File size: 4,772 Bytes
8d6afae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
"""Single-item inference CLI for OmniVoice.

Generates audio from a single text input using voice cloning,
voice design, or auto voice.

Usage:
    # Voice cloning
    omnivoice-infer --model k2-fsa/OmniVoice \
        --text "Hello, this is a text for text-to-speech." \
        --ref_audio ref.wav --ref_text "Reference transcript." --output out.wav

    # Voice design
    omnivoice-infer --model k2-fsa/OmniVoice \
        --text "Hello, this is a text for text-to-speech." \
        --instruct "male, British accent" --output out.wav

    # Auto voice
    omnivoice-infer --model k2-fsa/OmniVoice \
        --text "Hello, this is a text for text-to-speech." --output out.wav
"""

import argparse
import logging

import torch

import soundfile as sf

from omnivoice.models.omnivoice import OmniVoice
from omnivoice.utils.common import str2bool


def get_best_device():
    """Auto-detect the best available device: CUDA > MPS > CPU."""
    if torch.cuda.is_available():
        return "cuda"
    if torch.backends.mps.is_available():
        return "mps"
    return "cpu"


def get_parser() -> argparse.ArgumentParser:
    parser = argparse.ArgumentParser(
        description="OmniVoice single-item inference",
        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
    )
    parser.add_argument(
        "--model",
        type=str,
        default="k2-fsa/OmniVoice",
        help="Model checkpoint path or HuggingFace repo id.",
    )
    parser.add_argument(
        "--text",
        type=str,
        required=True,
        help="Text to synthesize.",
    )
    parser.add_argument(
        "--output",
        type=str,
        required=True,
        help="Output WAV file path.",
    )
    # Voice cloning
    parser.add_argument(
        "--ref_audio",
        type=str,
        default=None,
        help="Reference audio file path for voice cloning.",
    )
    parser.add_argument(
        "--ref_text",
        type=str,
        default=None,
        help="Reference text describing the reference audio.",
    )
    # Voice design
    parser.add_argument(
        "--instruct",
        type=str,
        default=None,
        help="Style instruction for voice design mode.",
    )
    parser.add_argument(
        "--language",
        type=str,
        default=None,
        help="Language name (e.g. 'English') or code (e.g. 'en').",
    )
    # Generation parameters
    parser.add_argument("--num_step", type=int, default=32)
    parser.add_argument("--guidance_scale", type=float, default=2.0)
    parser.add_argument("--speed", type=float, default=1.0)
    parser.add_argument(
        "--duration",
        type=float,
        default=None,
        help="Fixed output duration in seconds. If set, overrides the "
        "model's duration estimation. The speed factor is automatically "
        "adjusted to match while preserving language-aware pacing.",
    )
    parser.add_argument("--t_shift", type=float, default=0.1)
    parser.add_argument("--denoise", type=str2bool, default=True)
    parser.add_argument(
        "--postprocess_output",
        type=str2bool,
        default=True,
    )
    parser.add_argument("--layer_penalty_factor", type=float, default=5.0)
    parser.add_argument("--position_temperature", type=float, default=5.0)
    parser.add_argument("--class_temperature", type=float, default=0.0)
    parser.add_argument(
        "--device",
        type=str,
        default=None,
        help="Device to use for inference. Auto-detected if not specified.",
    )
    return parser


def main():
    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
    logging.basicConfig(format=formatter, level=logging.INFO, force=True)

    args = get_parser().parse_args()

    device = args.device or get_best_device()
    logging.info(f"Loading model from {args.model} on {device} ...")
    model = OmniVoice.from_pretrained(
        args.model, device_map=device, dtype=torch.float16
    )

    logging.info(f"Generating audio for: {args.text[:80]}...")
    audios = model.generate(
        text=args.text,
        language=args.language,
        ref_audio=args.ref_audio,
        ref_text=args.ref_text,
        instruct=args.instruct,
        duration=args.duration,
        num_step=args.num_step,
        guidance_scale=args.guidance_scale,
        speed=args.speed,
        t_shift=args.t_shift,
        denoise=args.denoise,
        postprocess_output=args.postprocess_output,
        layer_penalty_factor=args.layer_penalty_factor,
        position_temperature=args.position_temperature,
        class_temperature=args.class_temperature,
    )

    sf.write(args.output, audios[0], model.sampling_rate)
    logging.info(f"Saved to {args.output}")


if __name__ == "__main__":
    main()