File size: 3,544 Bytes
1852d76
 
b0744ce
1852d76
b0744ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1852d76
b0744ce
 
 
 
1852d76
fe80c45
b0744ce
 
1852d76
b0744ce
 
 
15812ac
b0744ce
15812ac
1852d76
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0744ce
 
 
 
 
1852d76
15812ac
 
 
1852d76
15812ac
 
 
 
 
1852d76
 
b0744ce
8a2ca70
 
1852d76
 
b0744ce
 
8a2ca70
b0744ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
456be9f
 
 
b0744ce
456be9f
b0744ce
1852d76
b0744ce
 
 
 
 
 
 
 
317a57f
 
 
b0744ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import re

import gradio as gr
import spaces
import torch
from omegaconf import OmegaConf
from transformers import pipeline

device = "cuda" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32


def load_pipe(model_id: str):
    return pipeline(
        "automatic-speech-recognition",
        model=model_id,
        max_new_tokens=128,
        chunk_length_s=30,
        batch_size=8,
        torch_dtype=torch_dtype,
        device=device,
    )


OmegaConf.register_new_resolver("load_pipe", load_pipe)

models_config = OmegaConf.to_object(OmegaConf.load("configs/models.yaml"))


@spaces.GPU
def automatic_speech_recognition(model_id: str, dialect_id: str, audio_file: str):
    model = models_config[model_id]["model"]

    generate_kwargs = {
        "task": "transcribe",
        "language": "id",
        "num_beams": 5,
    }
    if models_config[model_id]["dialect_mapping"] is not None:
        generate_kwargs["prompt_ids"] = torch.from_numpy(
            model.tokenizer.get_prompt_ids(dialect_id)
        ).to(device)

    result = model(audio_file, generate_kwargs=generate_kwargs)["text"].replace(
        f" {dialect_id}", ""
    )

    if result[-1] not in ".!?":
        result = result + "."

    sentences = re.split(r"[.!?] ", result)
    for i in range(len(sentences)):
        sentences[i] = sentences[i][0].upper() + sentences[i][1:]

    return " ".join(sentences)


def when_model_selected(model_id: str):
    model_config = models_config[model_id]

    if model_config["dialect_mapping"] is not None:
        dialect_drop_down_choices = [
            (k, v) for k, v in model_config["dialect_mapping"].items()
        ]

        return gr.update(
            choices=dialect_drop_down_choices,
            value=dialect_drop_down_choices[0][1],
        )
    else:
        return gr.update(visible=False)


def get_title():
    with open("DEMO.md") as tong:
        return tong.readline().strip("# ")


demo = gr.Blocks(
    title=get_title(),
    css="@import url(https://tauhu.tw/tauhu-oo.css);",
    theme=gr.themes.Default(
        font=(
            "tauhu-oo",
            gr.themes.GoogleFont("Source Sans Pro"),
            "ui-sans-serif",
            "system-ui",
            "sans-serif",
        )
    ),
)

with demo:
    default_model_id = list(models_config.keys())[0]
    model_drop_down = gr.Dropdown(
        models_config.keys(),
        value=default_model_id,
        label="模型",
    )

    dialect_drop_down = gr.Radio(
        choices=[
            "test"
            # (k, v)
            # for k, v in models_config[default_model_id]["dialect_mapping"].items()
        ],
        # value=list(models_config[default_model_id]["dialect_mapping"].values())[0],
        label="族別",
        visible=False,
    )

    model_drop_down.input(
        when_model_selected,
        inputs=[model_drop_down],
        outputs=[dialect_drop_down],
    )

    with open("DEMO.md") as tong:
        gr.Markdown(tong.read())

    gr.Interface(
        automatic_speech_recognition,
        inputs=[
            model_drop_down,
            dialect_drop_down,
            gr.Audio(
                label="上傳或錄音",
                type="filepath",
                waveform_options=gr.WaveformOptions(
                    sample_rate=16000,
                ),
            ),
        ],
        outputs=[
            gr.Text(interactive=False, label="辨識結果"),
        ],
        allow_flagging="auto",
    )

demo.launch()