o i2 @sdZddlmZddlZddlmZddlmZmZm Z m Z m Z ddl m Z ddlmZmZmZmZmZdZdd d d d d ddddddd Zdddddddddddd d! ZeeZeeeZgd"Zd#d$d%d&d'd(d)d*d+d,d- ZdXd1d2Zid3d4eDid5d(d6d(d7d'd8d'd9d'd:d*d;d*dd#d?d%d@d,dAd,dBd)dCd)dDd)dEd+d+d+d$dFZdYdHdIZ GdJdKdKe!eZ"GdLdMdMe!eZ#GdNdOdOeZ$GdPdQdQeZ%GdRdSdSeZ&GdTdUdUeZ'GdVdWdWeZ(dS)Zz Pydantic schemas for TTS API requests/responses. Ported from veena3srv/apps/api/serializers.py with full validation parity. No Django dependencies - pure Pydantic for Modal deployment. ) annotationsN)Enum)AnyDictListOptionalUnion)UUID) BaseModel ConfigDictFieldfield_validatormodel_validatoriP ) lipakshivardanreetNandinikrishnaanikaadarshNilayAarviAshaBittuMirarrrr!rrr rr"r#r$r%) MitraAaranyaTaruNeerDhruvaIraVedaAriar"r#r$r%) [angry] [curious] [excited][giggle][laughs harder][laughs] [screams][sighs][sings] [whispers]r.r/r0r1r2r3r4r5r6r7) zz z zzzzzzz namestrreturncCs6|tvrt|S|tvr|Std|ddt)a Resolve friendly speaker name to internal name. Examples: resolve_speaker_name("Mitra") -> "lipakshi" resolve_speaker_name("lipakshi") -> "lipakshi" Raises: ValueError: if name is not a valid speaker name zInvalid speaker name: z. Valid names: , )FRIENDLY_SPEAKER_MAPINDIC_SPEAKERS ValueErrorjoinALL_SPEAKER_NAMES)r8rA2/home/ubuntu/veenaModal/veena3modal/api/schemas.pyresolve_speaker_name;s rCcCsi|] \}}|d|qS)z<>)strip).0kvrArArB QsrHlaughinglaughz laughs harderz laugh harder laugh_hardersighssighgigglesgiggleangryexcitedwhisperswhisper screamingscreamscreamssinging)singsingscurioustextcCs4d}dd}t|||}d}dd}t|||S)u Normalize emotion tags to Spark TTS bracket format. Conversions: - → [laughs] - [laughing] → [laughs] - etc. z <([a-z_]+)>cSsJ|d}|tvrt|Sd|d}|tvrt|Sd|dS)Nr<>[])grouplowerrD_EMOTION_TO_TAGLEGACY_EMOTION_MAP)matchemotionold_tagrArArB replace_anglehs  z-normalize_emotion_tags..replace_anglez \[([^\]]+)\]cSs|d}d|d}|tvr|S|tvrt|S|dr2|ddtvr2t|ddS|dtvr>t|dS|S)Nrr^r_s)r`rarDINDIC_EMOTION_TAGSrbendswith)rdre bracket_tagrArArBreplace_bracketvs   z/normalize_emotion_tags..replace_bracket)resub)r[ angle_patternrgbracket_patternrmrArArBnormalize_emotion_tags\s   rrc@s$eZdZdZdZdZdZdZdZdS) AudioFormatzSupported audio output formats.wavopusmp3mulawflacN) __name__ __module__ __qualname____doc__WAVOPUSMP3MULAWFLACrArArArBrssrsc@seZdZdZdZdZdS)OutputSampleRatezOutput sample rate options.16khz48khzN)ryrzr{r|SR_16KHZSR_48KHZrArArArBrsrc@s.eZdZUdZeddeddZded<edd d d Zd ed <edddd Z d ed<edddZ ded<edd dd Z d ed<edddddZ ded<edddZ ded <eejd!dZd"ed#<ed$d%dZd&ed'<ed(d)d*d+dZd,ed-<ed dd.d/dZd&ed0<ed1d)d1d2dZd,ed3<ed4d5d4d6dZd&ed7<ed8d1d*d9dZd,ed:<ed;ddZded?<ed;d@dZdedA<eejdBdZdCedD<dZd edE<dZd edF<ed;dGZ e!de"dYdJdKZ#e!de"dZdLdMZ$e!d'e"d[dNdOZ%e&dPdQd\dSdTZ'd]dUdVZ(d^d]dWdXZ)dS)_TTSGenerateRequestzx Request schema for /v1/tts/generate. Mirrors validation rules from veena3srv/apps/api/serializers.py. .rz*Text to synthesize (max 50,000 characters)) min_length max_length descriptionr9r[N2z#Speaker name (internal or friendly))rr Optional[str]speakerizBVoice description (for voiceDesign model, NOT currently supported)rz9UUID of pre-created voice profile (for voiceDesign model))rOptional[UUID]voice_idz0Model to use (optional). Default: indic_speakersmodelriz.Random seed for reproducibility (0-2147483647))geler Optional[int]seedFzStream audio as it's generatedboolstreamzAudio output formatrsformat>z"Sample rate in Hz (default: 16000)int sample_rateg?gg@zSampling temperature (0.0-2.0)float temperaturedzTop-k sampling (1-100)top_kg?zNucleus sampling (0.0-1.0)top_piz"Maximum BiCodec tokens to generate max_tokensg?zRepetition penalty (1.0-2.0)repetition_penaltyTz#Apply text normalization before TTS normalizez2Return normalized text in X-Normalized-Text headernormalize_verbosez0Enable intelligent text chunking for long inputschunkingz9Output sample rate: '16khz' or '48khz' (super-resolution)routput_original_text_resolved_speaker)use_enum_valuesrGr:cCs2|}|s tddd|D}|rtd|S)z0Validate text field: no empty, no control chars.z'Text cannot be empty or whitespace onlycS$g|]}t|dkr|dvr|qS z ordrEcrArArB $z4TTSGenerateRequest.validate_text..z(Text contains invalid control characters)rDr>)clsrGstripped control_charsrArArB validate_textsz TTSGenerateRequest.validate_textcCs&|s|Sdd|D}|rtd|S)zValidate description field.cSrrrrrArArBr#rz;TTSGenerateRequest.validate_description..z/Description contains invalid control charactersr>)rrGrrArArBvalidate_descriptions z'TTSGenerateRequest.validate_descriptioncCs"gd}||vrtd||S)z2Validate sample rate is one of the allowed values.)@ri"Vi]iDiz$Invalid sample rate. Valid options: r)rrG valid_ratesrArArBvalidate_sample_rate(sz'TTSGenerateRequest.validate_sample_rateafter)mode'TTSGenerateRequest'cCst|j}t|j}|jdu}|stddt|jtvr.td|jddtt|j|_|s8|rr?r@rCrrrsrr)self has_speakerhas_description has_voice_idrArArBcross_field_validation1s*     z)TTSGenerateRequest.cross_field_validationcCs |jr|jS|jrt|jSdS)z*Return the resolved internal speaker name.)rrrC)rrArArBget_resolved_speakerYsz'TTSGenerateRequest.get_resolved_speakercCs*|j}|jr|r||_||}t|}|S)z Return normalized + emotion-normalized text. Args: normalizer_func: Optional custom normalizer (for testing). If None, returns text with only emotion normalization. )r[rrrr)rnormalizer_funcr[rArArBget_normalized_text_s  z&TTSGenerateRequest.get_normalized_text)rGr9r:r9)rGrr:r)rGrr:r)r:r)r:r9)N)*ryrzr{r|r MAX_TEXT_LENGTHr[__annotations__rrrrrrrsr}rrrrrrrrrrrrrrrr model_configr classmethodrrrrrrrrArArArBrs        'rc@seZdZUdZded<ded<ded<ded<dZd ed <dZd ed <dZd ed<dZd ed<dZ d ed<dZ ded<dZ d ed<dZ d ed<dZ ded<dZded<dS)TTSGenerateResponsez5Response metadata for non-streaming /v1/tts/generate.r9 request_idrrrrNrrrrr audio_url tokens_prompttokens_generatedzOptional[float]audio_duration_seconds audio_bytesttfb_msrtfcredits_consumed)ryrzr{r|rrrrrrrrrrrrArArArBrvs          rc@sLeZdZUdZded<ded<eedZded<dZd ed <dZ d ed <dS) ErrorDetailzError detail structure.r9codemessage)default_factoryzDict[str, Any]detailsNrrdocumentation_url) ryrzr{r|rr dictrrrrArArArBrs  rc@seZdZUdZded<dS) ErrorResponsezStandard error response.rerrorNryrzr{r|rrArArArBrs  rc@sBeZdZUdZded<ded<ded<ded<ded <ded <d S) HealthResponsez%Response for /v1/tts/health endpoint.r9statusr model_loaded model_versionruptime_seconds gpu_available app_versionNrrArArArBrs  r)r8r9r:r9)r[r9r:r9))r| __future__rrnenumrtypingrrrrruuidr pydanticr r r r rr SPEAKER_MAPr<listkeysr=r@rjrcrCitemsrbrrr9rsrrrrrrrArArArBs      - ]