Rule-based speech synthesis survives only where computational power is scarce.

Evidence

The largest problem with rule-based text-to-speech synthesizers is the quality of speech. When unlimited text is automatically transformed into speech, the synthesized speech almost always sounds ‘robotic’ and is barely intelligible. These methods have been largely abandoned after the introduction of data-based speech synthesis, which will be discussed below. However, in some special cases, knowledge-based speech synthesizers have their uses. For example, in a situation where data-based speech synthesis would require far more computational power than is available and where a lower quality of speech could be tolerated, the rule-based solutions could still be used. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2243

Categorization

• XREF: Connects to the broader pattern in computing where data-driven approaches displace rule-based heuristics once compute becomes cheap — parallels the shift seen in machine translation and image recognition.