python/example.py

#!/usr/bin/env python3
"""Inflect AX TTS SDK example.

Real text synthesis (requires phonemizer + espeak-ng, see README):

    python example.py \
        --encoder ../models/ax620e/encoder.axmodel \
        --decoder ../models/ax620e/decoder.axmodel \
        --text "The quick brown fox jumps over the lazy dog." \
        --output out.wav

Host dry-run without NPU and without eSpeak (onnxruntime stand-in on the
fp32 ONNX graphs, deterministic dummy tokens from
../models/model_meta.json numeric_baseline):

    python example.py --demo-tokens \
        --encoder ../model_convert/export/encoder.onnx \
        --decoder ../model_convert/export/decoder.onnx \
        --output demo.wav
"""

from __future__ import annotations

import argparse
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent))

from inflect_ax_tts import DUMMY_PHONEME_IDS, InflectTTS, write_wav


def main() -> None:
    parser = argparse.ArgumentParser(description="Inflect AX TTS SDK example")
    parser.add_argument("--encoder", required=True, help="encoder .axmodel (or .onnx)")
    parser.add_argument("--decoder", required=True, help="decoder .axmodel (or .onnx)")
    parser.add_argument("--text", help="text to synthesize (needs phonemizer + eSpeak-ng)")
    parser.add_argument(
        "--demo-tokens",
        action="store_true",
        help="skip the text frontend and use the deterministic EXPORT dummy "
        "phoneme ids (no eSpeak dependency)",
    )
    parser.add_argument("--output", required=True, help="output wav path")
    parser.add_argument("--speed", type=float, default=1.0)
    parser.add_argument("--variation", type=float, default=0.667)
    parser.add_argument("--seed", type=int, default=0)
    parser.add_argument(
        "--backend",
        default="auto",
        choices=["auto", "axengine", "onnxruntime"],
        help="inference backend (auto: .axmodel -> pyaxengine, .onnx -> onnxruntime)",
    )
    args = parser.parse_args()

    if bool(args.text) == bool(args.demo_tokens):
        parser.error("provide exactly one of --text or --demo-tokens")

    tts = InflectTTS(args.encoder, args.decoder, backend=args.backend)
    if args.demo_tokens:
        sample_rate, wav = tts.synthesize_tokens(
            DUMMY_PHONEME_IDS,
            speed=args.speed,
            variation=args.variation,
            seed=args.seed,
        )
    else:
        sample_rate, wav = tts.synthesize(
            args.text, speed=args.speed, variation=args.variation, seed=args.seed
        )
    write_wav(args.output, wav, sample_rate)
    print(
        f"wrote {args.output}: {wav.size} samples "
        f"({wav.size / sample_rate:.2f} s @ {sample_rate} Hz), "
        f"amp [{wav.min():.3f}, {wav.max():.3f}]"
    )


if __name__ == "__main__":
    main()