Spaces:
Paused
Paused
File size: 2,780 Bytes
bfb5ccb 8ca2e83 eaed2c2 8ca2e83 1c803c5 8ca2e83 0c872e7 bfb5ccb 8ca2e83 0c872e7 8ca2e83 eaed2c2 8ca2e83 eaed2c2 |
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 |
import spaces
import gradio as gr
import torch
import torchaudio
from transformers import AutoModelForCTC, Wav2Vec2BertProcessor
import yt_dlp
model = AutoModelForCTC.from_pretrained("anzorq/w2v-bert-2.0-kbd")
processor = Wav2Vec2BertProcessor.from_pretrained("anzorq/w2v-bert-2.0-kbd")
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
@spaces.GPU
def transcribe_speech(audio):
# Load the audio file
waveform, sr = torchaudio.load(audio)
# Resample the audio if needed
resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=16000)
waveform = resampler(waveform)
# Convert to mono if needed
if waveform.dim() > 1:
waveform = torch.mean(waveform, dim=0)
# Normalize the audio
waveform = waveform / torch.max(torch.abs(waveform))
# Extract input features
input_features = processor(waveform.unsqueeze(0), sampling_rate=16000).input_features
input_features = torch.from_numpy(input_features).to(device)
# Generate logits using the model
with torch.no_grad():
logits = model(input_features).logits
# Decode the predicted ids to text
pred_ids = torch.argmax(logits, dim=-1)[0]
pred_text = processor.decode(pred_ids)
return pred_text
@spaces.GPU
def transcribe_from_youtube(url):
# Download audio from YouTube using yt-dlp
audio_path = "downloaded_audio.wav"
ydl_opts = {
'format': 'bestaudio/best',
'outtmpl': audio_path,
'postprocessors': [{
'key': 'FFmpegExtractAudio',
'preferredcodec': 'wav',
'preferredquality': '192',
}],
'postprocessor_args': ['-ar', '16000'], # Ensure audio is at 16000 Hz
'prefer_ffmpeg': True,
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([url])
# Transcribe the downloaded audio
return transcribe_speech(audio_path)
with gr.Blocks() as demo:
with gr.Tab("Microphone Input"):
gr.Markdown("## Transcribe speech from microphone")
mic_audio = gr.Audio(source="microphone", type="filepath", label="Speak into your microphone")
transcribe_button = gr.Button("Transcribe")
transcription_output = gr.Textbox(label="Transcription")
transcribe_button.click(fn=transcribe_speech, inputs=mic_audio, outputs=transcription_output)
with gr.Tab("YouTube URL"):
gr.Markdown("## Transcribe speech from YouTube video")
youtube_url = gr.Textbox(label="Enter YouTube video URL")
transcribe_button = gr.Button("Transcribe")
transcription_output = gr.Textbox(label="Transcription")
transcribe_button.click(fn=transcribe_from_youtube, inputs=youtube_url, outputs=transcription_output)
demo.launch() |