File size: 4,473 Bytes
bc8cab0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import logging

import gradio as gr
import pytube as pt
import torch
from huggingface_hub import model_info
from transformers import pipeline

DEFAULT_MODEL_NAME = "bhuang/whisper-medium-cv11-french-case-punctuation"
MODEL_NAMES = [
    "bhuang/whisper-small-cv11-french",
    "bhuang/whisper-small-cv11-french-case-punctuation",
    "bhuang/whisper-medium-cv11-french",
    "bhuang/whisper-medium-cv11-french-case-punctuation",
]
CHUNK_LENGTH_S = 30

logging.basicConfig(
    format="%(asctime)s [%(levelname)s] [%(name)s] %(message)s",
    datefmt="%Y-%m-%dT%H:%M:%SZ",
)
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)

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

cached_models = {}

def maybe_load_cached_pipeline(model_name):
    pipe = cached_models.get(model_name)
    if pipe is None:
        # load pipeline
        pipe = pipeline(
            task="automatic-speech-recognition",
            model=model_name,
            chunk_length_s=CHUNK_LENGTH_S,
            device=device,
        )
        # set forced_decoder_ids
        pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="fr", task="transcribe")

        logger.info(f"`{model_name}` pipeline has been initialized")

        cached_models[model_name] = pipe
    return pipe


def transcribe(microphone, file_upload, model_name):
    warn_output = ""
    if (microphone is not None) and (file_upload is not None):
        warn_output = (
            "WARNING: You've uploaded an audio file and used the microphone. "
            "The recorded file from the microphone will be used and the uploaded audio will be discarded.\n"
        )

    elif (microphone is None) and (file_upload is None):
        return "ERROR: You have to either use the microphone or upload an audio file"

    file = microphone if microphone is not None else file_upload

    pipe = maybe_load_cached_pipeline(model_name)
    text = pipe(file)["text"]

    logger.info(f"Transcription: {text}")

    return warn_output + text


def _return_yt_html_embed(yt_url):
    video_id = yt_url.split("?v=")[-1]
    HTML_str = (
        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
        " </center>"
    )
    return HTML_str


def yt_transcribe(yt_url, model_name):
    yt = pt.YouTube(yt_url)
    html_embed_str = _return_yt_html_embed(yt_url)
    stream = yt.streams.filter(only_audio=True)[0]
    stream.download(filename="audio.mp3")

    pipe = maybe_load_cached_pipeline(model_name)
    text = pipe("audio.mp3")["text"]

    logger.info(f"Transcription: {text}")

    return html_embed_str, text


# load default model
maybe_load_cached_pipeline(DEFAULT_MODEL_NAME)

demo = gr.Blocks()

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(source="microphone", type="filepath", optional=True),
        gr.inputs.Audio(source="upload", type="filepath", optional=True),
        gr.inputs.Dropdown(choices=MODEL_NAMES, default=DEFAULT_MODEL_NAME, label="Whisper Model"),
    ],
    outputs="text",
    layout="horizontal",
    theme="huggingface",
    title="Whisper Demo: Transcribe Audio",
    description=(
        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the the fine-tuned"
        f" checkpoint [{DEFAULT_MODEL_NAME}](https://huggingface.co/{DEFAULT_MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length."
    ),
    allow_flagging="never",
)

yt_transcribe = gr.Interface(
    fn=yt_transcribe,
    inputs=[
        gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
        gr.inputs.Dropdown(choices=MODEL_NAMES, default=DEFAULT_MODEL_NAME, label="Whisper Model"),
    ],
    outputs=["html", "text"],
    layout="horizontal",
    theme="huggingface",
    title="Whisper Demo: Transcribe YouTube",
    description=(
        "Transcribe long-form YouTube videos with the click of a button! Demo uses the the fine-tuned checkpoint:"
        f" [{DEFAULT_MODEL_NAME}](https://huggingface.co/{DEFAULT_MODEL_NAME}) and 🤗 Transformers to transcribe audio files of"
        " arbitrary length."
    ),
    allow_flagging="never",
)

with demo:
    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Transcribe Audio", "Transcribe YouTube"])

# demo.launch(server_name="0.0.0.0", debug=True, share=True)
demo.launch(enable_queue=True)