Hugging Face's logo

Spaces:
litagin
/
anime-whisper-demo
Running on Zero

App Files Community
anime-whisper-demo
File size: 4,782 Bytes 
bef66de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
409084a
 
 
 
 
 
 
 
 
 
bef66de
 
 
409084a
bef66de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import os
import time

import gradio as gr
from pydub import AudioSegment
from transformers import pipeline

is_hf = os.getenv("SYSTEM") == "spaces"

generate_kwargs = {
    "language": "Japanese",
    "do_sample": False,
    "num_beams": 1,
    "no_repeat_ngram_size": 3,
}


model_dict = {
    "whisper-large-v2": "openai/whisper-large-v2",
    "whisper-large-v3": "openai/whisper-large-v3",
    "whisper-large-v3-turbo": "openai/whisper-large-v3-turbo",
    "kotoba-whisper-v1.0": "kotoba-tech/kotoba-whisper-v1.0",
    "kotoba-whisper-v2.0": "kotoba-tech/kotoba-whisper-v2.0",
    "galgame-whisper-wip": (
        "litagin/galgame-whisper-wip"
        if is_hf
        else "../whisper_finetune/galgame-whisper"
    ),
}

# Download models
for model in model_dict.values():
    pipeline("automatic-speech-recognition", model=model)


def transcribe_common(audio: str, model: str) -> tuple[str, float]:
    # Get duration of audio
    duration = AudioSegment.from_file(audio).duration_seconds
    if duration > 15:
        return "Audio too long, limit is 15 seconds", 0
    start_time = time.time()
    pipe = pipeline("automatic-speech-recognition", model=model)
    end_time = time.time()
    return pipe(audio, generate_kwargs=generate_kwargs)["text"], end_time - start_time


def transcribe_large_v2(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["whisper-large-v2"])


def transcribe_large_v3(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["whisper-large-v3"])


def transcribe_large_v3_turbo(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["whisper-large-v3-turbo"])


def transcribe_kotoba_v1(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["kotoba-whisper-v1.0"])


def transcribe_kotoba_v2(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["kotoba-whisper-v2.0"])


def transcribe_galgame_whisper(audio) -> tuple[str, float]:
    return transcribe_common(audio, model_dict["galgame-whisper-wip"])


initial_md = """
# Galgame-Whisper (WIP) Demo

- 日本語のみ対応
- 他の書き起こしとついでに比較できるようにいろいろ入れた
- 現在0.1エポックくらい
- 速度はCPUです
- 音声は15秒まで

pipeのハイパラ:
```python
generate_kwargs = {
    "language": "Japanese",
    "do_sample": False,
    "num_beams": 1,
    "no_repeat_ngram_size": 3,
}
```
"""

with gr.Blocks() as app:
    gr.Markdown(initial_md)
    audio = gr.Audio(type="filepath")
    with gr.Row():
        with gr.Column():
            gr.Markdown("### Whisper-Large-V2")
            button_v2 = gr.Button("Transcribe with Whisper-Large-V2")
            output_v2 = gr.Textbox()
            time_v2 = gr.Textbox("Time taken")
        with gr.Column():
            gr.Markdown("### Whisper-Large-V3")
            button_v3 = gr.Button("Transcribe with Whisper-Large-V3")
            output_v3 = gr.Textbox()
            time_v3 = gr.Textbox("Time taken")
        with gr.Column():
            gr.Markdown("### Whisper-Large-V3-Turbo")
            button_v3_turbo = gr.Button("Transcribe with Whisper-Large-V3-Turbo")
            output_v3_turbo = gr.Textbox()
            time_v3_turbo = gr.Textbox()
    with gr.Row():
        with gr.Column():
            gr.Markdown("### Kotoba-Whisper-V1.0")
            button_kotoba_v1 = gr.Button("Transcribe with Kotoba-Whisper-V1.0")
            output_kotoba_v1 = gr.Textbox()
            time_kotoba_v1 = gr.Textbox("Time taken")
        with gr.Column():
            gr.Markdown("### Kotoba-Whisper-V2.0")
            button_kotoba_v2 = gr.Button("Transcribe with Kotoba-Whisper-V2.0")
            output_kotoba_v2 = gr.Textbox()
            time_kotoba_v2 = gr.Textbox("Time taken")
    with gr.Row():
        with gr.Column():
            gr.Markdown("### Galgame-Whisper (WIP)")
            button_galgame = gr.Button("Transcribe with Galgame-Whisper (WIP)")
            output_galgame = gr.Textbox()
            time_galgame = gr.Textbox("Time taken")

    button_v2.click(transcribe_large_v2, inputs=audio, outputs=[output_v2, time_v2])
    button_v3.click(transcribe_large_v3, inputs=audio, outputs=[output_v3, time_v3])
    button_v3_turbo.click(
        transcribe_large_v3_turbo,
        inputs=audio,
        outputs=[output_v3_turbo, time_v3_turbo],
    )
    button_kotoba_v1.click(
        transcribe_kotoba_v1, inputs=audio, outputs=[output_kotoba_v1, time_kotoba_v1]
    )
    button_kotoba_v2.click(
        transcribe_kotoba_v2, inputs=audio, outputs=[output_kotoba_v2, time_kotoba_v2]
    )
    button_galgame.click(
        transcribe_galgame_whisper,
        inputs=audio,
        outputs=[output_galgame, time_galgame],
    )

app.launch(inbrowser=True)