Definition
Concatenative synthesis is a text-to-speech approach that generates speech by selecting and joining pre-recorded segments of human speech. A large database of recorded utterances is segmented into phonemes, diphones, or larger units. When the system needs to speak, it selects the best matching segments and concatenates them.
How it worked
The process involved recording a voice actor speaking thousands of carefully scripted utterances. These recordings were then cut into phonetic fragments and stored in a database. At synthesis time, the system selected the best matching fragments for the target text and joined them together, applying smoothing algorithms to reduce audible transitions.
Limitations
Despite decades of refinement, concatenative systems consistently produced speech that “sounded mechanical and contained artifacts such as glitches, buzzes and whistles.” The quality was constrained by the size and coverage of the recording database, and voices could not be easily modified or extended without recording new material.
Frequently Asked Questions
What is concatenative synthesis?
Concatenative synthesis is a text-to-speech method that works by cutting a large database of recorded speech into small phonetic units and reassembling them to form new utterances. It was the dominant commercial TTS approach from the 1990s until neural models replaced it around 2020-2022.
Why was concatenative synthesis replaced?
Concatenative synthesis produced speech with audible artifacts: glitches, buzzes, and unnatural transitions between phoneme segments. Neural TTS models generate audio sample by sample, producing much more natural-sounding speech with proper prosody and intonation.
Is concatenative synthesis still used?
Some legacy IVR systems and low-cost applications still use concatenative synthesis. Most commercial TTS providers have migrated to neural models, though basic non-neural voices remain available at lower price points.