| """Text -> 24 kHz WAV with the Maya 2 MLX port. |
| |
| .venv/bin/python -I speak.py "Hello there." -o hello.wav [--lang en] [--voice Malini] [--seed 7] |
| .venv/bin/python -I speak.py --voices # the named voices |
| |
| Same defaults as the official Maya3TTS.generate: plain voice, English, ~20 s pieces with 0.15 s gaps, no text |
| normalisation unless --normalize, a fresh engine-wide seed per piece unless --seed. |
| """ |
| import argparse |
| import os |
| import sys |
| import time |
| import wave |
|
|
| sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) |
|
|
| import numpy as np |
|
|
| from maya2_mlx import Maya2 |
|
|
|
|
| def write_wav(path, wav, sr=24000): |
| pcm = np.rint(np.asarray(wav, np.float32) * np.float32(32767)).astype("<i2") |
| with wave.open(path, "wb") as f: |
| f.setnchannels(1) |
| f.setsampwidth(2) |
| f.setframerate(sr) |
| f.writeframes(pcm.tobytes()) |
|
|
|
|
| def main(): |
| ap = argparse.ArgumentParser(description=__doc__.split("\n")[0]) |
| ap.add_argument("text", nargs="?") |
| ap.add_argument("-o", "--out", default="maya2.wav") |
| ap.add_argument("--lang", default="en") |
| ap.add_argument("--voice", default=None) |
| ap.add_argument("--seed", type=int, default=None) |
| ap.add_argument("--temperature", type=float, default=None) |
| ap.add_argument("--top-k", type=int, default=None) |
| ap.add_argument("--top-p", type=float, default=None) |
| ap.add_argument("--repetition-penalty", type=float, default=None) |
| ap.add_argument("--max-frames", type=int, default=None) |
| ap.add_argument("--normalize", action="store_true", help="run the bundled text normaliser first") |
| ap.add_argument("--voices", action="store_true", help="list the voices and exit") |
| ap.add_argument("--models", default=None, help="weights folder (default: models/maya2-8bit, else models/maya2)") |
| a = ap.parse_args() |
| tts = Maya2(a.models) if a.models else Maya2() |
| if a.voices: |
| for vid in tts.voices: |
| m = tts.registry.get(vid).meta |
| print(f"{vid:10s} {m.get('gender', ''):7s} {m.get('age_band', ''):7s} {m.get('description', '')}") |
| return |
| if not a.text: |
| ap.error("text is required") |
| t0 = time.perf_counter() |
| wav, info = tts.generate(a.text, a.lang, a.voice, max_frames=a.max_frames, return_info=True, normalize=a.normalize, |
| seed=a.seed, temperature=a.temperature, top_k=a.top_k, top_p=a.top_p, |
| repetition_penalty=a.repetition_penalty) |
| el = time.perf_counter() - t0 |
| write_wav(a.out, wav) |
| print(f"{a.out}: {len(wav) / 24000:.2f} s audio in {el:.1f} s ({os.path.basename(tts.models_dir)}), pieces {info['pieces']}, " |
| f"stops {[s['reason'] for s in info['stops']]}, seeds {[s['seed'] for s in info['stops']]}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
