fix download dataset

.gitignore

@@ -8,6 +8,9 @@ __pycache__/
 # Custom
 data/model_small/
 data/model_large/
+data/v4_ru.pt
+MyTTSDataset/
+vocal.wav
 
 # C extensions
 *.so

data/commands.yaml

@@ -76,5 +76,6 @@ home_assistant_execute:
   - включи телевизор
   - выключи телевизор
   - начни уборку
+  - убрать мою комнату
 home_assistant_get:
   - тест

modules/Jarvis.py

@@ -11,7 +11,7 @@ from fuzzywuzzy import fuzz
 from pvrecorder import PvRecorder
 
 from data import config
-from modules import HomeAssistant, MediaPlayerController
+from modules import HomeAssistant
 from utils import download_models, execute_cmd, play
 
 
@@ -25,7 +25,6 @@ class Jarvis:
         self.CDIR = os.getcwd()
         self.VA_CMD_LIST = yaml.safe_load(open('data/commands.yaml', encoding='utf8'))
         self.home_assistant = HomeAssistant.HomeAssistant()
-        self.media_player_controller = MediaPlayerController.MediaPlayerController()
         self.porcupine = pvporcupine.create(
             access_key=config.PICOVOICE_TOKEN,
             keywords=['jarvis'],
@@ -74,7 +73,7 @@ class Jarvis:
         :return: bool - распознана или нет команда
         """
         print(f"Распознано: {voice}")
-        for x in config.VA_ALIAS:
+        for x in config.VA_ALIAS + config.VA_TBR:
             voice = voice.replace(x, "").strip()
         rc = {'cmd': '', 'percent': 0}
         for c, v in self.VA_CMD_LIST.items():

modules/MediaPlayerController.py

@@ -1,90 +0,0 @@
-import platform
-import subprocess
-
-
-class MediaPlayerController:
-    """
-    Модуль для манипуляции музыкой
-    """
-    def __init__(self):
-        self.os_type = platform.system()
-
-    def play_pause(self) -> None:
-        """
-        Запуск/остановка музыки
-
-        :return:
-        """
-        if self.os_type == 'Windows':
-            self._windows_play_pause()
-        elif self.os_type == 'Linux':
-            self._linux_control("play-pause")
-
-    def next_track(self) -> None:
-        """
-        Включает следующею композицию
-
-        :return:
-        """
-        if self.os_type == 'Windows':
-            self._windows_control("next")
-        elif self.os_type == 'Linux':
-            self._linux_control("next")
-
-    def previous_track(self) -> None:
-        """
-        Включает предыдущею композицию
-
-        :return:
-        """
-        if self.os_type == 'Windows':
-            self._windows_control("previous")
-        elif self.os_type == 'Linux':
-            self._linux_control("previous")
-
-    def _windows_play_pause(self) -> None:
-        """
-        Запуск/остановка музыки в windows
-
-        :return:
-        """
-        import win32con
-        self.key_press(win32con.VK_MEDIA_PLAY_PAUSE)
-
-    def _windows_control(self, action: str) -> None:
-        """
-        Включает предыдущею или следующею композицию в windows
-
-        :return:
-        """
-        import win32con
-        if action == "next":
-            self.key_press(win32con.VK_MEDIA_NEXT_TRACK)
-        elif action == "previous":
-            self.key_press(win32con.VK_MEDIA_PREV_TRACK)
-
-    @staticmethod
-    def key_press(key_code: str) -> None:
-        """
-        Симуляция нажатия и отпускания клавиши
-
-        :param key_code: str - какую кнопку нажать
-        :return:
-        """
-        import win32api
-        import win32con
-        win32api.keybd_event(key_code, 0, 0, 0)
-        win32api.keybd_event(key_code, 0, win32con.KEYEVENTF_KEYUP, 0)
-
-    @staticmethod
-    def _linux_control(command: str) -> None:
-        """
-        Запускает команду для linux систем
-
-        :param command: str - команда для запуска
-        :return:
-        """
-        try:
-            subprocess.run(["playerctl", command], check=True)
-        except subprocess.CalledProcessError as e:
-            print(f"Failed to {command}: {e}")

pyproject.toml

@@ -6,7 +6,7 @@ authors = ["dmitrium12 <belicdima8@gmail.com>"]
 readme = "README.md"
 
 [tool.poetry.dependencies]
-python = "^3.11"
+python = ">=3.11,<3.12"
 vosk = "^0.3.45"
 pvporcupine = "^3.0.1"
 pvrecorder = "^1.2.1"
@@ -28,6 +28,8 @@ ollama = "^0.1.6"
 ruff = "^0.4.2"
 noisereduce = "^3.0.2"
 environs = "^11.0.0"
+webrtcvad = "^2.0.10"
+tts = "^0.22.0"
 
 
 [[tool.poetry.source]]

test.py

@@ -0,0 +1,55 @@
+import re
+
+import requests
+from bs4 import BeautifulSoup
+
+
+def filter_string(input_string: str) -> str:
+    allowed_chars = []
+    for j in "АБВГДЕЁЖЗИЙКЛМНОПРСТУФХЦЧШЩЪЫЬЭЮЯабвгдеёжзийклмнопрстуфхцчшщъыьэюя1234567890 !,.?-":
+        allowed_chars.append(j)
+    input_string = re.sub(r'^\d+.\s+', '', input_string)
+    return ''.join([char for char in input_string if char in allowed_chars])
+
+
+repetition = 0
+response = {}
+soup = BeautifulSoup(
+    requests.get('https://theportalwiki.com/wiki/GLaDOS_voice_lines/ru').text,
+    features='html.parser'
+)
+for li in soup.find_all('li'):
+    try:
+        i = li.find('i').text
+        url = li.find('span', class_=['audio-player']).find('a')['href']
+        if i not in response.keys():
+            response[i] = url
+        else:
+            repetition += 1
+    except AttributeError:
+        pass
+    try:
+        i = li.find('a').text
+        url = li.find('a')['href']
+        if i not in response.keys():
+            response[i] = url
+        else:
+            repetition += 1
+    except AttributeError:
+        pass
+print(f'Количество найденный элементов: {len(response)}')
+print(f'Количество повторении: {repetition}')
+with open('MyTTSDataset/transcript.txt', 'w') as f:
+    for index, (key, value) in enumerate(response.items()):
+        try:
+            response = requests.get(value)
+            if response.status_code == 200:
+                key = filter_string(key)
+                if key and len(key.replace(" ", "")) > 3:
+                    with open(f"MyTTSDataset/wavs/wav{index}.wav", 'wb') as file:
+                        file.write(response.content)
+                    f.write(f'wav{index}|{key}\n')
+        except requests.exceptions.MissingSchema:
+            pass
+        except requests.exceptions.InvalidSchema:
+            pass

utils/execute_cmd.py

@@ -16,14 +16,6 @@ def execute_cmd(self, cmd: str, recognized_phrase: str, voice: str) -> None:
         self.play("off", True)
         self.porcupine.delete()
         exit(0)
-    elif cmd == 'music_on':
-        self.media_player_controller.play_pause()
-    elif cmd == 'music_off':
-        self.media_player_controller.play_pause()
-    elif cmd == 'music_next':
-        self.media_player_controller.next_track()
-    elif cmd == 'music_previous':
-        self.media_player_controller.previous_track()
     elif cmd == 'home_assistant_execute':
         self.home_assistant.send_process(recognized_phrase)
     elif cmd == 'home_assistant_get':

utils/tts.py

@@ -0,0 +1,57 @@
+import os
+
+import torch
+import torchaudio
+
+
+def load_data(audio_folder):
+    audios = []
+    texts = []
+    for audio_file in os.listdir(audio_folder):
+        if audio_file.endswith('.wav'):
+            audio_path = os.path.join(audio_folder, audio_file)
+            waveform, sample_rate = torchaudio.load(audio_path)
+            text_path = audio_path.replace('.wav', '.txt')
+            with open(text_path) as f:
+                text = f.read().strip()
+            audios.append((waveform, sample_rate))
+            texts.append(text)
+    return audios, texts
+
+
+def train(model, audios, texts, epochs=3, learning_rate=1e-4):
+    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
+    criterion = torch.nn.MSELoss()  # Вам нужно будет настроить эту функцию под вашу задачу
+
+    model.train()
+    for epoch in range(epochs):
+        total_loss = 0
+        for waveform, text in zip(audios, texts):
+            optimizer.zero_grad()
+            # Предполагается, что модель принимает текст и возвращает аудио
+            predicted_waveform = model(text)
+            loss = criterion(predicted_waveform, waveform)
+            loss.backward()
+            optimizer.step()
+            total_loss += loss.item()
+        average_loss = total_loss / len(audios)
+        print(f'Epoch {epoch + 1}: Average Loss = {average_loss}')
+
+
+def main():
+    model_path = 'data/v4_ru.pt'
+    model = torch.load(model_path)
+    model.eval()
+    audio_folder = 'wav_files'
+    audios, texts = load_data(audio_folder)
+    train(model, audios, texts)
+    torch.save(model.state_dict(), 'fine_tuned_model.pth')
+    model.eval()
+    sample_text = "Пример текста для синтеза."
+    with torch.no_grad():
+        generated_waveform = model(sample_text)
+        torchaudio.save('output_audio.wav', generated_waveform, 16000)
+
+
+if __name__ == '__main__':
+    main()