Loads a configuration file and optionally merges it with a base configuration. Args: config_path (Path): Path to the configuration file.

Detokenizes the semantic and global tokens into a waveform. Args: semantic_tokens (tensor): Semantic tokens.

extract wav2vec2 features

Loads the model from a checkpoint. Args: model_dir (Path): Path to the model directory containing checkpoint and config.

Tokenizes the input audio into semantic and global tokens. Args: batch (dict): The input audio features and reference wa

Performs inference to generate speech from text, incorporating prompt audio and/or text. Args: text (str): The text inpu

Prepares inputs for Triton inference (offline or streaming).

Quantizes z, returns quantized zhat, same shape as z.

Removes weight normalization from all layers.

Perform TTS inference and save the generated audio.

Split a string using delimiter. Supports escaping. Args: string (str): The string to split. delimiter (str): The delimiter to spl

tokenize the audio

Initializes the tokenizer, model, and audio tokenizer for inference.

Load and initialize the BiCodec model and Wav2Vec2 feature extractor.

Turn age token.

return config dict with quantization info based on the command line args

Normalize the volume of an audio signal. Parameters: audio (numpy array): Input audio signal array. coeff (float): Target co

Bound `z`, an array of shape (..., d).

Converts a `code` to an index in the codebook.

Detect the start and end points of speech in an audio signal using RMS energy. Args: wav: Input audio signal array with values in [-1

detokenize the input indices to d-vector

detokenize the tokens to waveform Args: global_tokens: global tokens. shape: (batch_size, global_dim) semantic_tokens

Forward pass through the audio tokenizer component. Args: wav: Input waveform tensor wav_len: Waveform length

Prepares the response from the language model based on the provided inputs. Creates a `pb_utils.InferenceRequest` object with passed

Forward pass through the vocoder component. Args: global_token_ids: Global token IDs tensor pred_semantic_ids

Turn gender token.

Get reference audio clip for speaker embedding.

Extract reference audio clip for speaker embedding. Args: wav: Input waveform array Returns:

Converts indices to indices at each level, perhaps needed for a transformer with factorized embeddings

Initializes the MelSpectrogram transformer based on the provided configuration. Args: config (dict): Configuration param

Load the model once at the beginning.

r"""Load audio file with target sampling rate and lsength Args: adfile (Path): path to audio file. sampling_rate (int, optional):

Turn special token of loudness level.

Turn special toak of loudness value [0, 30]

Turn special token of mel level.

Turn special token of mel scale pitch.

Parse command-line arguments.

Parse command-line arguments such as model directory and device ID.

load auido and get reference audio from wav path

Process input for voice cloning. Args: text: The text input to be converted to speech. prompt_text: Transcript of the prompt

Process input for voice cloning. Args: text (str): The text input to be converted to speech. prompt_speech_p

Process input for voice creation. Args: gender (str): female | male. pitch (str): very_low | low | moderate

get an audio segment given the length Args: audio (np.ndarray): length (int): audio length = sampling_rate * duration