Spaces:
Running
Running
File size: 4,093 Bytes
0bb0d8e 16bc3e4 0bb0d8e da9138c 0bb0d8e 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 da9138c 16bc3e4 |
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 |
import gradio as gr
import os
import time
import sys
import subprocess
import tempfile
import requests
from urllib.parse import urlparse
# Clone and install faster-whisper from GitHub
subprocess.run(["git", "clone", "https://github.com/SYSTRAN/faster-whisper.git"], check=True)
subprocess.run(["pip", "install", "-e", "./faster-whisper"], check=True)
subprocess.run(["pip", "install", "yt-dlp"], check=True)
# Add the faster-whisper directory to the Python path
sys.path.append("./faster-whisper")
from faster_whisper import WhisperModel
from faster_whisper.transcribe import BatchedInferencePipeline
import yt_dlp
def download_audio(url):
parsed_url = urlparse(url)
if parsed_url.netloc == 'www.youtube.com' or parsed_url.netloc == 'youtu.be':
# YouTube video
ydl_opts = {
'format': 'bestaudio/best',
'postprocessors': [{
'key': 'FFmpegExtractAudio',
'preferredcodec': 'mp3',
'preferredquality': '192',
}],
'outtmpl': '%(id)s.%(ext)s',
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
info = ydl.extract_info(url, download=True)
return f"{info['id']}.mp3"
else:
# Direct MP3 URL
response = requests.get(url)
if response.status_code == 200:
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_file:
temp_file.write(response.content)
return temp_file.name
else:
raise Exception(f"Failed to download audio from {url}")
def transcribe_audio(input_source, batch_size):
# Initialize the model
model = WhisperModel("cstr/whisper-large-v3-turbo-int8_float32", device="auto", compute_type="int8")
batched_model = BatchedInferencePipeline(model=model)
# Handle input source
if isinstance(input_source, str) and (input_source.startswith('http://') or input_source.startswith('https://')):
# It's a URL, download the audio
audio_path = download_audio(input_source)
else:
# It's a local file path
audio_path = input_source
# Benchmark transcription time
start_time = time.time()
segments, info = batched_model.transcribe(audio_path, batch_size=batch_size)
end_time = time.time()
# Generate transcription
transcription = ""
for segment in segments:
transcription += f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}\n"
# Calculate metrics
transcription_time = end_time - start_time
real_time_factor = info.duration / transcription_time
audio_file_size = os.path.getsize(audio_path) / (1024 * 1024) # Size in MB
# Prepare output
output = f"Transcription:\n\n{transcription}\n"
output += f"\nLanguage: {info.language}, Probability: {info.language_probability:.2f}\n"
output += f"Duration: {info.duration:.2f}s, Duration after VAD: {info.duration_after_vad:.2f}s\n"
output += f"Transcription time: {transcription_time:.2f} seconds\n"
output += f"Real-time factor: {real_time_factor:.2f}x\n"
output += f"Audio file size: {audio_file_size:.2f} MB"
# Clean up downloaded file if it was a URL
if isinstance(input_source, str) and (input_source.startswith('http://') or input_source.startswith('https://')):
os.remove(audio_path)
return output
# Gradio interface
iface = gr.Interface(
fn=transcribe_audio,
inputs=[
gr.inputs.Textbox(label="Audio Source (Upload, MP3 URL, or YouTube URL)"),
gr.Slider(minimum=1, maximum=32, step=1, value=16, label="Batch Size")
],
outputs=gr.Textbox(label="Transcription and Metrics"),
title="Faster Whisper v3 turbo int8 transcription",
description="Enter an audio file path, MP3 URL, or YouTube URL to transcribe using Faster Whisper v3 turbo (int8). Adjust the batch size for performance tuning.",
examples=[
["https://www.youtube.com/watch?v=dQw4w9WgXcQ", 16],
["https://example.com/path/to/audio.mp3", 16],
["path/to/local/audio.mp3", 16]
],
)
iface.launch() |