Initialize StyleTTS2 model for text sythesizing. StyleTTS2 repository have to be cloned locally and a valid path to the repo have to be pas

Reset and extend the positional encodings if needed.

Perform an encoding, decoding and joint step on the given audio signal. Args: processed_signal: preprocessed audio chunk

Perform a single LLM inference step Args: input_token: either input token from user or previous generated token from LLM

Perform a single voice assistant forward path 1) If a processed signal is passed and speech is detected, the audio chunk is transcribed

Compute 'Scaled Dot Product Attention'. Args: query (torch.Tensor): (batch, time1, size) key (torch.Tensor): (batch, t

Compute Scaled Dot Product Local Attention with rel. positional encoding. using overlapping chunks Args: query (torch.Tensor): (ba

Adds positional encoding. Args: x (torch.Tensor): Input. Its shape is (batch, time, feature_size) cache_len (int): the

Compute positional encoding. Args: x (torch.Tensor): Input. Its shape is (batch, time, feature_size) cache_len (int):

Synthesize the entire given text. For good quality a whole sentence should be passed Args: text: String of text that should be sy

Wait for audio input, call voice assistant model and play synthesized speech Args: streaming_buffer: streaming buffer instance t

Play synthesized audio data in a separate process Args: audio_output_buffer: multiprocessing-queue to receive audio data

Signal handler for keyboard interrupt to terminate all subprocesses