# """Mock TTS implementation for local development""" # import numpy as np # class MockTTSModel: # def __init__(self): # self.model = None # def initialize(self): # """Mock initialization""" # self.model = "mock_model" # return True # def list_voices(self): # """Return mock list of voices""" # return ["mock_voice_1", "mock_voice_2"] # def generate_speech(self, text, voice_names, speed, gpu_timeout=90, progress_callback=None, progress_state=None, progress=None): # """Generate mock audio data""" # # Create mock audio data (1 second of silence) # sample_rate = 22050 # duration = 1.0 # t = np.linspace(0, duration, int(sample_rate * duration)) # audio_array = np.zeros_like(t) # # Mock metrics # metrics = { # "tokens_per_sec": [10.5, 11.2, 10.8], # "rtf": [0.5, 0.48, 0.52], # "total_time": 3, # "total_tokens": 100 # } # # Simulate progress updates # if progress_callback and progress_state and progress: # for i in range(3): # progress_callback(i+1, 3, metrics["tokens_per_sec"][i], # metrics["rtf"][i], progress_state, # progress_state.get("start_time", 0), # gpu_timeout, progress) # return audio_array, duration, metrics