o i&@svddlmZmZddlZddlZddlZddlZ ddl m Z ddZ ddZ dd d ZGd d d ZGdddZdS))enespeakN)TextPreprocessorcCsddl}|d|}|S)zBBasic English tokenizer that splits on whitespace and punctuation.rNz \w+|[^\w\s])refindall)textrtokensr H/home/ubuntu/.local/lib/python3.10/site-packages/kittentts/onnx_model.pybasic_english_tokenizes r cCs(|}|s|S|ddvr|d}|S)z7Ensure text ends with punctuation. If not, add a comma.z.!?,;:,)strip)rr r r ensure_punctuations  rc Csddl}|d|}g}|D]Q}|}|sqt||kr%|t|q|}d}|D]&}t|t|d|krF||rBd|n|7}q-|rQ|t||}q-|r_|t|q|S)z1Split text into chunks for processing long texts.rNz[.!?]+r )rsplitrlenappendr) rmax_lenr sentenceschunkssentencewords temp_chunkwordr r r chunk_texts*  rc@seZdZdddZddZdS) TextCleanerNc CsZd}d}d}d}|gt|t|t|}i}tt|D]}||||<q||_dS)N$u;:,.!?¡¿—…"«»"" 4ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyzuɑɐɒæɓʙβɔɕçɗɖðʤəɘɚɛɜɝɞɟʄɡɠɢʛɦɧħɥʜɨɪʝɭɬɫɮʟɱɯɰŋɳɲɴøɵɸθœɶʘɹɺɾɻʀʁɽʂʃʈʧʉʊʋⱱʌɣɤʍχʎʏʑʐʒʔʡʕʢǀǁǂǃˈˌːˑʼʴʰʱʲʷˠˤ˞↓↑→↗↘'̩'ᵻ)listrangerword_index_dictionary) selfdummy_pad _punctuation_letters _letters_ipasymbolsdictsir r r __init__8s zTextCleaner.__init__c Cs8g}|D]}z ||j|WqtyYqw|SN)rr$KeyError)r%rindexescharr r r __call__Fs zTextCleaner.__call__r/)__name__ __module__ __qualname__r.r3r r r r r7s  rc@seZdZddiifddZddededed efd d Zddedededed e j f ddZ ddededed e j fddZ ddedededede ded dfddZdS)KittenTTS_1_Onnxzkitten_tts_nano_preview.onnxz voices.npzcCsj||_t||_t||_tjj dddd|_t |_ ||_ gd|_ gd|_||_tdd|_dS) zInitialize KittenTTS with model and voice data. Args: model_path: Path to the ONNX model file voices_path: Path to the voices NPZ file zen-usT)languagepreserve_punctuation with_stress)zexpr-voice-2-mzexpr-voice-2-fzexpr-voice-3-mzexpr-voice-3-fzexpr-voice-4-mzexpr-voice-4-fexpr-voice-5-mzexpr-voice-5-f)BellaJasperLunaBrunoRosieHugoKikiLeoF)remove_punctuationN) model_pathnploadvoicesortInferenceSessionsession phonemizerbackend EspeakBackendr text_cleaner speed_priorsavailable_voicesall_voice_names voice_aliasesr preprocessor)r%rE voices_pathrPrSr r r r.Qs    zKittenTTS_1_Onnx.__init__?rvoicespeedreturnc Cs||jvr |j|}||jvrtd|d|j||jvr&||j|}|j|g}t|d}d|}||}| dd| d| dt j |gt j d}tt||j|jdd}|j|||d} || t j |gt jddS) z9Prepare ONNX model inputs from text and voice parameters.zVoice 'z' not available. Choose from: rr )dtyper) input_idsstylerX)rSrQ ValueErrorrPrL phonemizer joinrOinsertrrFarrayint64minrrHshapefloat32) r%rrWrX phonemes_listphonemesr r\ref_idref_sr r r _prepare_inputsls(          z KittenTTS_1_Onnx._prepare_inputsr;T clean_textcCsBg}|r ||}t|D] }|||||q tj|ddS)Nr )axis)rTrrgenerate_single_chunkrF concatenate)r%rrWrXrl out_chunks text_chunkr r r generates   zKittenTTS_1_Onnx.generatecCs4||||}|jd|}|ddddf}|S)aSynthesize speech from text. Args: text: Input text to synthesize voice: Voice to use for synthesis speed: Speech speed (1.0 = normal) Returns: Audio data as numpy array Nr.ix)rkrKrun)r%rrWrX onnx_inputsoutputsaudior r r rns z&KittenTTS_1_Onnx.generate_single_chunk] output_path sample_rateNcCs2|j||||d}t|||td|dS)aSynthesize speech and save to file. Args: text: Input text to synthesize output_path: Path to save the audio file voice: Voice to use for synthesis speed: Speech speed (1.0 = normal) sample_rate: Audio sample rate clean_text: If true, it will cleanup the text. Eg. replace numbers with words. )rlzAudio saved to N)rrsfwriteprint)r%rrxrWrXryrlrvr r r generate_to_files z!KittenTTS_1_Onnx.generate_to_file)rV)r;rVT)r;rV)r;rVrwT)r4r5r6r.strfloatdictrkboolrFndarrayrrrnintr}r r r r r7Ps ""r7)r)misakirrnumpyrFrL soundfilerz onnxruntimerI preprocessrr rrrr7r r r r s