Latency is not a static label
The same model behaves differently across hardware, audio length, streaming protocol, and concurrency. The KG only narrows the field.
Decision guide · TTS
How should you choose low-latency Chinese TTS?
Real-time narration, support, video voice-over, and local applications have different requirements. Compare first-packet latency, streaming, Chinese coverage, deployment cost, licensing, and integration—not just voice quality.
Three filters
- Interaction Can it stream, and what are the target first-packet and full-response latencies?
- Language and voice Are Mandarin, dialects, voice identity, emotion, or multiple speakers important?
- Operating conditions Can local CPU/GPU, cloud API, concurrency, and licensing requirements be met?
Do not rank by model name alone
Match the constraints first
Search uses Chinese, open-source, local, and real-time constraints while preserving dimensions that are not yet recorded.