import os import sys import gradio as gr device = "cuda" os.system('git clone https://github.com/Rudrabha/Wav2Lip.git') os.system('curl -o ./Wav2Lip/face_detection/detection/sfd/s3fd.pth https://www.adrianbulat.com/downloads/python-fan/s3fd-619a316812.pth') os.system('pip3 install moviepy') os.system('pip3 install speechRecognition') os.system('pip3 install gtts') os.system('pip3 install googletrans==3.1.0a0') os.system('pip install numba==0.48') os.system('pip3 install transformers') title = "Automatic translation and dubbing for Indic Languages" description = "A demo application to dub and translate videos spoken in Tamil, Hindi, Bengali and Telugu" article = "Official Repo: https://github.com/Rudrabha/Wav2Lip" def inference(language,speed,voice,video ): import moviepy.editor as mp clip = mp.VideoFileClip(video) clip.audio.write_audiofile(r"audio.wav") os.system('pip3 install pydub') os.system('pip3 install transformers==4.11.3 soundfile sentencepiece torchaudio librosa') speechlist = [] from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC import torch import torchaudio import librosa processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-960h-lv60-self") model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h-lv60-self") def get_transcription(audio_path): speech, sr = librosa.load(audio_path, sr=16000) resampler = torchaudio.transforms.Resample(sr, 16000) speech = resampler(speech) input_values = processor(speech, return_tensors="pt", sampling_rate=16000)["input_values"] logits = model(input_values)["logits"] predicted_ids = torch.argmax(logits, dim=-1) transcription = processor.decode(predicted_ids[0]) return transcription.lower() speechtext = get_transcription("audio.wav") speechlist.append(speechtext) text = " ".join(speechlist) from googletrans import Translator from gtts import gTTS translator= Translator() if speed == "Slow": con = True elif speed == "Fast": con = False if language == "Hindi": translation = translator.translate(text, src = 'en', dest='hi', slow=con) tts = gTTS(translation.text, lang= "hi") tts.save('input_audio.wav') elif language == "Tamil": translation = translator.translate(text, src = 'en', dest='ta', slow=con) tts = gTTS(translation.text, lang= "ta") tts.save('input_audio.wav') elif language == "Bengali": translation = translator.translate(text, src = 'en', dest='bn', slow=con) tts = gTTS(translation.text, lang= "hi") tts.save('input_audio.wav') elif language == "Telugu": translation = translator.translate(text, src = 'en', dest='te', slow=con) tts = gTTS(translation.text, lang= "hi") tts.save('input_audio.wav') audio = "input_audio.wav" os.system('mv ./Wav2Lip/* .') os.system("python inference.py --checkpoint_path ./wav2lip_gan.pth --face {} --audio {}".format(video, audio)) return "./results/result_voice.mp4" iface = gr.Interface(inference, inputs=[gr.Radio(["Tamil", "Hindi", "Bengali", "Telugu"], label = "Enter language to translate to"), gr.Radio(["Slow", "Fast"], label = "Enter speaking speed"), gr.Radio(["Male", "Female"], label = "Enter preferred voice"), gr.Video(format="mp4", sources="upload", label="Video to be Translated")], outputs=["video"], title=title, description=description, article=article) iface.launch()