Linguistic aspects of speech synthesis

Area of Science:

• Speech Technology
• Computational Linguistics
• Natural Language Processing

Background:

• Text-to-speech (TTS) synthesis requires a deep understanding of linguistic structures.
• Accurate pronunciation and prosody are crucial for natural-sounding synthesized speech.

Purpose of the Study:

• To outline the comprehensive linguistic analysis required for text-to-speech conversion.
• To detail the methods for deriving pronunciation, stress, and prosody from unrestricted text.

Main Methods:

• Morphological analysis and letter-to-sound conversion for word pronunciation.
• Part-of-speech tagging and phrase-level parsing for prosodic structure determination.
• Analysis of discourse factors (new/old information, contrast) to refine prosody.

Main Results:

• A multi-stage analysis pipeline is presented, starting from text input to a complete speech synthesis specification.
• The method accounts for word-level details (morphology, stress) and utterance-level prosody (duration, fundamental frequency).
• Discourse context is shown to significantly impact prosodic modifications.

Conclusions:

• A robust text-to-speech system relies on detailed linguistic analysis, including morphology, syntax, and discourse.
• The described framework enables the generation of natural and contextually appropriate synthesized speech.
• Future work may involve multilingual systems and rule-based frameworks.