Speaking like a nativeConvert the English speech rhythms of native Japanese speaker into native-like speechNTT Communication Science Laboratories

This is a technique called non-negative temporal decomposition, which extracts speech rhythms from English speech spoken by native Japanese speakers and transforms them to sound like those of a native English speaker. Speech rhythm is thought to play an important role in smooth speech communication when using English, and in the future, this technology may reduce the difficulty of communicating in English by native Japanese speakers.

Difficulties in spoken communication using a foreign language exist both when speaking and listening. Many native Japanese speakers, in particular, have difficulty understanding what English speakers are saying, or find that their sentences are not understood by native English speakers. This technology provides a method for converting the speech rhythm of English speech when spoken by native Japanese speakers, focusing on the fact that even if there are some errors in pronunciation, it is easy to understand as long as the rhythm is correct.

• The vocal-tract spectrum extracted from speech is decomposed into pronunciation and speech rhythms
• An algorithm that takes into account speech mechanisms not found in conventional methods (non-negative matrix factorization)
• Introduction of a model that expresses the speech rhythm in phonemic units
• Considers the fact that only adjacent phonemes affect temporal information
• Capable of converting speech rhythms for various languages (e.g., Japanese for foreign students in Japan)

• Use with efficient English learning methods
• Conversion to native English rhythms through mobile devices
• Conversion to English rhythms for video conferences and presentations

The vocal-tract spectrum, a speech parameter that corresponds to the position of the speech organs, contains both the pronunciation and the speech rhythm. Using non-negative temporal decomposition, the vocal-tract spectrum is decomposed into pronunciation and speech rhythms, and the speech rhythms are converted to sound like those of a native English speaker in one’s own voice.

NTT Communication Science Laboratories - Human Information Science Laboratory