Files
Michael Treadgold 6e1da4ddfb Add smoothness training pipeline with STFT/adversarial/vocoder-consistency losses
- inflect_nano/train_smooth.py: enhanced training with multi-res STFT loss,
  adversarial mel discriminator, vocoder consistency loss, deeper residual
  postnet, and cosine LR schedule
- preprocess_dataset.py: convert HF datasets, local dirs, or LJSpeech CSVs
  to the durations.jsonl format needed by training
- inference.py: add --smooth-prosody, --mel-smooth-sigma, --lowpass-hz flags
  for zero-cost inference-time quality improvements
- test_inference.py: smoke tests for model loading and synthesis
- colab_smooth_finetune.ipynb: Colab notebook for T4 GPU fine-tuning
- requirements.txt: add numba, scipy, datasets
2026-06-18 19:46:40 +12:00

186 lines
7.2 KiB
Python

from __future__ import annotations
import argparse
import math
import sys
from pathlib import Path
import numpy as np
import soundfile as sf
import torch
REPO_ROOT = Path(__file__).resolve().parent
VENDORED_FRONTEND = REPO_ROOT / "third_party" / "tiny_tts_frontend"
sys.path.insert(0, str(REPO_ROOT))
sys.path.insert(0, str(VENDORED_FRONTEND))
from tiny_tts.nn import commons
from tiny_tts.text import phonemes_to_ids
from tiny_tts.text.english import grapheme_to_phoneme, normalize_text
from tiny_tts.utils import ADD_BLANK
from inflect_nano.text_cleaning import clean_tinytts_text
from inflect_nano.vocoder import HifiGanGenerator, make_config
from inflect_nano.acoustic import MicroFastSpeech, MicroFastSpeechConfig
DEFAULT_ACOUSTIC = REPO_ROOT / "weights" / "inflect_nano_v1_acoustic.pt"
DEFAULT_VOCODER = REPO_ROOT / "weights" / "inflect_nano_v1_vocoder.pt"
def text_to_tokens(text: str) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
cleaned = clean_tinytts_text(text)
normalized = normalize_text(cleaned)
phones, tones, _ = grapheme_to_phoneme(normalized)
phone_ids, tone_ids, lang_ids = phonemes_to_ids(phones, tones, "EN")
if ADD_BLANK:
phone_ids = commons.insert_blanks(phone_ids, 0)
tone_ids = commons.insert_blanks(tone_ids, 0)
lang_ids = commons.insert_blanks(lang_ids, 0)
return torch.LongTensor(phone_ids), torch.LongTensor(tone_ids), torch.LongTensor(lang_ids)
def load_acoustic(path: Path, device: torch.device) -> tuple[MicroFastSpeech, dict[str, int], int]:
ckpt = torch.load(path, map_location=device, weights_only=False)
cfg = MicroFastSpeechConfig(**ckpt["config"])
model = MicroFastSpeech(cfg).to(device)
model.load_state_dict(ckpt["model"])
model.eval()
params = int(ckpt.get("params") or sum(p.numel() for p in model.parameters()))
return model, ckpt.get("speakers") or {"mark": 0}, params
def load_vocoder(path: Path, device: torch.device) -> tuple[HifiGanGenerator, int]:
ckpt = torch.load(path, map_location=device, weights_only=False)
cfg = make_config((ckpt.get("config") or {}).get("variant", "snake_v2mid"))
model = HifiGanGenerator(cfg).to(device)
model.load_state_dict(ckpt["generator"])
model.remove_weight_norm()
model.eval()
params = int(ckpt.get("generator_params") or sum(p.numel() for p in model.parameters()))
return model, params
def rms_db(audio: np.ndarray) -> float:
return 20.0 * math.log10(float(np.sqrt(np.mean(audio**2, dtype=np.float64))) + 1e-9)
def normalize_audio(audio: np.ndarray, target_rms_db: float = -20.0, peak_db: float = -1.0) -> np.ndarray:
audio = np.asarray(audio, dtype=np.float32).reshape(-1)
if audio.size == 0:
audio = np.zeros(1, dtype=np.float32)
audio = audio - float(audio.mean())
audio *= 10 ** ((target_rms_db - rms_db(audio)) / 20.0)
peak = float(np.max(np.abs(audio)) + 1e-9)
peak_limit = 10 ** (peak_db / 20.0)
if peak > peak_limit:
audio *= peak_limit / peak
return np.clip(audio, -1.0, 1.0)
def smooth_mel(mel: torch.Tensor, sigma: float = 1.0) -> torch.Tensor:
"""Gaussian temporal smoothing on mel frames to reduce frame-to-frame jitter."""
if sigma <= 0:
return mel
kernel_size = int(2 * math.ceil(2 * sigma) + 1)
if kernel_size < 3:
return mel
kernel = torch.exp(-0.5 * (torch.arange(kernel_size, device=mel.device, dtype=mel.dtype) - kernel_size // 2) ** 2 / sigma**2)
kernel = kernel / kernel.sum()
# [B, n_mels, T] -> pad last dim (time), then conv1d over time
pad = kernel_size // 2
mel_padded = torch.nn.functional.pad(mel, (pad, pad), mode="replicate")
kernel_expanded = kernel.view(1, 1, -1).expand(mel.shape[1], 1, -1)
return torch.nn.functional.conv1d(mel_padded, kernel_expanded, groups=mel.shape[1])
def apply_lowpass(wav: np.ndarray, cutoff_hz: float, sample_rate: int = 24000) -> np.ndarray:
"""Simple low-pass filter to reduce vocoder buzz above cutoff."""
if cutoff_hz <= 0 or cutoff_hz >= sample_rate / 2:
return wav
from scipy import signal
sos = signal.butter(4, cutoff_hz, btype="low", fs=sample_rate, output="sos")
return signal.sosfiltfilt(sos, wav).astype(np.float32)
@torch.inference_mode()
def synthesize(
text: str,
acoustic: MicroFastSpeech,
vocoder: HifiGanGenerator,
speakers: dict[str, int],
device: torch.device,
length_scale: float = 1.0,
pitch_scale: float = 1.0,
energy_scale: float = 1.0,
smooth_prosody: bool = False,
mel_smooth_sigma: float = 0.0,
lowpass_hz: float = 0.0,
) -> np.ndarray:
phone, tone, lang = text_to_tokens(text)
phone = phone.unsqueeze(0).to(device)
tone = tone.unsqueeze(0).to(device)
lang = lang.unsqueeze(0).to(device)
speaker = torch.LongTensor([int(speakers.get("mark", next(iter(speakers.values()), 0)))]).to(device)
mel = acoustic.infer(
phone,
tone,
lang,
speaker,
length_scale=float(length_scale),
pitch_scale=float(pitch_scale),
energy_scale=float(energy_scale),
smooth_predictors=smooth_prosody,
)
if mel_smooth_sigma > 0:
mel = smooth_mel(mel, mel_smooth_sigma)
wav = vocoder(mel).squeeze().detach().cpu().numpy()
wav = normalize_audio(wav)
if lowpass_hz > 0:
wav = apply_lowpass(wav, lowpass_hz)
return wav
def main() -> None:
ap = argparse.ArgumentParser(description="Run Inflect-Nano-v1 text-to-speech.")
ap.add_argument("--text", required=True)
ap.add_argument("--out", type=Path, default=Path("inflect_nano_v1_output.wav"))
ap.add_argument("--acoustic", type=Path, default=DEFAULT_ACOUSTIC)
ap.add_argument("--vocoder", type=Path, default=DEFAULT_VOCODER)
ap.add_argument("--device", default="cuda" if torch.cuda.is_available() else "cpu")
ap.add_argument("--length-scale", type=float, default=1.0)
ap.add_argument("--pitch-scale", type=float, default=1.0)
ap.add_argument("--energy-scale", type=float, default=1.0)
ap.add_argument("--smooth-prosody", action="store_true",
help="Apply 3-frame averaging to pitch/energy/brightness (reduces jitter)")
ap.add_argument("--mel-smooth-sigma", type=float, default=0.0,
help="Gaussian temporal smooth on mel frames (0.5-1.5 reduces buzzing)")
ap.add_argument("--lowpass-hz", type=float, default=0.0,
help="Low-pass cutoff in Hz (e.g. 8000-11000 reduces vocoder buzz)")
args = ap.parse_args()
device = torch.device(args.device)
acoustic, speakers, acoustic_params = load_acoustic(args.acoustic, device)
vocoder, vocoder_params = load_vocoder(args.vocoder, device)
audio = synthesize(
args.text,
acoustic,
vocoder,
speakers,
device,
length_scale=args.length_scale,
pitch_scale=args.pitch_scale,
energy_scale=args.energy_scale,
smooth_prosody=args.smooth_prosody,
mel_smooth_sigma=args.mel_smooth_sigma,
lowpass_hz=args.lowpass_hz,
)
args.out.parent.mkdir(parents=True, exist_ok=True)
sf.write(str(args.out), audio, 24000, subtype="PCM_16")
print(f"Wrote {args.out}")
print(f"Params: acoustic={acoustic_params:,} vocoder={vocoder_params:,} total={acoustic_params + vocoder_params:,}")
if __name__ == "__main__":
main()