o WliMB@sFUdZddlZddlZddlZddlZddlZddlZddlZddlm Z ddl m Z ddl m Z mZmZmZmZmZddlZddlZddlmZmZmZddlmZmZmZdd lmZdd l m!Z!m"Z"dd l#m$Z$da%ee"e&d <e'Z(eZ)d Z*e+dZ,e-e.Z/e GdddZ0e GdddZ1e GdddZ2dS)z&Phonemization and synthesis for Piper.N) dataclass)Path)AnyIterableOptionalSequenceTupleUnion) PhonemeType PiperConfigSynthesisConfig)BOSEOSPADphonemes_to_ids)ESPEAK_DATA_DIREspeakPhonemizer)TashkeelDiacritizer_ESPEAK_PHONEMIZERg@z (\[\[.*?\]\])c@s*eZdZUeed<eeed<eed<dS)PhonemeAlignmentphoneme phoneme_ids num_samplesN)__name__ __module__ __qualname__str__annotations__rintr!r!E/home/kim/smarthome/.venv/lib/python3.10/site-packages/piper/voice.pyr!s   rc@seZdZUdZeed< eed< eed< ejed< ee ed< eeed< dZ e ejed < dZ e ee ed < dZe ejed <dZe eed <dZe ee ed <edejfddZedefddZdS) AudioChunkzChunk of raw audio. sample_rate sample_widthsample_channelsaudio_float_arrayphonemesrNphoneme_id_samplesphoneme_alignments_audio_int16_array_audio_int16_bytes_phoneme_alignmentsreturncCs0|jdurt|jtt ttj|_|jS)zg Get audio as an int16 numpy array. :return: Audio data as int16 numpy array. N)r+npclipr'_MAX_WAV_VALUEastypeZint16selfr!r!r"audio_int16_arrayMs zAudioChunk.audio_int16_arraycCs |jS)zl Get audio as 16-bit PCM bytes. :return: Audio data as signed 16-bit sample bytes. )r5tobytesr3r!r!r"audio_int16_bytes[s zAudioChunk.audio_int16_bytes)rrr__doc__r rr/ndarraylistrr)rr*rr+r,bytesr-propertyr5r7r!r!r!r"r#(s2     r#c@seZdZUdZejed< eed< eZ e ed< e Z e ed< dZ eed<dZeeed <d Zeeed <edd edfd eee fdeeee fdedeee fdeeee fddf ddZdedeeefddZdeedeefddZ  d$dedeededeefddZ   d%dede j!deedededeee"f dd Z#  d$d!eedeededee$j%e&e$j%ee$j%fffd"d#Z'dS)& PiperVoicezA voice for Piper.sessionconfigespeak_data_dir download_dirT use_tashkeelNtashkeel_diacritizierg?taskeen_thresholdF model_path config_pathuse_cudar.cCs|dur|d}td|t|ddd }t|}Wdn1s&wY|r:ddd ifg}td nd g}|durEt}tt |t j t |t |d t|t|d S)a Load an ONNX model and config. :param model_path: Path to ONNX voice model. :param config_path: Path to JSON voice config (defaults to model_path + ".json"). :param use_cuda: True if CUDA (GPU) should be used instead of CPU. :param espeak_data_dir: Path to espeak-ng data dir (defaults to internal data). :param download_dir: Path to download resources (defaults to current directory). :return: Voice object. Nz.jsonzGuessing voice config path: %srzutf-8)encodingZCUDAExecutionProviderZcudnn_conv_algo_searchZ HEURISTICz Using CUDAZCPUExecutionProvider)Z sess_options providers)r?r>r@rA)_LOGGERdebugopenjsonloadrcwdr=r from_dict onnxruntimeInferenceSessionrZSessionOptions)rErFrGr@rA config_fileZ config_dictrJr!r!r"rOzs2    zPiperVoice.loadtextc Cs|jjtjkrttd|gS|jjtjkr9ddlm }t |dd}|dur4||j d}t |d|| |S|jjtjkrItd|jjg}t|}d}t|D]\}}|d rd }|sh|g|d kr|||dd r||d d |d |dd|t|dkr||dd r|d d qV|jjdkr|jr|jdurt|_|j||jd}t1tdurt |j!at |jj|} |r| r|d | d | dd} || Wdn1swYd}qV|r|d s|"|S)z Text to phonemes grouped by sentence. :param text: Text to phonemize. :return: List of phonemes for each sentence. ZNFDr )ChinesePhonemizerZ_chinese_phonemizerNZg2pWzUnexpected phoneme type: Fz[[Tr ar)rD)#r? phoneme_typer ZTEXTr: unicodedata normalizePINYINphonemize_chineserVgetattrrAsetattr phonemizeZESPEAK ValueError_PHONEME_BLOCK_PATTERNsplit enumerate startswithappendendswithextendstriplenZ espeak_voicerBrCrrD_ESPEAK_PHONEMIZER_LOCKrrr@pop) r4rUrVZ phonemizerr(Z text_partsZprev_raw_phonemesiZ text_partZtext_part_phonemesr!r!r"rcs\        "    zPiperVoice.phonemizer(cCs6|jjtjkrddlm}|||jjSt||jjS)zt Phonemes to ids. :param phonemes: List of phonemes. :return: List of phoneme ids. r r)r?r\r r_r`rphoneme_id_map)r4r(Zchinese_phonemes_to_idsr!r!r"rs zPiperVoice.phonemes_to_ids syn_configinclude_alignmentsc cs|durt}||}td|||D]}|sq||}d}|j|||d}t|tr3|\} }n|} |jrNt t | } | dkrJt | } n| | } |j dkrX| |j } t | ddt j} d} |durt|t|kr|jjtg} d} g} d}ttg|tgD]L}|jj|g}|tkrtt|| }n|}| }|D]}| t|krd }n||| krd }n| d 7} q|rn| t||t||| d q|rd} td t|jjd d | |||| dVqdS)a Synthesize one audio chunk per sentence from from text. :param text: Text to synthesize. :param syn_config: Synthesis configuration. :param include_alignments: If True and the model supports it, include phoneme/audio alignments. Nztext=%s, phonemes=%s)rsg:0yE>g?grFTr )rrrzPhoneme alignment failedrY)r$r%r&r'r(rr)r*) _DEFAULT_SYNTHESIS_CONFIGrcrKrLrphoneme_ids_to_audio isinstancetupleZnormalize_audior/maxabsZ zeros_likevolumer0r2float32rmr?rqgetr itertoolschainrrr:rirsumr#r$)r4rUrrrsZsentence_phonemesr(rr)Z audio_resultaudioZmax_valr*Zpad_idsZphoneme_id_idxZalignment_failedrZ expected_idsZ ids_to_checkZstart_phoneme_id_idxZ phoneme_idr!r!r" synthesize s               zPiperVoice.synthesizewav_fileset_wav_formatc Cs|g}d}|j|||dD]+}|r&|r$||j||j||jd}||j|r7|j r7| |j q |r<|SdS)a Synthesize and write WAV audio from text. :param text: Text to synthesize. :param wav_file: WAV file writer. :param syn_config: Synthesis configuration. :param set_wav_format: True if the WAV format should be set automatically. :param include_alignments: If True and the model supports it, return phoneme/audio alignments. :return: Phoneme/audio alignments if include_alignments is True, otherwise None. T)rrrsFN) rZ setframerater$Z setsampwidthr%Z setnchannelsr&Z writeframesr7r*rk) r4rUrrrrrsZ alignmentsZ first_chunkZ audio_chunkr!r!r"synthesize_wav|s$       zPiperVoice.synthesize_wavrcCsJ|durt}|j}|j}|j}|j}|dur|jj}|dur"|jj}|dur*|jj}ttj|tj dd}tj|j dgtj d} tj|||gtj d} || | d} |jj dkr[d}|jj dkrg|durgd}|durxtj|gtj d} | | d<|j d| } | d}|s|St| dkr|dfS| d|jjtj }||fS)a4 Synthesize raw audio from phoneme ids. :param phoneme_ids: List of phoneme ids. :param syn_config: Synthesis configuration. :param include_alignments: Return samples per phoneme id if True. :return: Audio float numpy array from voice model (unnormalized, in range [-1, 1]). If include_alignments is True and the voice model supports it, the return value will be a tuple instead with (audio, phoneme_id_samples) where phoneme_id_samples contains the number of audio samples per phoneme id. N)Zdtyperr )inputZ input_lengthsscalessid)rt speaker_id length_scale noise_scale noise_w_scaler?r/Z expand_dimsarrayZint64shaper{Z num_speakersr>runZsqueezermZ hop_lengthr2)r4rrrrsrrrrZphoneme_ids_arrayZphoneme_ids_lengthsrargsrresultrr)r!r!r"rusT   zPiperVoice.phoneme_ids_to_audio)NF)NTF)(rrrr8rRrSrr rr@rrPrArBboolrCrrrDfloat staticmethodr rrOr:rcr rr rr#rwaveZ Wave_writerrr/r9rrur!r!r!r"r=es       2Q s  .r=)3r8r}rNloggingre threadingr]r dataclassesrpathlibrtypingrrrrrr numpyr/rRr?r r r constrrrrrZphonemize_espeakrrZtashkeelrrrLockrnrtr1compilere getLoggerrrKrr#r=r!r!r!r"s<       <