o y i'<@sdZddlZddlZddlZddlZddlZddlmZmZm Z m Z ddl m Z ddl Z ddl Z ddlmZddlmZddlmZGdd d ZeZdS) za Speech-to-Text Service for MCube Voice Bot Handles audio transcription using OpenAI Whisper API N)OptionalDictAnyList)datetime)Config)Log)AudioNoiseServicec @seZdZdZddZdededeefddZd eededeefd d Z d e dedee fd dZ de de fddZ dde de de fddZde dedeefddZdeeeefdedeeeeffddZdS)SpeechToTextServicez Service for converting audio to text using OpenAI Whisper API. Handles audio format conversion and transcription. cCsrtj|_d|_g|_t|_tj|_ tj |_ tj |_ t|_|j r2td|j d|j ddStddS)Nz.https://api.openai.com/v1/audio/transcriptionsu0🎤 Speech-to-Text Service initialized (Model: z , Language: )u4🎤 Speech-to-Text Service initialized but disabled)rZOPENAI_API_KEYopenai_api_key whisper_urlZ audio_bufferasyncioZLockZ buffer_lockZSPEECH_TO_TEXT_ENABLEDenabledZ STT_LANGUAGElanguageZ STT_MODELmodelr noise_servicerinfo)selfr"services/speech_to_text_service.py__init__s  zSpeechToTextService.__init__ audio_data session_idreturnc sB|jstd|dSztt|}td|dt|dt|dkr6tdt|dWdS|||IdH}|sLtd|WdStd ||||IdH}|rx| rxt d |d |dd d | WSt d|WdSt y}zt d|d|WYd}~dSd}~ww)a& Transcribe base64 encoded audio data to text using OpenAI Whisper. Args: audio_data: Base64 encoded audio data session_id: Session ID for logging Returns: Transcribed text or None if transcription fails uA🎤 Speech-to-text disabled, skipping transcription for session Nu🎤 Decoded audio for session :  bytesiu(🎤 Audio too short for transcription: u2⚠️ Failed to process audio format for session u%🎤 Calling Whisper API for session u#🎤 Transcribed audio for session : '2...'u-🎤 No speech detected in audio for session u)❌ Error transcribing audio for session ) rrdebugbase64 b64decodelen_process_audio_formatwarning_call_whisper_apistripr Exceptionerror)rrr audio_bytesprocessed_audio transcriptionerrrtranscribe_audio*s4      z$SpeechToTextService.transcribe_audio audio_chunksc s~|jstd|dSz|sWdSd}|D]*}z t|}||7}WqtyB}ztd|d|WYd}~qd}~ww|sPtd|WdSt|dkrdtdt|d WdS|||IdH}|srWdS| ||IdH}|r| rt d |d |dd d | WSt d|WdSty}zt d|d|WYd}~dSd}~ww)a) Transcribe multiple audio chunks by combining them first. Args: audio_chunks: List of base64 encoded audio chunks session_id: Session ID for logging Returns: Transcribed text or None if transcription fails uH🎤 Speech-to-text disabled, skipping buffer transcription for session Nu0⚠️ Failed to decode audio chunk for session ru)⚠️ No valid audio chunks for session @u1🎤 Combined audio too short for transcription: ru,🎤 Transcribed combined audio for session rrru6🎤 No speech detected in combined audio for session u0❌ Error transcribing audio buffer for session ) rrr r!r"r(r%r#r$r&r'rr)) rr/rZcombined_audiochunkr*r-r+r,rrrtranscribe_audio_bufferXsJ        z+SpeechToTextService.transcribe_audio_bufferr*c szF||}tjr|j||}|dur|}|j|dd}t|dkr2tdt|dWdStd|dt|d t|d|WSt ye}zt d |d|WYd}~dSd}~ww) uY Process audio format for Whisper API compatibility. Converts from MCube format (μ-law) to WAV format and applies noise reduction. Args: audio_bytes: Raw audio bytes session_id: Session ID for logging Returns: Processed audio bytes in WAV format Nr1) sample_ratei>u🎵 WAV file too short: ru(🎵 Processed audio format for session rz -> u.❌ Error processing audio format for session ) _mulaw_to_pcmrZAUDIO_NOISE_REDUCTION_ENABLEDrZprocess_audio_chunk_create_wav_filer#rr r(r))rr*rpcm_dataZ processed_pcmZwav_datar-rrrr$s$  &z)SpeechToTextService._process_audio_format mulaw_datacCs>gd}t}|D]}||}||jddddq t|S)u Convert μ-law encoded audio to PCM. Args: mulaw_data: μ-law encoded audio bytes Returns: PCM audio bytes (iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiDiiDiiDiiDiiDiiDiiDiiDiii$idiii$idiii$idiii$idiiiiii4iTitiiiiii4iTitiiiiiiiii ii,ii|<i|:i|8i|6i|4i|2i|0i|.i|,i|*i|(i|&i|$i|"i| iiiiiiiiiiiiiiiii<ii<i i< i i< i i< i i< i i< ii<ii\iiii\iiii\iiii\iiiiliLi,i iiiiiliLi,i iiiiitidiTiDi4i$iixph`XPH@80( rlittleT) byteorderZsigned) bytearrayextendto_bytesbytes)rr8Z mulaw_tabler7ZbyteZ pcm_valuerrrr5s  #z!SpeechToTextService._mulaw_to_pcmr1r7r4cCs&d}d}|||d}||d}t}|d|t|djddd|d |d |djddd|djd dd||jd dd||jddd||jddd||jd dd||jd dd|d |t|jddd|||S) z Create WAV file from PCM data. Args: pcm_data: PCM audio data sample_rate: Sample rate in Hz Returns: WAV file bytes rNrOsRIFF$rQ)rRsWAVEsfmt rPsdata)ioBytesIOwriter#rUgetvalue)rr7r4ZchannelsZbits_per_sampleZ byte_rateZ block_alignZ wav_bufferrrrr6s(       z$SpeechToTextService._create_wav_filec szdd|ji}t}|jd|ddd|d|j|d|j|d d t4Id H}|j|j||tj d d d4Id Hm}|j dkr|| Id H}t d|d|d dd|Wd Id HWd Id HWS| Id H}t d|d|j d|t dt|d Wd Id HWd Id HWd S1Id HswYWd Id HWd S1Id HswYWd Stjyt d|Yd Sty} zt d|d| WYd } ~ d Sd } ~ ww)z Call OpenAI Whisper API for transcription. Args: audio_data: WAV audio data session_id: Session ID for logging Returns: Transcribed text or None if failed Z AuthorizationzBearer filez audio.wavz audio/wav)filenameZ content_typerrZresponse_formattextN)total)headersdataZtimeoutu&🎤 Whisper API response for session rdru"❌ Whisper API error for session rz - u(🎤 Audio data size that caused error: ru$❌ Whisper API timeout for session u*❌ Error calling Whisper API for session )r aiohttpZFormDataZ add_fieldrrZ ClientSessionZpostr Z ClientTimeoutstatusr`rr r'r)r#r TimeoutErrorr() rrrrcrdZsessionZresponser,Z error_textr-rrrr&sN      2z%SpeechToTextService._call_whisper_apiaudio_segmentsc sz>g}|D])}|d}|d}|r/|||IdH}|||t|o&|d}||qtdt|d||WSty^} zt d|d| gWYd} ~ Sd} ~ ww) a Transcribe multiple audio segments in batch. Args: audio_segments: List of audio segments with metadata session_id: Session ID for logging Returns: List of transcribed segments with text r timestampN)rkrr,Z has_speechu🎤 Batch transcribed z segments for session u-❌ Error in batch transcription for session r) getr.boolr'appendrrr#r(r)) rrjrZtranscribed_segmentsZsegmentrrkr,Ztranscribed_segmentr-rrrbatch_transcribeEs,    z$SpeechToTextService.batch_transcribeN)r1)__name__ __module__ __qualname____doc__rstrrr.rr3rVr$r5intr6r&rrrorrrrr s.9'6&21r )rsrr!rZZjsonZwavetypingrrrrrrgZaiofilesZconfigrZservices.log_utilsrZservices.audio_noise_servicer r Zspeech_to_text_servicerrrrs"     Y