o oiK6@srdZddlmZmZmZddlmZddlZddl Z ddl Z ddl Z ddl m Z mZmZddZGd d d ZdS) zSimple API for Kani-TTS.)TupleOptionalUnion)PathN) TTSConfigNemoAudioPlayer KaniModelcCstdzddl}|j|jWn tyYnwtdtj tdtj tdtj tdtj tdtj td tj td tj ttj dS) z Suppress all logging output from transformers, NeMo, PyTorch, and other libraries. Only print() statements from user code will be visible. ignorerNnemo nemo_loggertorchpytorchnumba matplotlibPIL) warningsfilterwarnings transformersloggingset_verbosity_errordisable_progress_bar ImportError getLoggersetLevelERROR)rr@/home/ubuntu/.local/lib/python3.10/site-packages/kani_tts/api.pysuppress_all_logs s   rc@sneZdZdZ            d1deded ed ed ed ed eedeedeedeedeedeedeefddZ ddZ     d2dedeedee e j eefdeded ed!eejeffd"d#Z     d2dedeedee e j eefdeded ed!eejeffd$d%Zd&e eefd!e j fd'd(Zd)ejd*efd+d,Zd-d.Zd3d/d0ZdS)4KaniTTSz Simple interface for Kani text-to-speech model. Example: >>> model = KaniTTS('your-model-name') >>> audio, text = model("Hello, world!") auto TN model_name device_mapmax_new_tokenstokeniser_length suppress_logs show_infotext_vocab_sizetokens_per_frame audio_stepuse_learnable_rope alpha_min alpha_maxspeaker_emb_dimc Cs|rtt|||||| | | | | d |_||_t|j|_t|j||j|_|jj|_|jj |_ |jj |_ | |rA| dSdS)a Initialize Kani-TTS model. Args: model_name: Hugging Face model ID or path to local model device_map: Device mapping for model (default: "auto") max_new_tokens: Maximum number of tokens to generate (default: 3000) tokeniser_length: Length of text tokenizer vocabulary (default: 64400) suppress_logs: Whether to suppress library logs (default: True) show_info: Whether to display model info on initialization (default: True) text_vocab_size: Text vocabulary size for position encoding. If None, reads from model config. (default: None) tokens_per_frame: Number of audio tokens per frame. If None, reads from model config. (default: None) audio_step: Position step size per audio frame. If None, reads from model config. (default: None) use_learnable_rope: Enable learnable RoPE with per-layer alpha. If None, reads from model config. (default: None) alpha_min: Minimum alpha value for learnable RoPE. If None, reads from model config. (default: None) alpha_max: Maximum alpha value for learnable RoPE. If None, reads from model config. (default: None) speaker_emb_dim: Dimension of speaker embeddings. If None, reads from model config. (default: None) ) r$r&r%r)r*r+r,r-r.r/N) rrconfigr#rplayerr modelstatuslanguage_tags_list sample_rate_sync_config_from_modelshow_model_info)selfr#r$r%r&r'r(r)r*r+r,r-r.r/rrr__init__5s0)     zKaniTTS.__init__cCs|jj}|jjdurt|dd|j_|jjdur t|dd|j_|jjdur.t|dd|j_|jjdurtd|d|j d|S|jdkr]|j d|jj kr[td |jj d |j |Std |j ) z Load speaker embedding from a .pt file. Args: path: Path to .pt file containing speaker embedding Returns: Speaker embedding tensor [speaker_emb_dim] or [1, speaker_emb_dim] z"Speaker embedding file not found: z.ptz+Speaker embedding must be a .pt file, got: rrz0Speaker embedding has wrong dimension: expected z, got z6Speaker embedding has wrong dimension: expected [..., z], got z/Speaker embedding must be 1D or 2D, got shape: ) rexistsFileNotFoundErrorsuffix ValueErrorr loadrMr0r/shape)r8rPrA expected_dimrrrrLs0      zKaniTTS.load_speaker_embeddingaudio output_pathcCs6zddl}||||jWdStytdw)z Save audio waveform to file. Args: audio: Audio waveform as numpy array output_path: Path to save audio file (e.g., "output.wav") rNzKsoundfile is required to save audio. Install it with: pip install soundfile) soundfilewriter5r)r8rYrZsfrrr save_audios zKaniTTS.save_audiocCsttdtdtdtdtdttdtdtdttd|j}t|d kr>d |d d }td |tjrLdnd}td||jdkrtdt|jd|jr}t|jdkr}d|j}td|n|jrtd|jdd|jddntdttdtd|j dtd|j j td|j j pd td!|j j pd td"|j jpd td#|j jpd |j jrtd$td%|j jpd d|j jpd d&ntd'tdttd(td S))z= Display beautiful model information banner. u╔════════════════════════════════════════════════════════════╗uB║ ║uD║ N I N E N I N E S I X 😼 ║u╚════════════════════════════════════════════════════════════╝z /\_/\ z ( o.o )z > ^ )rEN)__name__ __module__ __qualname____doc__strintboolrfloatr9r6rr rKrrnpndarrayrIrGrLr^r7rprrrrr,s       F     $&9r)rttypingrrrpathlibrnumpyryrrr corerrr rrrrrrs !