Spaces:
Running
on
Zero
Running
on
Zero
File size: 3,815 Bytes
701a08b 5d06d89 ea81981 5d06d89 cb9cf5b 86bce7e 5d06d89 a8f0fd9 3555ad9 ea81981 5d06d89 ea81981 5d06d89 ea81981 5d06d89 ea81981 5d06d89 ea81981 6b58aeb 5d06d89 ea81981 |
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 |
import os
import torch
import gradio as gr
import pytube as pt
import spaces
from transformers import pipeline
from huggingface_hub import model_info
MODEL_NAME = os.environ.get("MODEL_NAME", "NbAiLab/whisper-large-sme")
lang = "fi"
share = (os.environ.get("SHARE", "False")[0].lower() in "ty1") or None
auth_token = os.environ.get("AUTH_TOKEN") or True
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")
@spaces.GPU(duration=120)
def pipe(file, return_timestamps=False):
asr = pipeline(
task="automatic-speech-recognition",
model=MODEL_NAME,
chunk_length_s=30,
device=device,
token=auth_token,
)
asr.model.config.forced_decoder_ids = asr.tokenizer.get_decoder_prompt_ids(
language=lang,
task="transcribe",
no_timestamps=not return_timestamps,
)
# asr.model.config.no_timestamps_token_id = asr.tokenizer.encode("<|notimestamps|>", add_special_tokens=False)[0]
return asr(file, return_timestamps=return_timestamps)
def transcribe(file, return_timestamps=False):
if not return_timestamps:
text = pipe(file)["text"]
else:
chunks = pipe(file, return_timestamps=True)["chunks"]
text = []
for chunk in chunks:
start_time = time.strftime('%H:%M:%S', time.gmtime(chunk["timestamp"][0])) if chunk["timestamp"][0] is not None else "??:??:??"
end_time = time.strftime('%H:%M:%S', time.gmtime(chunk["timestamp"][1])) if chunk["timestamp"][1] is not None else "??:??:??"
line = f"[{start_time} -> {end_time}] {chunk['text']}"
text.append(line)
text = "\n".join(text)
return 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, return_timestamps=False):
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")
text = transcribe("audio.mp3", return_timestamps=return_timestamps)
return html_embed_str, text
demo = gr.Blocks()
mf_transcribe = gr.Interface(
fn=transcribe,
inputs=[
gr.components.Audio(sources=['upload', 'microphone'], type="filepath"),
# gr.components.Checkbox(label="Return timestamps"),
],
outputs="text",
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 [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
" of arbitrary length."
),
allow_flagging="never",
)
yt_transcribe = gr.Interface(
fn=yt_transcribe,
inputs=[
gr.components.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
# gr.components.Checkbox(label="Return timestamps"),
],
examples=[["https://www.youtube.com/watch?v=mukeSSa5GKo"]],
outputs=["html", "text"],
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" [{MODEL_NAME}](https://huggingface.co/{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(share=True).queue()
|