Spaces:
Running
Running
| import gradio as gr | |
| import torchaudio | |
| import torch | |
| import os | |
| import time | |
| import soundfile as sf | |
| languages = { | |
| "English": "eng", | |
| "Hindi": "hin", | |
| "Portuguese": "por", | |
| "Russian": "rus", | |
| "Spanish": "spa" | |
| } | |
| welcome_message = """ | |
| # Welcome to Tonic's Unity On Device! | |
| Tonic's Unity On Device uses [facebook/seamless-m4t-unity-small](https://huggingface.co/facebook/seamless-m4t-unity-small) for audio translation & accessibility. | |
| Tonic's Unity On Device!🚀 on your own data & in your own way by cloning this space. Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/TeamTonic/SeamlessOnDevice?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a></h3> | |
| ### Join us : | |
| TeamTonic is always making cool demos! Join our active builder's community on Discord: [Discord](https://discord.gg/GWpVpekp) On Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On Github: [Polytonic](https://github.com/tonic-ai) & contribute to [PolyGPT](https://github.com/tonic-ai/polygpt-alpha)" | |
| """ | |
| def save_and_resample_audio(input_audio_path, output_audio_path, resample_rate=16000): | |
| waveform, sample_rate = torchaudio.load(input_audio_path) | |
| resampler = torchaudio.transforms.Resample(sample_rate, resample_rate, dtype=waveform.dtype) | |
| resampled_waveform = resampler(waveform) | |
| torchaudio.save(output_audio_path, resampled_waveform, resample_rate) | |
| def save_audio(audio_input, output_dir="saved_audio", resample_rate=16000): | |
| if not os.path.exists(output_dir): | |
| os.makedirs(output_dir) | |
| sample_rate, audio_data = audio_input | |
| file_name = f"audio_{int(time.time())}.wav" | |
| file_path = os.path.join(output_dir, file_name) | |
| sf.write(file_path, audio_data, sample_rate) | |
| resampled_file_path = os.path.join(output_dir, f"resampled_{file_name}") | |
| save_and_resample_audio(file_path, resampled_file_path, resample_rate) | |
| return resampled_file_path | |
| def speech_to_text(audio_data, tgt_lang): | |
| file_path = save_audio(audio_data) | |
| audio_input, _ = torchaudio.load(file_path) | |
| s2t_model = torch.jit.load("unity_on_device.ptl", map_location=torch.device('cpu')) | |
| with torch.no_grad(): | |
| model_output = s2t_model(audio_input, tgt_lang=languages[tgt_lang]) | |
| transcribed_text = model_output[0] if model_output else "" | |
| print("Speech to Text Model Output:", transcribed_text) | |
| return transcribed_text | |
| def speech_to_speech_translation(audio_data, tgt_lang): | |
| file_path = save_audio(audio_data) | |
| audio_input, _ = torchaudio.load(file_path) | |
| s2st_model = torch.jit.load("unity_on_device.ptl", map_location=torch.device('cpu')) | |
| with torch.no_grad(): | |
| translated_text, units, waveform = s2st_model(audio_input, tgt_lang=languages[tgt_lang]) | |
| output_file = "/tmp/result.wav" | |
| torchaudio.save(output_file, waveform.unsqueeze(0), sample_rate=16000) | |
| print("Translated Text:", translated_text) | |
| print("Units:", units) | |
| print("Waveform Shape:", waveform.shape) | |
| return translated_text, output_file | |
| def create_interface(): | |
| with gr.Blocks(theme='ParityError/Anime') as interface: | |
| gr.Markdown(welcome_message) | |
| input_language = gr.Dropdown(list(languages.keys()), label="Select Target Language", value="English") | |
| with gr.Accordion("Speech to Text", open=False) as stt_accordion: | |
| audio_input_stt = gr.Audio(label="Upload or Record Audio") | |
| text_output_stt = gr.Text(label="Transcribed Text") | |
| stt_button = gr.Button("Transcribe") | |
| stt_button.click(speech_to_text, inputs=[audio_input_stt, input_language], outputs=text_output_stt) | |
| gr.Examples([["audio1.wav"]], inputs=[audio_input_stt], outputs=[text_output_stt]) | |
| with gr.Accordion("Speech to Speech Translation", open=False) as s2st_accordion: | |
| audio_input_s2st = gr.Audio(label="Upload or Record Audio") | |
| text_output_s2st = gr.Text(label="Translated Text") | |
| audio_output_s2st = gr.Audio(label="Translated Audio", type="filepath") | |
| s2st_button = gr.Button("Translate") | |
| s2st_button.click(speech_to_speech_translation, inputs=[audio_input_s2st, input_language], outputs=[text_output_s2st, audio_output_s2st]) | |
| gr.Examples([["audio1.wav"]], inputs=[audio_input_s2st], outputs=[text_output_s2st, audio_output_s2st]) | |
| return interface | |
| app = create_interface() | |
| app.launch(show_error=True, debug=True) |