| |
| """Run several synthesis cases on the board and save wavs (for ASR / listening).""" |
| import os, sys, time |
| sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) |
| from cutetts_ax import CuteTTSAX |
| from cutetts_ax import audio as A |
|
|
| MODEL = os.environ.get("CUTETTS_MODEL_DIR", "compile") |
| ASSETS = os.environ.get("CUTETTS_ASSETS", "sdk/assets") |
| OUT = os.environ.get("CUTETTS_OUT", "scratch/board_samples") |
| os.makedirs(OUT, exist_ok=True) |
|
|
| CASES = [ |
| ("en_default", "The voice generated by this model sounds amazing!", |
| "sdk/samples/reference.wav", 42), |
| ("en_long", "Hello, this is a test of the CuteTTS text to speech system running on the AX650 neural processing unit.", |
| "sdk/samples/reference.wav", 7), |
| ("zh_default", "今天天气很好，我们一起去公园散步吧。", |
| "sdk/samples/reference.wav", 123), |
| ("en_other_voice", "Quick brown fox jumps over the lazy dog.", |
| "/ai_data/AX615/hy/workspace/Speech-Translation-ZipVoice/samples/en_16k.wav", 2024), |
| ] |
|
|
| tts = CuteTTSAX(MODEL, ASSETS, engine=os.environ.get("CUTETTS_ENGINE", "ax")) |
| for tag, text, ref, seed in CASES: |
| if not os.path.exists(ref): |
| print("skip (no ref)", tag); continue |
| t0 = time.time() |
| wav = tts.synthesize(text, ref, seed=seed) |
| A.write_wav(os.path.join(OUT, tag + ".wav"), wav, tts.sr) |
| tm = tts.last_timing |
| print(f"{tag}: {tm['audio_seconds']:.2f}s audio, {tm['steps']} patches, " |
| f"total {tm['total']:.2f}s, RTF {tm['total']/max(tm['audio_seconds'],1e-6):.2f} | " |
| f"vae_enc {tm['vae_enc']:.2f} spk {tm['speaker_enc']:.2f} prefill {tm['prefill']:.2f} " |
| f"ar {tm['ar_loop']:.2f} (dit {tm['dit']:.2f} dec {tm['lm_decode']:.2f}) vae_dec {tm['vae_dec']:.2f}") |
|
|
