intermediate commit

.gitignore

@@ -1,5 +1,5 @@
-.env
+.env
-__pycache__/
+__pycache__/
-outputs/*
+outputs/*
-venv/
+venv/
-
+

app.py

@@ -5,13 +5,11 @@ from config import *
 
 # Подгрузка сервисов
 from services.llm_factory import get_llm_provider
-
+from services.fasterWhisper import FasterWhisper
-
 
 # Загрузка доп. модулей
 from handlers.gradioHandler import GradioHandlers
 from handlers.fileHandlers import FileHandlers
-from services.fasterWhisper import FasterWhisper
 from handlers.convertMdToPdf import ConvertMdToPdf
 from handlers.glueAudio import GlueAudio
 
@@ -115,13 +113,20 @@ def main():
         saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filename)
         saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filenamePdf)
 
-        recognizeBtn.click(gh.handleRecognizeBtn, outputs=[recognizedText], inputs=[audioFiles, fastWhisperModel, device, compute_type, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText, gr.State(GLUED_AUDIO_FILENAME), gr.State(OUTPUT_PATH)])
+        recognizeBtn.click(
+            gh.handleRecognizeBtn, 
+            inputs=[audioFiles, fastWhisperModel, device, compute_type, beamSize, 
+                    vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, 
+                    wordTimestamps, noSpeechThreshold, conditionOnPreviousText, gr.State(GLUED_AUDIO_FILENAME), gr.State(OUTPUT_PATH)],
+            outputs=[recognizedText], 
+            )
         
         # Если пайплайн включен то тогда делаем автоматически
         # автоматический пайплайн
         recognizedText.change(
             gh.generateByCondition,
-            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
+            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, 
+                    isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
             outputs=[refinedText, refinedTextMD]
         )
         
@@ -131,11 +136,12 @@ def main():
         llmProvider.change(
             gh.update_model_dropdown, 
             inputs=llmProvider, 
-            outputs=llmModel
+            outputs=[llmModel, apiKey]
         )
         refineTextBtn.click(
             gh.generateByCondition,
-            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
+            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, 
+                    isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
             outputs=[refinedText, refinedTextMD]
         )
 

config.py

@@ -8,7 +8,10 @@ DEVICES = ['cpu', 'cuda']
 COMPUTE_TYPE = ['auto', 'int8', 'float16', 'float32']
 
 # Стандартный API ключ
-DEFAULT_API_KEY=os.getenv('API_KEY')
+IO_API_KEY=os.getenv('IO_API_KEY')
+GEMINI_API_KEY=os.getenv('GEMINI_API_KEY')
+DEFAULT_API_KEY=IO_API_KEY
+
 
 # Словарь провайдеров и их моделей
 LLM_PROVIDERS = ['io.net', 'Gemini', 'gpt4free']
@@ -46,42 +49,38 @@ OUTPUT_PATH='outputs'
 
 GLUED_AUDIO_FILENAME='glued.mp3'
 
-DEFAULT_SYSTEM_PROMPT='''
+DEFAULT_SYSTEM_PROMPT='''You are a diligent university student who has recorded a lecture as an audio file and later transcribed it into raw text. 
-You are a smart university student creating easy-to-understand study notes summary of lesson  for a classmate who is a beginner. Your source is a raw text/audio transcript.
+Your task is to rewrite this unstructured transcript into a clear, logically organized, and detailed lecture summary (lecture notes).  
 
-GOAL: rewrite the information into a clear, structured summary in RUSSIAN.
+Guidelines:
+1. Structure:
+   - Organize the text into a hierarchy of sections and subsections.  
+   - Use headings, bullet points, or numbering where appropriate.  
+   - Present the material in a logical flow (from introduction → main points → details → examples → conclusion).  
 
-KEY RULES FOR CONTENT:
+2. Clarity & Cohesion:
-1.  **Logical Structure:** Use Markdown headers (#, ##), bullet points, and short paragraphs.
+   - Remove filler words, repetitions, and irrelevant fragments.  
-2.  **No "Water":** Remove filler words. Keep only practical information.
+   - Rewrite incomplete sentences into full, grammatically correct sentences.  
-3.  **Student Tone:** Write naturally, as if sharing notes with a friend. Avoid robotic phrases like "It is important to note".
+   - Ensure smooth transitions between topics, making the summary feel continuous and well-connected.  
 
-KEY RULES FOR LATEX (CRITICAL FOR PYLATEXENC):
+3. Depth & Detail: 
-1.  **Math Mode:** ANY variable (like t, L, C), number in a formula, or equation MUST be wrapped in dollar signs `$`.
+   - Capture all important concepts, definitions, examples, and explanations from the lecture.  
-    *   BAD: i(t) = i_pr + i_sv
+   - Expand shorthand or fragmented thoughts into full, precise explanations.  
-    *   GOOD: $i(t) = i_{pr} + i_{sv}$
+   - Where appropriate, rephrase or clarify confusing passages for better understanding.  
-2.  **Subscripts:** Always use curly braces `{}` for subscripts longer than one character.
-    *   BAD: $i_pr$
-    *   GOOD: $i_{pr}$ (or $i_{пр}$ if using cyrillic)
-3.  **Symbols:** Use standard LaTeX commands for symbols.
-    *   Arrow: use `\to` (e.g., $t \to \infty$).
-    *   Infinity: use `\infty`.
-    *   Multiplication: use `\cdot` or just space.
-4.  **Consistency:** Never leave a mathematical symbol as plain text. If you mention "current i", write "ток $i$".
 
-EXAMPLE OF DESIRED OUTPUT FORMAT:
+4. Accuracy:
-# Тема лекции
+   - Preserve the lecturer’s original meaning, intent, and terminology.  
-## Основные понятия
+   - Avoid adding personal opinions or new information that was not in the lecture.  
-*   **Переходный процесс** — это когда цепь перестраивается с одного режима на другой (например, щелкнули выключателем).
-*   Математически это описывается дифференциальными уравнениями. Порядок уравнения = количеству реактивных элементов ($L$ и $C$).
 
-## Классический метод
+5. Style:
-Решение ищется в виде суммы двух частей:
+   - Write in a formal, academic tone suitable for study notes.  
-$$i(t) = i_{pr} + i_{sv}$$
+   - Aim for readability: concise sentences, but thorough coverage of concepts.  
+   - Use emphasis (e.g., bold or italic text) only when it improves comprehension.  
 
-1.  **Принужденная составляющая** ($i_{pr}$) — это режим, который установится в будущем, когда все успокоится ($t \to \infty$).
+Final Output: A cohesive, detailed, and well-structured lecture summary, suitable for later studying and revision.
-2.  **Свободная составляющая** ($i_{sv}$) — это то, что происходит "само по себе" из-за энергии, запасенной в $L$ и $C$.
+Use only russian language!
-
+USE LATEX IN DOLLAR SIGN ($)!
-***
+EXTRA BIG LENTH OF CONSPECT!
-STRICTLY FOLLOW THESE FORMATTING RULES. OUTPUT IN RUSSIAN.
+MAKE AS LONG AS POSIBLE AND AS BE GOOD!
 '''
+

handlers/glueAudio.py

@@ -1,11 +1,61 @@
-from pydub import AudioSegment
+import subprocess
+from pathlib import Path
+
 
 class GlueAudio():
-    def glue(self, audioFiles):
+    def glue(self, audio_files: list, output_path: str, output_filename: str) -> Path:
-        glued = AudioSegment.empty()
+        """
+        Склеивает аудиофайлы с помощью FFmpeg, используя промежуточный список файлов.
+        Этот метод чрезвычайно эффективен по памяти и скорости.
 
-        for audioFile in audioFiles:
+        Args:
-            audio = AudioSegment.from_file(audioFile)
+            audio_files (list): Список путей к исходным аудиофайлам.
-            glued += audio
+            output_path (str): Директория для сохранения итогового файла.
+            output_filename (str): Имя итогового склеенного файла.
 
-        return glued
+        Returns:
+            Path: Путь к созданному склеенному файлу.
+        """
+        output_dir = Path(output_path)
+        output_dir.mkdir(parents=True, exist_ok=True)
+        final_audio_path = output_dir / output_filename
+
+        if not audio_files:
+            raise ValueError("Список аудиофайлов для склейки пуст.")
+
+        # 1. Формируем часть команды с входными файлами (-i file1 -i file2 ...)
+        input_args = []
+        for file_path in audio_files:
+            input_args.extend(['-i', str(Path(file_path).resolve())])
+
+        # 2. Формируем строку для filter_complex
+        num_files = len(audio_files)
+        stream_specifiers = "".join([f"[{i}:a]" for i in range(num_files)])
+        filter_complex_str = f"{stream_specifiers}concat=n={num_files}:v=0:a=1[outa]"
+
+        # 3. Собираем полную команду
+        command = [
+            'ffmpeg',
+            *input_args,  # Распаковываем список входных файлов
+            '-filter_complex', filter_complex_str,
+            '-map', '[outa]',
+            '-c:a', 'libmp3lame',
+            '-q:a', '2',
+            str(final_audio_path),
+            '-y'
+        ]
+
+        try:
+            # 4. Выполняем команду
+            print(f"Выполнение команды FFmpeg: {' '.join(command)}")
+            subprocess.run(command, check=True, capture_output=True, text=True)
+            print("FFmpeg успешно завершил склейку.")
+
+        except FileNotFoundError:
+            raise FileNotFoundError("FFmpeg не найден. Убедитесь, что он установлен и доступен в системной переменной PATH.")
+        except subprocess.CalledProcessError as e:
+            print("Ошибка при выполнении FFmpeg!")
+            print("Stderr:", e.stderr)
+            raise RuntimeError(f"Ошибка FFmpeg при склейке файлов: {e.stderr}")
+        
+        return final_audio_path

handlers/gradioHandler.py

@@ -1,6 +1,6 @@
 from config import LLM_MODELS # Импортируем словарь моделей
 import gradio as gr 
-
+from config import GEMINI_API_KEY, IO_API_KEY
 
 class GradioHandlers:
     def __init__(self, llm_factory, ConvertMdToPdf, FileHandlers, FasterWhisper, GlueAudio):
@@ -9,23 +9,39 @@ class GradioHandlers:
         self.ga = GlueAudio()
         self.ConvertMdToPdf = ConvertMdToPdf()
         self.FasterWhisper = FasterWhisper()
-        self.llm_factory = llm_factory # Сохраняем фабрику
+        self.llm_factory = llm_factory
 
-    def handleRecognizeBtn(self, audioFiles, model, device, compute_type, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText, filename, outPath):
+    def handleRecognizeBtn(
-        audioFile = self.ga.glue(audioFiles)
+            self, audioFiles, model, device, compute_type, beamSize, vadFilter, 
-        file = self.fh.saveFile(filename, audioFile, outPath)
+            minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, 
+            wordTimestamps, noSpeechThreshold, conditionOnPreviousText, filename, outPath
+            ):
+        try:
+            glued_audio_path = self.ga.glue(
+                audio_files=[f.name for f in audioFiles], # Передаем список путей
+                output_path=outPath,
+                output_filename=filename
+            )
+        except (FileNotFoundError, RuntimeError) as e:
+            # Если FFmpeg не найден или произошла ошибка, сообщаем пользователю
+            gr.Warning(str(e))
+            return "" # Возвращаем пустую строку в текстовое поле
 
-        return self.FasterWhisper.recognize(model, device, compute_type, file, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText)
+        # Передаем путь к склеенному файлу в FasterWhisper
+        return self.FasterWhisper.recognize(model, device, compute_type, str(glued_audio_path), beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText)
 
     # Функция улучшения текста
-    def generateByCondition(self, api_key, llm_provider, llm_model, system_prompt, recognized_text, llm_temperature, is_pipeline_enabled, trigger, isSaveFile, filename, filenamePdf, output_path):
+    def generateByCondition(self, api_key, llm_provider, 
+                            llm_model, system_prompt, recognized_text, 
+                            llm_temperature, is_pipeline_enabled, trigger, 
+                            isSaveFile, filename, filenamePdf, output_path):
         try:
             # Получаем нужный провайдер через фабрику
             provider = self.llm_factory(llm_provider, api_key)
         except ValueError as e:
             # Если API ключ не предоставлен для нужного провайдера, выводим ошибку
-            self.gr.Warning(str(e))
+            gr.Warning(str(e))
-            return self.gr.skip(), self.gr.skip()
+            return gr.skip(), gr.skip()
 
         def process():
             result, md = provider.generate(llm_model, system_prompt, recognized_text, llm_temperature)
@@ -53,7 +69,9 @@ class GradioHandlers:
         default_value = models[0] if models else None
         
         # Возвращаем обновленный компонент. Используем 'gr' напрямую.
-        return gr.update(choices=models, value=default_value)
+        if provider == 'io.net': return gr.update(choices=models, value=default_value), gr.update(label='API key (required for io.net, Gemini)', value=IO_API_KEY, interactive=True)
+        if provider == 'Gemini': return gr.update(choices=models, value=default_value), gr.update(label='API key (required for Oio.net, Gemini)', value=GEMINI_API_KEY, interactive=True)
+        if provider == 'gpt4free': return gr.update(choices=models, value=default_value), gr.update(label='API key (required for Oio.net, Gemini)', value="", interactive=True)
 
      # Функция для динамического обновления кнопки
     def updateButton(self, isChecked):

requirements.txt

@@ -100,3 +100,118 @@ uvicorn==0.35.0
 webencodings==0.5.1
 websockets==15.0.1
 zopfli==0.2.3.post1
+aiofiles==24.1.0
+aiohappyeyeballs==2.6.1
+aiohttp==3.13.0
+aiosignal==1.4.0
+annotated-types==0.7.0
+anyio==4.10.0
+attrs==25.4.0
+audioop-lts==0.2.2
+av==15.1.0
+beautifulsoup4==4.13.5
+Brotli==1.1.0
+bs4==0.0.2
+certifi==2025.8.3
+cffi==2.0.0
+charset-normalizer==3.4.3
+click==8.2.1
+coloredlogs==15.0.1
+colour==0.1.5
+cssselect2==0.8.0
+ctranslate2==4.6.0
+distro==1.9.0
+dotenv==0.9.9
+exceptiongroup==1.3.0
+fastapi==0.116.1
+faster-whisper==1.2.0
+ffmpeg-python==0.2.0
+ffmpy==0.6.1
+filelock==3.19.1
+flatbuffers==25.2.10
+flatlatex==0.15
+fonttools==4.59.2
+frozenlist==1.8.0
+fsspec==2025.9.0
+future==1.0.0
+g4f==0.6.3.5
+gradio==5.44.1
+gradio_client==1.12.1
+groovy==0.1.2
+h11==0.16.0
+hf-xet==1.1.9
+httpcore==1.0.9
+httpx==0.28.1
+huggingface-hub==0.34.4
+humanfriendly==10.0
+idna==3.10
+iso639-lang==2.6.3
+Jinja2==3.1.6
+jiter==0.10.0
+joblib==1.5.2
+langdetect==1.0.9
+littleutils==0.2.4
+markdown-it-py==3.0.0
+markdown_pdf==1.9
+MarkupSafe==3.0.2
+mdurl==0.1.2
+mpmath==1.3.0
+multidict==6.7.0
+nest-asyncio==1.6.0
+nltk==3.9.1
+numpy==2.2.6
+onnxruntime==1.22.1
+openai==1.106.1
+orjson==3.11.3
+outdated==0.2.2
+packaging==25.0
+pandas==2.3.2
+pillow==11.3.0
+propcache==0.4.0
+protobuf==6.32.0
+pycparser==2.22
+pycryptodome==3.23.0
+pydantic==2.11.7
+pydantic_core==2.33.2
+pydub==0.25.1
+pydyf==0.11.0
+Pygments==2.19.2
+pylatexenc==2.10
+pymultidictionary==1.3.2
+PyMuPDF==1.26.4
+pyperclip==1.9.0
+pyphen==0.17.2
+python-dateutil==2.9.0.post0
+python-dotenv==1.1.1
+python-multipart==0.0.20
+pytz==2025.2
+PyYAML==6.0.2
+regex==2025.9.1
+requests==2.32.5
+rich==14.1.0
+ruff==0.12.12
+safehttpx==0.1.6
+semantic-version==2.10.0
+setuptools==80.9.0
+shellingham==1.5.4
+six==1.17.0
+sniffio==1.3.1
+soupsieve==2.8
+starlette==0.47.3
+sympy==1.14.0
+tinycss2==1.4.0
+tinyhtml5==2.0.0
+tkmacosx==1.0.5
+tokenizers==0.22.0
+tomlkit==0.13.3
+tqdm==4.67.1
+typer==0.17.4
+typing-inspection==0.4.1
+typing_extensions==4.15.0
+tzdata==2025.2
+urllib3==2.5.0
+uvicorn==0.35.0
+webencodings==0.5.1
+websockets==15.0.1
+yarl==1.22.0
+zopfli==0.2.3.post1

services/fasterWhisper.py

@@ -1,7 +1,11 @@
 from faster_whisper import WhisperModel
 
 class FasterWhisper:
-    def recognize(self, model, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText):
+    def recognize(self, model, device, compute_type, 
+                  audioFile, beamSize, vadFilter, 
+                  minSilenceDurationMs, speechPadMs, 
+                  temp0, temp1, temp2, wordTimestamps, 
+                  noSpeechThreshold, conditionOnPreviousText):
         model = WhisperModel(model, device=device, compute_type=compute_type) # Задаем модель
 
         segments, _ = model.transcribe( # Распознаем текст
@@ -22,6 +26,7 @@ class FasterWhisper:
 
         for seg in segments:
             text += f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}" + '\n'
+            print(f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}")
         
         return text
 

test.py

@@ -0,0 +1,19 @@
+from handlers.convertMdToPdf import ConvertMdToPdf
+import re
+# Создаем экземпляр класса
+converter = ConvertMdToPdf()
+
+# Тестовые примеры дробей
+test_cases = [
+    r"$\frac{1}{2}$",          # простая дробь
+    r"$\dfrac{3}{4}$",         # дробь с displaystyle
+    r"$\frac{a}{b} + \frac{c}{d}$",  # сложение дробей
+    r"$$\frac{x^2}{y^3}$$",    # блочная дробь
+    r"$\frac{\partial f}{\partial x}$"  # частная производная
+]
+
+for latex in test_cases:
+    result = converter.replace_math(re.search(r'\$\$(.*?)\$\$|\$(.*?)\$', latex))
+    print(f"Input: {latex}")
+    print(f"Output: {result}")
+    print("---")