PrimeTTS v2.1 · live on a Jetson Nano …

zh-TW + English TTS. Every waveform is synthesized on a single Jetson Nano Gen1 — one continuous utterance per request, streamed back as it is produced. UI inspired by the PrimeTTS-Streaming demo; inference runs on real Jetson hardware.

Type & speak

voice speed first-audio – · audio 0.0s · gaps 0 ms

Backend

checking…

Use it in your voice bot

The backend implements the OpenAI text-to-speech API (POST /v1/audio/speech, pcm/wav, 16 kHz, streaming), so it works as the TTS of a Hugging Face speech-to-speech voice bot:

speech-to-speech local --tts openai \
  --openai_tts_base_url http://<your-gateway>:8091/v1 \
  --openai_tts_voice 0 --openai_tts_response_format pcm \
  --openai_tts_sample_rate 16000

Self-host the Nano server and gateway from the repo.

Shared demo box. The first request after the Nano idles can take ~1 s longer (GPU clock ramp). Code: custom-runtime · docs/server.md.