File size: 2,213 Bytes
1f204d5
 
 
 
e1f8860
1f204d5
 
fba0c3b
 
1f204d5
 
 
 
 
c394a78
1f204d5
c394a78
1f204d5
 
 
c394a78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5149cfb
 
52e4ef5
1f204d5
 
 
c394a78
0ce0895
c394a78
1f204d5
 
 
 
c394a78
 
1f204d5
 
 
 
 
 
4b1ac93
c394a78
 
 
 
 
 
 
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
import torch
from transformers import pipeline
import gradio as gr

MODEL_NAME = "JackismyShephard/whisper-tiny-finetuned-minds14"
BATCH_SIZE = 8



device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
    return_timestamps='word'
)


# Copied from https://github.com/openai/whisper/blob/c09a7ae299c4c34c5839a76380ae407e7d785914/whisper/utils.py#L50
def format_timestamp(
    seconds: float, always_include_hours: bool = False, decimal_marker: str = "."
):
    if seconds is not None:
        milliseconds = round(seconds * 1000.0)

        hours = milliseconds // 3_600_000
        milliseconds -= hours * 3_600_000

        minutes = milliseconds // 60_000
        milliseconds -= minutes * 60_000

        seconds = milliseconds // 1_000
        milliseconds -= seconds * 1_000

        hours_marker = f"{hours:02d}:" if always_include_hours or hours > 0 else ""
        return f"{hours_marker}{minutes:02d}:{seconds:02d}{decimal_marker}{milliseconds:03d}"
    else:
        # we have a malformed timestamp so just return it as is
        return seconds


def transcribe(file, return_timestamps):
    outputs = pipe(
        file,
        batch_size=BATCH_SIZE,
        return_timestamps=return_timestamps,
    )
    text = outputs["text"]
    if return_timestamps:
        timestamps = outputs["chunks"]
        timestamps = [
            f"[{format_timestamp(chunk['timestamp'][0])} -> {format_timestamp(chunk['timestamp'][1])}] {chunk['text']}"
            for chunk in timestamps
        ]
        text = "\n".join(str(feature) for feature in timestamps)
    return text


demo = gr.Interface(
    fn=transcribe,
    inputs=[
        #gr.Audio(label="Audio", type="filepath"),
        gr.Audio(sources=["upload", "microphone"], type="filepath"),
        gr.Checkbox(label="Return timestamps"),
    ],
    outputs=gr.Textbox(show_copy_button=True, label="Text"),
    title="Automatic Speech Recognition",
    examples=[
        ["examples/example.wav", False],
        ["examples/example.wav", True],
    ],
    cache_examples=True,
    allow_flagging="never",
)

demo.launch()