Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from transformers import pipeline | |
| import torch | |
| import numpy as np | |
| device = "cuda:0" if torch.cuda.is_available() else "cpu" | |
| wav2_ft = pipeline("automatic-speech-recognition",model='sanchit-gandhi/wav2vec2-large-tedlium',device=device,trust_remote_code=True) | |
| app = gr.Blocks() | |
| def inference(path): | |
| out = wav2_ft( | |
| path, | |
| max_new_tokens=256, | |
| chunk_length_s=30, | |
| batch_size=8, | |
| ) | |
| return out['text'] | |
| def transcribe(stream, new_chunk): | |
| sr, y = new_chunk | |
| y = y.astype(np.float32) | |
| y /= np.max(np.abs(y)) | |
| if stream is not None: | |
| stream = np.concatenate([stream, y]) | |
| else: | |
| stream = y | |
| return stream, wav2_ft({"sampling_rate": sr, "raw": stream})["text"] | |
| mic_mode = gr.Interface( | |
| fn=inference, | |
| inputs=gr.Audio(sources="microphone", type='filepath', label="Record Your Lecture"), | |
| outputs=gr.Textbox(label="Transcription Output"), | |
| title="🎙️ Recording & Transcribe", | |
| description="Record through your mic. When you're done, hit stop and wait a moment. Feel free to trim the recording. Then, hit Submit!", | |
| examples=[], | |
| ) | |
| upload_mode = gr.Interface( | |
| fn=inference, | |
| inputs=gr.Audio(sources="upload", type='filepath', label="Upload Your Lecture Recording"), | |
| outputs=gr.Textbox(label="Transcription Output"), | |
| title="📂 Upload & Transcribe", | |
| description="Have a recorded lecture? Upload the audio file here, and it'll be transcribed in seconds!", | |
| ) | |
| # inspired by Gradio App Real Time Speech Recognition: https://www.gradio.app/guides/real-time-speech-recognition | |
| live_mode = gr.Interface( | |
| transcribe, | |
| ["state", gr.Audio(sources=["microphone"], streaming=True)], | |
| ["state", "text"], | |
| title="🎤 Live Transcription", | |
| description="Transcribe your lecture in real-time! Start speaking into your microphone, and watch the transcription appear instantly.", | |
| live=True, | |
| ) | |
| with app: | |
| gr.Markdown( | |
| """ | |
| # Lecture Transcription 📝 | |
| Welcome to **Lecture Transcription**, the go-to tool for transcribing lectures accurately. Whether you’re attending a live lecture or revisiting a recorded one, this app will ensure you don’t miss a single detail. | |
| ## How It Works | |
| - **Recording Mode:** Record the lecture as it happens. When you stop, your transcription will be generated. | |
| - **Upload Mode:** Upload your pre-recorded lecture audio files, and receive a precise transcription. Supports various audio formats including WAV, MP3, and more. | |
| - **Live Mode:** That's right, low-latency live transcription! | |
| ## Optimized for Technical Oration | |
| Under the hood, this is a Wav2Vec2 model fine-tuned on the TED-Lium dataset. It's well-versed for | |
| accurately transcribing technical speech. | |
| """ | |
| ) | |
| gr.TabbedInterface( | |
| [mic_mode, upload_mode,live_mode], | |
| ["🎙️ Record & Transcribe", "📂 Upload & Transcribe","🎤 Live Transcribe"] | |
| ) | |
| app.launch(debug=True) | |