#!/usr/bin/env python3 """M5-TTS (AX650) command line demo. python3 demo.py --model-dir models --assets assets \ --text "你好,欢迎使用我们的语音合成模型。" \ --prompt-wav samples/aslp_default_voice.wav \ --prompt-text "猪携带该疾病,然后通过蚊子传染给人类。" --output out.wav --stream writes audio packets while the LM is still generating (TTFA). """ import argparse import json import os import sys import time import wave import numpy as np sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) from m5_ax import M5AX, write_wav # noqa: E402 def main(): here = os.path.dirname(os.path.abspath(__file__)) ap = argparse.ArgumentParser() ap.add_argument("--model-dir", default=os.environ.get("M5_MODEL_DIR", "models")) local_assets = os.path.join(here, "assets") default_assets = local_assets if os.path.isdir(local_assets) else os.path.join("models", "assets") ap.add_argument("--assets", default=os.environ.get("M5_ASSETS", default_assets)) ap.add_argument("--engine", default="ax", choices=["ax", "onnx"]) ap.add_argument("--text", default="你好,欢迎使用我们的语音合成模型。") ap.add_argument("--prompt-wav", default=os.path.join(here, "samples", "aslp_default_voice.wav")) ap.add_argument("--prompt-text", default="猪携带该疾病,然后通过蚊子传染给人类。") ap.add_argument("--lang", default="Chinese") ap.add_argument("--seed", type=int, default=1234) ap.add_argument("--max-frames", type=int, default=400) ap.add_argument("--stream", action="store_true", help="write packets while generating (lower TTFA)") ap.add_argument("--first-frames", type=int, default=4, help="codes in the first packet") ap.add_argument("--chunk-frames", type=int, default=8, help="codes per packet afterwards") ap.add_argument("--no-overlap", action="store_true", help="decode the codec sequentially (benchmark A/B)") ap.add_argument("--repeat", type=int, default=1, help="repeat runs (warm reference cache)") ap.add_argument("--output", default="out.wav") a = ap.parse_args() tts = M5AX(a.model_dir, a.assets, engine=a.engine) for _ in range(max(1, a.repeat) - 1): tts.generate(a.text, a.prompt_wav, a.prompt_text, lang=a.lang, seed=a.seed, max_frames=min(a.max_frames, 4)) started = time.perf_counter() if a.stream: with wave.open(a.output, "wb") as handle: handle.setnchannels(1) handle.setsampwidth(2) handle.setframerate(tts.sr) for pcm in tts.synthesize_stream(a.text, a.prompt_wav, a.prompt_text, lang=a.lang, seed=a.seed, max_frames=a.max_frames, first_frames=a.first_frames, chunk_frames=a.chunk_frames, overlap=not a.no_overlap): handle.writeframes((np.clip(pcm, -1, 1) * 32767.0).astype("