o .-iq&@sNdZddlZddlmZddlmZGdddZGdddZed d Z dS) z Voice Enhancement Service for ElevenLabs VAD Filters background noise and enhances human voice frequencies to improve VAD accuracy and reduce false positives from background noise. Uses only numpy (no scipy) for maximum compatibility. N)Tuple)Logc @sBeZdZdZddedededefddZd ejd ejfd d Z d S)BandpassFilterz;Simple IIR bandpass filter implementation using numpy only.low_freq high_freq sample_rateordercCsJ||_||_||_tdtj|||_tdtj|||_dS)z Initialize bandpass filter. Args: low_freq: Low cutoff frequency in Hz high_freq: High cutoff frequency in Hz sample_rate: Sample_rate in Hz order: Filter order (default 4) gN)rrrnpexppi low_alpha high_alpha)selfrrrr rT/home/vmc/milan/call_management/calls/homebook/services/voice_enhancement_service.py__init__s  zBandpassFilter.__init__audioreturncCst|}tdt|D]}|j||d||||d||<q t|}tdt|D]}d|j|||j||d||<q0|S)z Apply bandpass filter to audio. Args: audio: Input audio signal Returns: Filtered audio signal )r copyrangelenr r)rrZ high_passediZ low_passedrrrapply#s , *zBandpassFilter.applyN)r) __name__ __module__ __qualname____doc__floatintrr ndarrayrrrrrrsrc @seZdZdZddefddZdejdefdd Z dejdefd d Z dejdejfd d Z  d dejdededejfddZ dejde fddZ  d!dede de deee ffddZddZdS)"VoiceEnhancementServicea Advanced audio preprocessing to enhance human voice and suppress background noise. Human voice characteristics: - Fundamental frequency: 85-255 Hz (male), 165-255 Hz (female) - Formants (resonances): 300-3400 Hz - Most energy: 100-4000 Hz Background noise typically: - Low frequency rumble: < 100 Hz - High frequency hiss: > 4000 Hz - Random/non-periodic patterns >rcCsx||_d|_d|_t|j|j||_d|_d|_d|_d|_t dt d|d t d |jd |jd dS) z Initialize voice enhancement service. Args: sample_rate: Audio sample rate in Hz (default 16000 for ElevenLabs) diNg\(\?g{Gz?333333?u:🎙️ Voice Enhancement Service initialized (numpy-only)z Sample Rate: z Hzz Voice Band: -) rZvoice_low_freqZvoice_high_freqrbandpass_filternoise_floor_estimatenoise_floor_alphaenergy_thresholdzero_crossing_rate_thresholdrinfo)rrrrrrIs  z VoiceEnhancementService.__init__rrcCstt|dS)z,Calculate normalized energy of audio signal.)r sqrtmean)rrrrr_calculate_energyisz)VoiceEnhancementService._calculate_energyc Cs,tttt|d}|t|S)zz Calculate zero crossing rate (ZCR). Voice has moderate ZCR, noise has very high or very low ZCR. r-)r sumabsdiffsignr)rrZzero_crossingsrrr_calculate_zero_crossing_ratems  z5VoiceEnhancementService._calculate_zero_crossing_ratecCstj|}t|}t|}|jdur||_n|j|jd|j||_|d|j}t|d|}|td|}tjj |t |d}|S)z Apply spectral subtraction to reduce stationary background noise. Args: audio: Input audio signal Returns: Noise-reduced audio signal Nrg@皙?y?)n) r fftrfftr2angler(r)maximumr irfftr)rrZspectrum magnitudeZphaseZclean_magnitudeZclean_spectrumZ clean_audiorrr_spectral_subtractionus     z-VoiceEnhancementService._spectral_subtractionr%@ thresholdratiocCsFt|}t|}||k}t|||||||||<|S)ad Apply dynamic range compression to make quiet speech louder while preventing loud sounds from clipping. Args: audio: Input audio signal threshold: Compression threshold (0-1) ratio: Compression ratio (e.g., 4:1) Returns: Compressed audio signal )r r2rr4)rrr@rAZ abs_audio compressedmaskrrr _apply_dynamic_range_compressions   z8VoiceEnhancementService._apply_dynamic_range_compressioncCs@||}||}||jkod|ko|jk}|S}|S)z Determine if audio chunk contains voice activity. Args: audio: Input audio signal (normalized -1 to 1) Returns: True if voice activity detected, False otherwise r6)r0r5r*r+)rrZenergyZzcris_voicerrris_voice_activitys  z)VoiceEnhancementService.is_voice_activityT audio_bytesapply_spectral_subtractionapply_compressionc Csz`tj|tjdtj}|d}||}|s,t|}|dtj}|dfWS|j |}|r9| |}|r@| |}t t |}|dkrR|d|}|dtj} | dfWSty~} ztd| |dfWYd} ~ Sd} ~ ww)a Enhance voice in audio by filtering noise and boosting speech frequencies. Args: audio_bytes: Raw PCM audio bytes (16-bit signed) apply_spectral_subtraction: Whether to apply spectral subtraction apply_compression: Whether to apply dynamic range compression Returns: Tuple of (enhanced_audio_bytes, is_voice_detected) )dtypeg@Fgffffff?Tu❌ Voice enhancement failed: N)r frombufferint16astypefloat32rF zeros_liketobytesr'rr>rDmaxr2 Exceptionrerror) rrGrHrIrrEZsilenceZ silence_bytesmax_valZ audio_int16errr enhance_voices.       z%VoiceEnhancementService.enhance_voicecCsd|_tddS)zFReset the noise floor estimate (useful when call environment changes).Nu🔄 Noise floor estimate reset)r(rdebug)rrrrreset_noise_floorsz)VoiceEnhancementService.reset_noise_floorN)r#)r%r?)TT)rrrrr rr r!rr0r5r>rDboolrFbytesrrVrXrrrrr":s4 &    8r"r#)r) rnumpyr typingrservices.log_utilsrrr"voice_enhancement_servicerrrrs  ,T