sopho-poetry-tts (m3_v6)
On-device Chinese classical-poetry TTS for the SuFei
Android app. A compact FastSpeech2 acoustic model (7.6 M params) plus the
PaddleSpeech HiFi-GAN CSMSC vocoder, both exported to ONNX. Total deploy size
~68 MB, runs on CPU (onnxruntime) with no network at runtime.
Files
| File | Bytes | SHA-256 (first 12) | Role |
|---|---|---|---|
fastspeech2_sufei.onnx | 35,591,617 | 9e58042c158a | acoustic model: phone ids → mel[T,80] |
hifigan_csmsc.onnx | 32,432,756 | eb305fdc8be9 | vocoder: mel → 24 kHz pcm |
phone_id_map.txt | 2,128 | ca00d60618a0 | phone → id (vocab 268) |
norm_stats.npz | 1,156 | 19304d7074d2 | mel mean/std for de-normalization |
manifest.json | — | — | metadata + per-file sha256 |
SHA256SUMS.txt carries the canonical checksums; the app's Gradle fetchTtsModels
task pins this revision and verifies each file's sha256 before bundling it.
Pipeline
text → G2P (PaddleSpeech phones) → phone_ids
→ fastspeech2_sufei.onnx → mel[T,80]
→ * mel_std + mel_mean (norm_stats.npz)
→ hifigan_csmsc.onnx → 24 kHz pcm
- Sample rate 24000 Hz, mel bins 80, phone vocab 268.
- G2P convention: PaddleSpeech phones; the interpunct
·produces no token (the m3_v6 dot-token fix); punctuation，。？！kept,；：→，.
On-device notes (fp32 + mobile-safe length regulator)
Two changes make this model run correctly under onnxruntime-android:
- Length regulator — exported with a vectorized
cumsum + GreaterOrEqual-sum + Gatherformulation instead oftorch.repeat_interleave. The latter exports toLoop+SplitToSequence/ConcatFromSequence; the vectorized form uses only standard tensor ops and is numerically identical (max|diff| = 0). - fp32, not fp16 — an fp16 FS2 yields a 0-frame mel on the ARM
onnxruntime-android CPU EP (the fp16 duration path collapses), while the same
fp16 graph is correct on x86. fp32 is exact and verified on-device (instrumented
NarOnnxEngineTest). fp16 halves the acoustic model but is unusable on the target runtime, so fp32 ships.
Quality
ASR-CER (FunASR), holdout-20 / train-30 seed42: train 20.6 % / holdout 21.0 %
(fp32 baseline; production v9 was 24.0 % / 21.5 %). Checkpoint m3_v6_final.pt
(step 30000).
Training data
Distilled from the sopho-poetry-tts-data
cosyvoice3/ lineage (Fun-CosyVoice3-0.5B teacher).
License
Apache-2.0. The vocoder hifigan_csmsc.onnx is PaddleSpeech's pretrained HiFi-GAN
CSMSC, redistributed unchanged under Apache-2.0; the FastSpeech2 acoustic model was
trained by this project. Teacher audio was produced by Apache-2.0 models, which place
no restriction on their outputs.