Vietnamese Piper voices (medium)
Three single-speaker Vietnamese narrator voices in Piper/VITS format (22,050 Hz, ~63 MB each), ready for
sherpa-onnx: the .onnx files carry sherpa-onnx metadata and each
folder ships a tokens.txt. Phonemization uses eSpeak NG (vi); espeak-ng-data is not included.
| Folder | Voice |
|---|---|
vits-piper-vi_VN-tramvang-medium |
Male, Northern accent, deep and resonant storyteller |
vits-piper-vi_VN-diemdam-medium |
Male, Northern accent, calm and mellow |
vits-piper-vi_VN-altotram-medium |
Female, Northern accent, warm low alto |
None of the voices belongs to a real person: each was designed from a text description with VoxCPM2 Voice Design.
Usage (sherpa-onnx, Python)
import sherpa_onnx
d = "vits-piper-vi_VN-altotram-medium"
tts = sherpa_onnx.OfflineTts(sherpa_onnx.OfflineTtsConfig(model=sherpa_onnx.OfflineTtsModelConfig(
vits=sherpa_onnx.OfflineTtsVitsModelConfig(
model=f"{d}/vi_VN-altotram-medium.onnx", tokens=f"{d}/tokens.txt",
data_dir="espeak-ng-data"))))
audio = tts.generate("Xin chào, đây là giọng đọc tiếng Việt.", sid=0, speed=1.0)
How they were made
- Training data. For each voice, roughly 5,000 Vietnamese sentences were synthesized with VoxCPM2 (Apache-2.0). About 30% of the sentences contain English names and terms, so the voices learn eSpeak's English phonemes. Every clip was transcribed back with Whisper, and clips with a high character error rate, an off-target pitch or an outlier speaking rate were discarded.
- Training. Each voice was fine-tuned with piper1-gpl for 80 epochs, starting from an earlier Vietnamese Piper checkpoint.
Lineage and license
The warm-start checkpoint descends from Piper's vi_VN-vais1000-medium. That voice was trained on the
VAIS-1000 corpus
(CC BY 4.0) and was itself fine-tuned from Piper's en_US-lessac-medium. The Lessac / Blizzard 2013 data
behind en_US-lessac-medium is licensed for research use only. Check whether that lineage is acceptable for
your use before relying on these weights.