Multilingual, multi-voice real-time TTS
Core model parameters ~0.6B
Model files 1.62 GB
Starts speaking as text arrives; 10 languages, 9 voices, any length as one continuous stream
- Fast first sound
- 10 languages
- 9 voices
Recorded on the board
Screen recording of the board web demo, recorded on Qualcomm QCS8550.
Highlights
Plays while generating
Playback starts with the first chunk; the rest is generated while it plays.
Text of any length
Long text is split at sentence ends and played as one continuous stream with natural pauses.
9 built-in voices
Male and female voices in several styles, including Beijing and Sichuan dialect speakers.
10 languages
Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian, or Auto from the text.
Fully on-device
Everything runs locally on the board's NPU: no cloud, no GPU.
Easy to integrate
A small C API and a Python binding; the same text, voice and seed give the same audio.
Pipeline
From input to output: where every step runs.
- 01Browser
Text in
Text of any length, with a voice and a language
- 02CPU
Text processing
Numbers and dates normalised, split at sentence ends; a short first segment for a quick start
- 03NPU
Speech codes
Generates the speech codes frame by frame at 12 Hz
- 04NPU
Detail codes
Fills in the remaining code layers of every frame
- 05NPU
Decode and play
Decodes 0.64 s of audio at a time; the browser plays each chunk as it arrives
Supported hardware
- Core model parameters
- ~0.6B
- Model files
- 1.62 GB
- License
- Pandora License