Update app.py

app.py

@@ -1,35 +1,262 @@
-gradio
-torch
-torchvision
-torchaudio
-spaces
-imageio-ffmpeg
-# SoniTranslate Core Dependencies
-git+https://github.com/m-bain/whisperX.git
-pyannote.audio>=3.3.2
-fairseq
-yt-dlp
-pysrt
-pydub
-faster-whisper
-audiostretchy
-
-# Translation and TTS
-google-generativeai
-openai
-edge-tts
-piper-tts==1.2.0
-TTS==0.21.1
-
-# Other utilities
-# Important: numpy must be <2 for audio libraries to work
-numpy<2
-soundfile
-librosa
-onnxruntime-gpu
-tqdm
-demucs
-python-multipart
-tenacity
-youtube-transcript-api
-ffmpeg-python
+import os
+import sys
+import subprocess
+
+# --- FFmpeg Setup (Replaces packages.txt) ---
+try:
+    import imageio_ffmpeg
+    ffmpeg_path = imageio_ffmpeg.get_ffmpeg_exe()
+    ffmpeg_dir = os.path.dirname(ffmpeg_path)
+    # Add ffmpeg binary directory to system PATH so os.system("ffmpeg") works
+    os.environ["PATH"] += os.pathsep + ffmpeg_dir
+    # Ensure it's executable
+    subprocess.run(["chmod", "+x", ffmpeg_path])
+    print(f"✅ FFmpeg configured at: {ffmpeg_path}")
+except ImportError:
+    print("⚠️ imageio-ffmpeg not found. Please add it to requirements.txt")
+
+# --- Main Imports ---
+import gradio as gr
+import torch
+import spaces  # Required for ZeroGPU
+from soni_translate.logging_setup import logger, set_logging_level, configure_logging_libs
+configure_logging_libs()
+import whisperx
+from soni_translate.preprocessor import audio_video_preprocessor, audio_preprocessor
+from soni_translate.postprocessor import media_out, get_no_ext_filename, sound_separate, get_subtitle_speaker
+from soni_translate.speech_segmentation import transcribe_speech, align_speech, diarize_speech, ASR_MODEL_OPTIONS, find_whisper_models, diarization_models, COMPUTE_TYPE_CPU, COMPUTE_TYPE_GPU
+from soni_translate.translate_segments import translate_text, TRANSLATION_PROCESS_OPTIONS
+from soni_translate.text_to_speech import audio_segmentation_to_voice, edge_tts_voices_list, coqui_xtts_voices_list, piper_tts_voices_list
+from soni_translate.audio_segments import create_translated_audio, accelerate_segments
+from soni_translate.language_configuration import LANGUAGES, LANGUAGES_LIST
+from soni_translate.utils import remove_files, get_link_list, get_valid_files, is_audio_file, is_subtitle_file
+from soni_translate.text_multiformat_processor import process_subtitles, srt_file_to_segments, break_aling_segments
+from soni_translate.languages_gui import language_data
+import hashlib
+import json
+import copy
+from pydub import AudioSegment
+
+# Check for API key from Hugging Face Secrets
+if "GOOGLE_API_KEY" in os.environ:
+    print("✅ Google API Key found in secrets.")
+else:
+    print("⚠️ Google API Key not found. Please set it in the Space secrets.")
+
+if "OPENAI_API_KEY" in os.environ:
+    print("✅ OpenAI API Key found in secrets.")
+else:
+    print("⚠️ OpenAI API Key not found. Please set it in the Space secrets if you use OpenAI models.")
+
+
+# Create necessary directories
+directories = ["downloads", "logs", "weights", "clean_song_output", "_XTTS_", "audio", "outputs"]
+for directory in directories:
+    if not os.path.exists(directory):
+        os.makedirs(directory)
+
+class SoniTranslate:
+    def __init__(self):
+        # Device detection moved inside the function for ZeroGPU compatibility
+        self.result_diarize = None
+        self.align_language = None
+        self.result_source_lang = None
+        self.tts_info = self._get_tts_info()
+
+    def _get_tts_info(self):
+        # Simplified for this example
+        class TTS_Info:
+            def tts_list(self):
+                try:
+                    return edge_tts_voices_list()
+                except Exception as e:
+                    logger.warning(f"Could not get Edge-TTS voices: {e}")
+                    return ["en-US-JennyNeural-Female"] # fallback
+        return TTS_Info()
+
+    # --- ZeroGPU Decorator ---
+    # duration=300 means 5 minutes max per request. Adjust if needed.
+    @spaces.GPU(duration=300) 
+    def multilingual_media_conversion(
+        self,
+        media_file,
+        link_media,
+        directory_input,
+        origin_language,
+        target_language,
+        tts_voice,
+        transcriber_model,
+        max_speakers,
+        is_gui=True,
+        progress=gr.Progress(),
+    ):
+        # Check device inside the GPU decorated function
+        self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        logger.info(f"Working on device: {self.device}")
+
+        try:
+            progress(0.05, desc="Starting process...")
+            
+            # 1. Handle Input
+            input_media = None
+            if media_file is not None:
+                input_media = media_file.name
+            elif link_media:
+                input_media = link_media
+            elif directory_input and os.path.exists(directory_input):
+                input_media = directory_input
+            
+            if not input_media:
+                raise ValueError("No input media specified. Please upload a file or provide a URL.")
+
+            base_audio_wav = "audio.wav"
+            base_video_file = "video.mp4"
+
+            remove_files(base_audio_wav, base_video_file)
+
+            progress(0.1, desc="Processing input media...")
+            if is_audio_file(input_media):
+                audio_preprocessor(False, input_media, base_audio_wav)
+            else:
+                audio_video_preprocessor(False, input_media, base_video_file, base_audio_wav)
+
+            # 2. Transcription
+            progress(0.25, desc="Transcribing audio with WhisperX...")
+            source_lang_code = LANGUAGES[origin_language] if origin_language != "Automatic detection" else None
+            
+            # Force float16 if cuda is available (ZeroGPU)
+            compute_type = "float16" if self.device == "cuda" else "int8"
+
+            audio, result = transcribe_speech(
+                base_audio_wav,
+                transcriber_model,
+                compute_type,
+                16,
+                source_lang_code
+            )
+            
+            progress(0.4, desc="Aligning transcription...")
+            self.align_language = result["language"]
+            result = align_speech(audio, result)
+
+            # 3. Diarization
+            progress(0.5, desc="Separating speakers...")
+            hf_token = os.environ.get("HF_TOKEN") 
+            if not hf_token:
+                logger.warning("Hugging Face token not found. Diarization might fail.")
+            
+            self.result_diarize = diarize_speech(
+                base_audio_wav,
+                result,
+                1,
+                max_speakers,
+                hf_token,
+                diarization_models["pyannote_3.1"]
+            )
+            self.result_source_lang = copy.deepcopy(self.result_diarize)
+
+            # 4. Translation
+            progress(0.6, desc="Translating text...")
+            translate_to_code = LANGUAGES[target_language]
+            self.result_diarize["segments"] = translate_text(
+                self.result_diarize["segments"],
+                translate_to_code,
+                "google_translator_batch", 
+                chunk_size=1800,
+                source=self.align_language,
+            )
+
+            # 5. Text-to-Speech
+            progress(0.75, desc="Generating dubbed audio...")
+            valid_speakers = audio_segmentation_to_voice(
+                self.result_diarize,
+                translate_to_code,
+                is_gui,
+                tts_voice
+            )
+
+            # 6. Audio Processing & Merging
+            progress(0.85, desc="Synchronizing and mixing audio...")
+            dub_audio_file = "audio_dub_solo.ogg"
+            remove_files(dub_audio_file)
+            audio_files, _ = accelerate_segments(self.result_diarize, 1.8, valid_speakers)
+            create_translated_audio(self.result_diarize, audio_files, dub_audio_file, False, False)
+            
+            mix_audio_file = "audio_mix.mp3"
+            remove_files(mix_audio_file)
+            
+            # Using os.system which relies on the PATH set at the top
+            command_volume_mix = f'ffmpeg -y -i {base_audio_wav} -i {dub_audio_file} -filter_complex "[0:0]volume=0.1[a];[1:0]volume=1.5[b];[a][b]amix=inputs=2:duration=longest" -c:a libmp3lame {mix_audio_file}'
+            os.system(command_volume_mix)
+
+            # 7. Final Video Creation
+            progress(0.95, desc="Creating final video...")
+            output_filename = "video_dub.mp4"
+            remove_files(output_filename)
+            
+            if os.path.exists(base_video_file):
+                os.system(f"ffmpeg -i {base_video_file} -i {mix_audio_file} -c:v copy -c:a copy -map 0:v -map 1:a -shortest {output_filename}")
+                final_output = media_out(input_media, translate_to_code, "", "mp4", file_obj=output_filename)
+            else: 
+                final_output = media_out(input_media, translate_to_code, "", "mp3", file_obj=mix_audio_file)
+
+            progress(1.0, desc="Done!")
+            return final_output
+
+        except Exception as e:
+            logger.error(f"An error occurred: {e}")
+            gr.Error(f"An error occurred: {e}")
+            return None
+
+# Instantiate the class
+SoniTr = SoniTranslate()
+
+# Create Gradio Interface
+with gr.Blocks(theme="Taithrah/Minimal") as app:
+    gr.Markdown("<center><h1>📽️ ابزار دوبله ویدیو با هوش مصنوعی 🈷️</h1></center>")
+    gr.Markdown("ساخته شده توسط [aigolden](https://youtube.com/@aigolden) - بر پایه [SoniTranslate](https://github.com/r3gm/SoniTranslate)")
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("### ۱. ورودی ویدیو")
+            video_file_input = gr.File(label="آپلود ویدیو")
+            link_media_input = gr.Textbox(label="یا لینک یوتیوب", placeholder="https://www.youtube.com/watch?v=...")
+            
+            gr.Markdown("### ۲. تنظیمات دوبله")
+            origin_language_input = gr.Dropdown(LANGUAGES_LIST, value="Automatic detection", label="زبان اصلی ویدیو")
+            target_language_input = gr.Dropdown(LANGUAGES_LIST[1:], value="Persian (fa)", label="زبان مقصد دوبله")
+            tts_voice_input = gr.Dropdown(SoniTr.tts_info.tts_list(), value="fa-IR-FaridNeural", label="صدای گوینده")
+            
+            with gr.Accordion("تنظیمات پیشرفته", open=False):
+                transcriber_model_input = gr.Dropdown(
+                    ASR_MODEL_OPTIONS + find_whisper_models(), 
+                    value="large-v3", 
+                    label="مدل استخراج متن (Whisper)",
+                    info="مدل‌های بزرگتر دقیق‌تر اما کندتر هستند."
+                )
+                max_speakers_input = gr.Slider(1, 10, value=2, step=1, label="حداکثر تعداد گوینده")
+
+            process_button = gr.Button("شروع دوبله", variant="primary")
+
+        with gr.Column():
+            gr.Markdown("### ۳. خروجی")
+            output_video = gr.Video(label="ویدیوی دوبله شده")
+            output_file = gr.File(label="دانلود فایل")
+
+    process_button.click(
+        SoniTr.multilingual_media_conversion,
+        inputs=[
+            video_file_input,
+            link_media_input,
+            gr.Textbox(visible=False), 
+            origin_language_input,
+            target_language_input,
+            tts_voice_input,
+            transcriber_model_input,
+            max_speakers_input,
+        ],
+        outputs=[output_file]
+    )
+
+if __name__ == "__main__":
+    app.launch(server_name="0.0.0.0", server_port=7860)