o eiwF @sddlZddlZddlmZmZejejdedZddl m Z m Z m Z ddl mZddlmZmZzddlmZmZmZmZmZWn)eymddlZejejejeddlmZmZmZmZmZYnwzdd lm Z m!Z!m"Z"m#Z#m$Z$Wn)eyddlZejejejedd lm Z m!Z!m"Z"m#Z#m$Z$YnwGd d d eZ%Gd d d eZ&GdddeZ'GdddeZ(GdddeZ)GdddeZ*GdddeZ+GdddeZ,e ddedd-ede$dd-e"d d!Z.e.j/ed"gd#d"gd"gd$e.j0d%e&d&d'ed(d)d*id)d+id)d,id-d.gd/d0e%fd1d2Z1e.j0d3e*d4d5e"d6d)d7id)d8id)d9id-d:gd/d0e)fd;d<Z2e.j3d=e(d>d?d@gdAdBdCZ4e.j3dDe,dEdFd@gdAdGdHZ5e.j3dIdJdKd@gdLdMdNZ6dS)ON)ListOptional)levelresponse_iq_api)FastAPI HTTPExceptionstatus)CORSMiddleware) BaseModelField)analyze_responseLANGUAGE_CONFIGDEFAULT_LANGUAGESUPPORTED_LANGUAGES OLLAMA_MODEL) generate_ttsTTS_LANGUAGE_CONFIGTTS_SUPPORTED_LANGUAGESTTS_DEFAULT_LANGUAGE TTS_MODEL_IDc@seZdZUdZedddZeed<edddZeed<edddZ eed <ed d dZ e eed <ee d e de ddZe eed<GdddZd S)ResponseIQRequestuf Request payload for the /response_iq_data endpoint. Fields ------ question : The interview / assessment question. predefinedAnswer : The ideal / expected answer. userAnswer : The candidate's actual text answer. userVoice : Optional — URL or Base64-encoded audio of the candidate's voice. language : Language key controlling the script and output language. Supported values: english, hindi, marathi, tamil, telugu, kannada, punjabi. Defaults to 'english' when omitted. .z$The interview / assessment question. descriptionquestionzThe ideal / expected answer.predefinedAnswerzThe candidate's actual answer. userAnswerNz-URL or Base64 audio of the candidate's voice. userVoicez$Language of the content. Supported: . Defaults to ''.defaultrlanguagec@eZdZdZdS)zResponseIQRequest.ConfigignoreN__name__ __module__ __qualname__extrar)r)H/var/www/eduai.edurigo.com/responseiq/production/response_iq_fast_api.pyConfigYr+)r%r&r'__doc__r rstr__annotations__rrrrrrr!r+r)r)r)r*r<s  rc@seZdZUdZedddZeed<edddZe ed<edddZ e ed <ed d dZ e e ed <edd dZ ee ed<ed ddZe e ed<d S)ResponseIQResponseu Structured analysis result returned by /response_iq_data. Fields ------ matchScore : Semantic similarity score 0–100. comparison : Rationale for the score (in requested language). behavioralAnalysis: Communication style insights (in requested language). tonalAnalysis : Voice delivery or text-tone analysis (in requested language). visuals : Exactly 3 vibe keywords (in requested language). .uSemantic match score (0–100).r matchScorez*Score rationale in the requested language. comparisonzCommunication style insights.behavioralAnalysisNz%Voice delivery or text-tone analysis. tonalAnalysisz*3 vibe keywords in the requested language.visualszGWhisper transcription of the user's voice (when userVoice is provided).voiceTranscription)r%r&r'r-r r1intr/r2r.r3r4rr5rr6r)r)r)r*r0]s  r0c@JeZdZUedddZeed<edddZeed<edddZeed<d S) SupportedLanguageItem.-Language key to pass in the 'language' field.rkeyHuman-readable language name. display_namez Language and script description. script_nameN) r%r&r'r r;r.r/r=r>r)r)r)r*r9r r9c@s"eZdZUeeed<eed<dS)SupportedLanguagesResponsesupported_languagesdefault_languageN)r%r&r'rr9r/r.r)r)r)r*r@xs  r@c@s\eZdZUdZeddddZeed<eede ded d Z e eed <Gd d d Z dS)ResponseIQTTSRequestuj Request payload for the /response_iq_tts endpoint. Fields ────── predefinedAnswer : The text to be converted to speech. Typically the ideal/model answer for a question. language : Language key that controls the voice, accent, and script used during synthesis. Supported values: english, hinglish, hindi, marathi, punjabi, tamil, telugu, kannada, gujarati, bengali, odia, assamese, maithili, urdu. Defaults to 'english' when omitted. .z#The text to synthesise into speech.)r min_lengthrz$Target language for TTS. Supported: rrrr!c@r")zResponseIQTTSRequest.Configr#Nr$r)r)r)r*r+r,r+N) r%r&r'r-r rr.r/rrr!rr+r)r)r)r*rCs  rCc@sveZdZUdZedddZeed<edddZeed<edddZ e ed <edd dZ e ed <edd dZ e ed <dS)ResponseIQTTSResponseu TTS result returned by /response_iq_tts. Fields ────── audio_tts : Base64-encoded WAV audio string. Decode and play directly. language : Normalised language key used for synthesis. sample_rate : Sample rate of the returned audio (Hz). duration_sec : Duration of the synthesised audio in seconds. char_count : Number of characters that were synthesised. .z3Base64-encoded WAV audio of the synthesised speech.r audio_ttsz Language key used for synthesis.r!zAudio sample rate in Hz. sample_ratez)Duration of synthesised audio in seconds. duration_secz!Number of characters synthesised. char_countN)r%r&r'r-r rGr.r/r!rHr7rIfloatrJr)r)r)r*rFs  rFc@r8) TTSSupportedLanguageItem.r:rr;r<r=z'BCP-47 language tag used for synthesis. bcp47_tagN) r%r&r'r r;r.r/r=rMr)r)r)r*rLr?rLc@s*eZdZUeeed<eed<eed<dS)TTSSupportedLanguagesResponserArBmodel_idN)r%r&r'rrLr/r.r)r)r)r*rNs  rNzResponse IQ APIu## Response IQ — Intelligent Answer Analysis & Text-to-Speech ### Analysis (`/response_iq_data`) Analyses a candidate's interview answer against an ideal answer using the **z{** model via Ollama. Supports multi-language content and optional voice tonal analysis. **Supported analysis languages:** z, zw --- ### Text-to-Speech (`/response_iq_tts`) Converts text to natural human-like speech using **Indic Parler-TTS** (`zd`) running on the local GPU server. Returns Base64-encoded WAV audio. **Supported TTS languages:** z3.0.0)titlerversion*T) allow_originsallow_credentials allow_methods allow_headersz/response_iq_datazAnalyse a candidate's responsezAccepts a question, the ideal answer, and the candidate's answer. Optionally accepts a voice recording (URL or Base64). The **`language`** field controls which language the analysis is written in. Supported values: `u`. Returns a structured evaluation with match score, behavioural analysis, tonal analysis, and 3 vibe keywords — all in the requested language.rzSuccessful analysis result.z"Unsupported language key supplied.zLLM or audio-processing error.)iiAnalysis)response_modelsummaryr responsestagspayloadc s|jpt}|tvrttjd|dtddt d|d|j r'dndd |j d d d t |j |j|j|j |d }d|vrYt d|dttj|dd|S)z 1. Validates the language key. 2. Optionally extracts acoustic features from the voice recording. 3. Calls the LLM to produce a language-aware analysis. 4. Returns the structured result. Unsupported language 'z'. Valid values are: . status_codedetailz"POST /response_iq_data | language=z | has_voice=yesnoz | question='N2z...')rpredefined_answer user_answeruser_voice_urlr!errorzAnalysis failed: )r!rstriplowerr rrHTTP_400_BAD_REQUESTrloggerinforrr rrriHTTP_500_INTERNAL_SERVER_ERROR)r]requested_languageanalysis_resultr)r)r*response_iq_data_endpoints<   rrz/response_iq_ttsz(Convert predefined answer text to speechaSynthesises the `predefinedAnswer` text into natural, human-like speech using **Indic Parler-TTS** (`ai4bharat/indic-parler-tts`) running on the local GPU server. The `language` field selects the speaker's accent, voice profile, and script-aware phoneme rendering. Supported values: `a`. The response contains a **Base64-encoded WAV** string in `audio_tts`. Decode it on the client and play directly, or save as a `.wav` file. **Example client usage (PHP):** ```php $audioData = base64_decode($response['audio_tts']); file_put_contents('answer.wav', $audioData); ``` **Example client usage (JavaScript):** ```js const blob = new Blob( [Uint8Array.from(atob(response.audio_tts), c => c.charCodeAt(0))], { type: 'audio/wav' } ); const url = URL.createObjectURL(blob); new Audio(url).play(); ```z'Base64 WAV audio returned successfully.z0Unsupported language key or empty text supplied.z*TTS model error or GPU processing failure.zText-to-Speechc s|jpt}|tvrttjd|dtdd|j p d}|s,ttjddt d|dt |d |d d t |d krEd ndd t ||d}d|vrht d|dttj|ddt d|d|ddt |dd|S)a( 1. Validates the language key against TTS_LANGUAGE_CONFIG. 2. Normalises and chunks the predefinedAnswer text. 3. Synthesises each chunk via Indic Parler-TTS (GPU inference). 4. Concatenates chunks, resamples, encodes as Base64 WAV. 5. Returns the structured TTS result. r^z' for TTS. Valid values are: r_r`z%'predefinedAnswer' must not be empty.z!POST /response_iq_tts | language=z | chars=z | preview='N<z...')textr!riz TTS failed: zTTS response ready | language=z | duration=rIzs | base64_len=rG)r!rrjrkrrrrlrrrmrnlenrriroget)r]rprv tts_resultr)r)r*response_iq_tts_endpoint-sT-  rzz/supported_languagesz!List supported analysis languageszReturns all language keys accepted by the `language` field of `/response_iq_data`, along with their display names and script descriptions. Use this to populate language selector dropdowns in your frontend.Metadata)rYrZrr\csddtDtdS)zQReturns every supported analysis language key along with human-readable metadata.cS$g|]\}}||d|ddqS)r=r>)r;r=r>r).0r;cfgr)r)r* z0supported_languages_endpoint..)rArB)r itemsrr)r)r)r*supported_languages_endpoints rz/tts_supported_languageszList supported TTS languageszReturns all language keys accepted by the `language` field of `/response_iq_tts`, along with display names and BCP-47 tags. Use this to populate language selector dropdowns in your frontend.csddtDttdS)zLReturns every supported TTS language key along with human-readable metadata.cSr|)r=bcp47_lang_tag)r;r=rMr)r}r)r)r*rrz4tts_supported_languages_endpoint..)rArBrO)rrrrr)r)r)r* tts_supported_languages_endpointsrz/healthz Health checkz>Returns API status, active LLM model name, and TTS model name.)rZrr\csdttttdS)zSimple liveness probe.ok)r llm_model tts_modelrAtts_supported_languages)rrrrr)r)r)r* health_checks r)7osloggingtypingrr basicConfigINFO getLoggerrmfastapirrrfastapi.middleware.corsr pydanticr r response_iq_analysisr r rrr ImportErrorsyspathappenddirnameabspath__file__response_iq_ttsrrrrrrr0r9r@rCrFrLrNjoinappadd_middlewarepostrrrzrxrrrr)r)r)r*s        ! #     2#=