…A study conducted by Yvon et al (1998) evaluated the quality of automatic G2P conversions produced by eight different systems. The study revealed that G2P conversions are still problematic in French, since even the best systems are prone to make at least one error in every 10 sentences.…”

“…Initially, we compare the predictions of liaisons produced by our prototype system with those generated by six other text-to-speech synthesizers (Yvon et al, 1998). This part is addressed in this section.…”

“…Roughly, it contains 1500 liaison candidates, of which about 600 are obligatory. Further details concerning the description of this data can be found in (Yvon et al, 1998). Although context classification labels are not available, corresponding phonetic transcriptions accompany this corpus.…”

“…The phonetic sequences were elaborated by human experts, including various acceptable pronunciations for the words/tokens of the corpus. Yvon et al (1998) used this data to conduct an objective evaluation of G2P conversion for French TTS synthesis in eight systems (A, B, C, D, E, F, G, and H). However, the performance of the systems A and E could not be reported due to misalignment problems.…”

Modified Grapheme Encoding and Phonemic Rule to Improve PNNR-Based Indonesian G2P

“…In general, a G2P is developed using machine learning-based methods, such as instance-based learning [1], table lookup with defaults [1], self-learning techniques [2], hidden Markov model [3], morphology and phoneme history [4], joint multigram models [5], conditional random fields [6], Kullback-Leibler divergence-based hidden Markov model [7]. These methods are commonly very complex and designed to be language independent, but they give varying performances for some phonemically complex languages, such as English, Dutch, French, and Germany.…”