Spaces:
Running
Running
| import gradio as gr | |
| import torch | |
| import torchaudio | |
| import scipy.io.wavfile | |
| import numpy as np | |
| from transformers import AutoProcessor, SeamlessM4Tv2Model | |
| from pathlib import Path | |
| from typing import Optional, Union | |
| class SeamlessTranslator: | |
| def __init__(self, model_name: str = "facebook/seamless-m4t-v2-large"): | |
| try: | |
| self.processor = AutoProcessor.from_pretrained(model_name) | |
| self.model = SeamlessM4Tv2Model.from_pretrained(model_name) | |
| self.sample_rate = self.model.config.sampling_rate | |
| except Exception as e: | |
| raise RuntimeError(f"Failed to initialize model: {str(e)}") | |
| # Available language pairs | |
| self.language_codes = { | |
| "English": "eng", | |
| "Spanish": "spa", | |
| "French": "fra", | |
| "German": "deu", | |
| "Italian": "ita", | |
| "Portuguese": "por", | |
| "Russian": "rus", | |
| "Chinese": "cmn", | |
| "Japanese": "jpn", | |
| "Korean": "kor", | |
| "Arabic": "ara", | |
| "Hindi": "hin", | |
| } | |
| def translate_text(self, text: str, src_lang: str, tgt_lang: str) -> tuple[int, np.ndarray]: | |
| try: | |
| inputs = self.processor(text=text, src_lang=src_lang, return_tensors="pt") | |
| audio_array = self.model.generate(**inputs, tgt_lang=tgt_lang)[0].cpu().numpy().squeeze() | |
| return self.sample_rate, audio_array | |
| except Exception as e: | |
| raise RuntimeError(f"Text translation failed: {str(e)}") | |
| def translate_audio(self, audio_path: str, tgt_lang: str) -> tuple[int, np.ndarray]: | |
| try: | |
| # Load and resample audio | |
| audio, orig_freq = torchaudio.load(audio_path) | |
| audio = torchaudio.functional.resample( | |
| audio, | |
| orig_freq=orig_freq, | |
| new_freq=16_000 | |
| ) | |
| # Process and generate translation | |
| inputs = self.processor(audios=audio, return_tensors="pt") | |
| audio_array = self.model.generate(**inputs, tgt_lang=tgt_lang)[0].cpu().numpy().squeeze() | |
| return self.sample_rate, audio_array | |
| except Exception as e: | |
| raise RuntimeError(f"Audio translation failed: {str(e)}") | |
| class GradioInterface: | |
| def __init__(self): | |
| self.translator = SeamlessTranslator() | |
| self.languages = list(self.translator.language_codes.keys()) | |
| def text_to_speech(self, text: str, src_lang: str, tgt_lang: str) -> tuple[int, np.ndarray]: | |
| src_code = self.translator.language_codes[src_lang] | |
| tgt_code = self.translator.language_codes[tgt_lang] | |
| return self.translator.translate_text(text, src_code, tgt_code) | |
| def speech_to_speech(self, audio_path: str, tgt_lang: str) -> tuple[int, np.ndarray]: | |
| tgt_code = self.translator.language_codes[tgt_lang] | |
| return self.translator.translate_audio(audio_path, tgt_code) | |
| def launch(self): | |
| # Create the Gradio interface | |
| with gr.Blocks(title="SeamlessM4T Translator") as demo: | |
| gr.Markdown("# 🌐 SeamlessM4T Translator") | |
| gr.Markdown("Translate text or speech to different languages using Meta's SeamlessM4T model") | |
| with gr.Tabs(): | |
| # Text-to-Speech tab | |
| with gr.TabItem("Text to Speech"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| text_input = gr.Textbox( | |
| label="Input Text", | |
| placeholder="Enter text to translate...", | |
| lines=3 | |
| ) | |
| src_lang = gr.Dropdown( | |
| choices=self.languages, | |
| value="English", | |
| label="Source Language" | |
| ) | |
| tgt_lang_text = gr.Dropdown( | |
| choices=self.languages, | |
| value="Spanish", | |
| label="Target Language" | |
| ) | |
| translate_btn = gr.Button("Translate", variant="primary") | |
| with gr.Column(): | |
| audio_output = gr.Audio( | |
| label="Translated Speech", | |
| type="numpy" | |
| ) | |
| translate_btn.click( | |
| fn=self.text_to_speech, | |
| inputs=[text_input, src_lang, tgt_lang_text], | |
| outputs=audio_output | |
| ) | |
| # Speech-to-Speech tab | |
| with gr.TabItem("Speech to Speech"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| audio_input = gr.Audio( | |
| label="Input Speech", | |
| type="filepath" | |
| ) | |
| tgt_lang_speech = gr.Dropdown( | |
| choices=self.languages, | |
| value="Spanish", | |
| label="Target Language" | |
| ) | |
| translate_audio_btn = gr.Button("Translate", variant="primary") | |
| with gr.Column(): | |
| audio_output_s2s = gr.Audio( | |
| label="Translated Speech", | |
| type="numpy" | |
| ) | |
| translate_audio_btn.click( | |
| fn=self.speech_to_speech, | |
| inputs=[audio_input, tgt_lang_speech], | |
| outputs=audio_output_s2s | |
| ) | |
| gr.Markdown( | |
| """ | |
| ### Notes | |
| - Text-to-Speech: Enter text and select source/target languages | |
| - Speech-to-Speech: Upload an audio file and select target language | |
| - Processing may take a few moments depending on input length | |
| """ | |
| ) | |
| # Launch the interface | |
| demo.launch(share=True) | |
| if __name__ == "__main__": | |
| interface = GradioInterface() | |
| interface.launch() |