import torch import spaces import gradio as gr import soundfile as sf import numpy as np import pytube as pt import librosa from transformers import AutoProcessor, Wav2Vec2BertForCTC MODEL_NAME = "mikr/w2v-bert-2.0-czech-colab-cv16" device = 0 if torch.cuda.is_available() else "cpu" processor = AutoProcessor.from_pretrained(MODEL_NAME) model = Wav2Vec2BertForCTC.from_pretrained(MODEL_NAME).to(device) @spaces.GPU def text_from_audio(audio_path): a, s = librosa.load(audio_path, sr=16_000) input_values = processor(a, sampling_rate=s, return_tensors="pt").input_features with torch.no_grad(): logits = model(input_values.to(device)).logits predicted_ids = torch.argmax(logits, dim=-1) # transcribe speech transcription = processor.batch_decode(predicted_ids) text = transcription[0] return text def transcribe(microphone, file_upload): 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" audio_path = microphone if microphone is not None else file_upload text = text_from_audio(audio_path) return warn_output + text def _return_yt_html_embed(yt_url): video_id = yt_url.split("?v=")[-1] HTML_str = ( f'
' "
" ) return HTML_str def yt_transcribe(yt_url): 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 = text_from_audio("audio.mp3") return html_embed_str, text demo = gr.Blocks() mf_transcribe = gr.Interface( fn=transcribe, inputs=[ gr.Audio(sources="microphone", type="filepath"), gr.Audio(sources="upload", type="filepath"), ], outputs="text", title="W2V Bert 2.0 Demo: Transcribe Czech Audio", description=( "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses 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.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")], outputs=["html", "text"], title="W2V Bert 2.0 Demo: Transcribe Czech YouTube Video", 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(server_name="0.0.0.0")