pocket-tts

Verdict: pocket-tts is worth adopting first if you specifically need local/offline, CPU-friendly, embeddable speech synthesis. If you need a turnkey commercial stack, official mobile SDKs, or clearly proven high-throughput production performance, the current evidence is not enough. It is not a full voice assistant, not ASR, and not a chat model; a better analogy is a lightweight deployable speech output engine.

In practice, the official GitHub repo and multiple reposts consistently highlight a roughly 100M-parameter model that runs on CPU, with interest around streaming TTS, voice cloning, and privacy-friendly local use cases. Social posts also mention first-audio latency, 6 languages, and community Android experiments, but those are mostly reposts or demo-style claims. They prove attention, not reliability. Stronger evidence for capability judgment comes from the official repository itself and benchmark commentary calling it “the slowest but the most interesting,” which suggests its appeal is local deployment and flexibility rather than raw speed leadership.