+ 6ESj=7Rt^RIHt^RIt^RIt^RIt^RIHtHt^RI t Rt Rt Rt Rt^KRutt^t^tR tR tR t]P,!R 4t]!R R44tRRuttR.R/uttRR.uttRtRt Rt!RRRRRRRRRRRR0RR1RR2/t"RRRRRR0/t#]!RR7!R R!44t$]PJ!4t&]$!4s'R"R#lt(R$R%lt)R&R'lt*R(R)lt+R*R+lt,!R,R-4t-R#)3uKVocal-tone reader: HOW Ahmed said it, not WHAT he said. The STT pipeline reduces every utterance to bare text and throws the voice away — pace, pitch, loudness, laughter, all gone before any brain sees it. This module recovers a few cheap acoustic facts from the SAME utterance buffer the speaker-lock already runs on (main.py) and turns them into a one-line descriptor like "fast, urgent" or "slow, flat" that rides along with the transcript into both brains. Design notes: - **Pure numpy, ~2 ms per utterance.** No model, no torch, no GPU, no numba JIT (librosa.yin would add a 6.5 s first-call compile — not worth it). About 7x cheaper than the TitaNet speaker embedding that already runs every turn. - **We do NOT classify emotion.** Sarcasm/joking aren't decidable from sound alone — the descriptor reports the delivery (e.g. "slow, flat") and Claude infers intent from delivery + words + context, the way a person does. - **Relative to the speaker.** Everyone sits at a different pitch/pace/volume, so we track a rolling EMA baseline of Ahmed's own NEUTRAL delivery and describe deviations from it. The baseline only learns from neutral, clean, loud-enough turns so one shouty or one whisper-quiet utterance can't drag it. - **Pitch uses a periodicity gate.** A frame only contributes an F0 if its autocorrelation peak is a strong fraction of zero-lag (real voicing); noise — including a quiet clip that main.py amplified ~10x — has a weak peak and is rejected, so amplified hiss can't fake a huge pitch range. Pitch is only trusted when enough voiced frames survive; otherwise pitch tags are skipped. - **Rate & pitch dynamics are the reliable dimensions** (the AEC mic's AGC flattens absolute volume, so loudness is a soft, baseline-relative hint). - **Never throws into the run loop.** Any failure returns a neutral read. ) annotationsN) dataclassfieldi>ig{Gz?gffffff?iJg333333?g?g{Gz?z3(?:ha){2,}|(?:he){2,}|\bl(?:ol|mao|mfao)\b|\bhaha\bcb]tRt^9t$RtRtR]R&]!]R7t R]R&]!] R7t R]R &R t R#) ProsodyuOne utterance's read. `note` is None when delivery is unremarkable — the caller injects nothing on neutral turns so the brain isn't spammed.Nz str | Nonenote)default_factoryz list[str]tagsdictfeats) __name__ __module__ __qualname____firstlineno____doc__r__annotations__rlistr r r __static_attributes__r voice/prosody.pyrr9s0OD*D1D)1-E4-rrg?g?g?g?皙?animated?loudgffffff?urgentfastlaughingflatquietslowT)frozenc]tRt^Yt$RtRtR]R&RtR]R&RtR]R&^t R]R&] R R l4t ] R R l4t ] R Rl4t ] RRl4t] RRl4t] RRl4t] RRl4tRtR#) ToneStateuA smoothed read of Ahmed's recent vocal energy — the snapshot the TTS colours delivery with. `energy` < 0 = flat/tired, > 0 = animated; `pace` < 0 = slow; `laugh` decays from 1.0 the turn he laughs. Frozen so a worker thread can read it while the run loop swaps in the next one.floatenergypacelaughintncV^8dQhRR/#returnboolr )formats"r __annotate__ToneState.__annotate__es 88D8rc (VP\8#N)r&_TONE_UPselfs&rup ToneState.upds!%!88rcV^8dQhRR/#r,r )r0s"rr1r2g <rDrIrr rrr rrr#r#YsDFED%E5 AsJ 88 << << :: :: :: ??rr#cV^8dQhRR/#)r-r.r#r )r0s"rr1r1xsIrc\#)uLatest smoothed vocal-tone snapshot (never None; neutral until warmed). O(1), lock-free read — safe from the TTS worker thread at synth time.) _TONE_STATEr rr tone_stater]xs  rc V^8dQhRRRR/#)r-prz 'Prosody'r.Noner )r0s"rr1r1~siDrc VP;'g.pRV9;'g%\VPPR44p\ R \ R\ RV4444p\ R \ R\ RV4444p\;_uu_4\p\VP\W5P, ,,VP\WEP, ,,\ V'dRMRVP\,4VP^,R7sR R R 4R # +'giR #;i) zFold one read into the rolling tone EMA and publish the new snapshot. Side-effect only; reuses the tags read() already produced, so there's no extra DSP and no new call site anywhere.rr(g333333?c3N"TFp\PVR4xK R#5ir$N)_TONE_ENERGY_Wget.0ts& r _publish_tone..s #M1N$6$6q#$>$>#%rc3N"TFp\PVR4xK R#5irc) _TONE_PACE_Wrerfs& rrirjs #KdL$4$4Q$<$)lenFRAMEnpzerosfloat32reshapesqrtmean)rr*frs& r _frame_rmsrsl E ee#AAvxx,, r  2u %B 77B!G>>q>)D0 11rc V^8dQhRRRR/#rr )r0s"rr1r1s-- -z-rcrVR^\,p\V4\,\,pV^8Xd'\P!^\P R7#VRVP R\4p\\,\\,rC.pVFp\P!V^,P44\8dK9WfP4, p\P!WfR4\^, RpV^,pVR8:dKWsVp V P^8XdK\V P44p W,V, \ 8dKVP#\W:,, 4K \P$!V\P R7#)u3Per-voiced-frame F0 via autocorrelation with a PERIODICITY gate: a frame only counts if its best in-band autocorr peak is >= _PERIODICITY of zero-lag. That rejects noise/unvoiced frames (weak peak) — including amplified silence — so hiss can't fake pitch. Analyses up to the first 5 s for a stable read.Nrfullgư>r) SAMPLE_RATErrrrrr_F0_HI_F0_LOrr _VOICED_GATE correlatesizer)argmax _PERIODICITYappendasarray) rr*frameslohioutfaczsegks & r _pitch_trackrs7 "1{? #E E ee#AAvxx,, 2AY  r5 )F F "K6$9C  77AF==? #l 2  L \\! ' 3 qE 9 Bi 88q=    6A: $  ;"&)* ::c ,,rcN]tRt^tRtRRltRRltR RRlltR R ltR t R#) ProsodyReaderzReads vocal tone off raw utterance audio, relative to a rolling baseline of the owner's own neutral delivery. One instance per session; call read() per turn (mirrors SpeakerGate's construct-once/check-per-turn shape).cV^8dQhRR/#r-r.r`r )r0s"rr1ProsodyReader.__annotate__s==$=rc ^VnRVnRVnRVnRVn\ P PRR4R8gVnR#)rNPROSODY10) _n _base_wpm _base_prange _base_loud _base_evarosenvironreenabledr6s&r__init__ProsodyReader.__init__sA'+*.(,(,zz~~i5< rcV^8dQhRR/#rr )r0s"rr1rs  4 rc VVPRJpRpV'dTMV!VPV4VnV!VPV4VnV'dTMV!VPV4VnV'dTMV!VPV4VnV;P^, unR#)NcNVfV#VfV#V\W, ,,#r4)_EMA)rwnews&&rstep,ProsodyReader._update_baseline..steps'{ +3 C31C+C Cr)rrrrr)r7wpmprangerevarfirstrs&&&&& r_update_baselineProsodyReader._update_baselines|$& D!&4+D !2!2F;"'$T$//4-H"'$T$//4-H 1 rNc ,V^8dQhRRRRRRRRR R /#) r-rrspeech_sr%rzr{rz float | Noner.rr )r0s"rr1rs1*S+2rc VP'g \4#VPWW44p\ T4T# \d\4u#i;i \dT#i;i)zaudio: float32 mono @ 16 kHz utterance. speech_s: net voiced seconds. text: transcript. loud: PRE-normalization peak/level (main.py has it); falls back to this clip's RMS. Never raises.)rr_read Exceptionrx)r7rrrzrr_s&&&&& rreadProsodyReader.readso |||9  ET8B  "   9      s > AAA A('A(cV^8dQhRR/#)r-r.rr )r0s"rr1rsW:W:GW:rc a\V4p\T;'gRP44pVf_\\P !\P !VP\P4^,4R,44pV\8gV^8Xd)V'd\RR.RR/4#\RRRV/R7#R V,\VR 4, p\V4pVP'd6\VP4VP 4R,, 4MR p \V4p V P\ 8dM\P"!V ^^K.4wr\W, 4p \\P$!V 44pMR;rR \'V4R \'V^4RV'd \'V4M^RV e \'V 4MRR\'V ^4R\)V P4RVRVP*/pV\,8;'dV P\ 8pVP*\.8gVP0fDV'dVP3W}WI4V'dRMRp\TV'd R.V4#.V4#.oV'dSP5R4\7VP8;'dW@P8R,84pV^8d\WpP0R,8dSP5V'dRMR4M(WpP0R,8dSP5R4V eVP:'dVP<eWP:R,8;'gWP<R,8pWP:R,8;'dWP<R,8pV'dSP5R4MV'dSP5R4V'dS\>;QJdV3RlR'4F 'gK RM R M!V3RlR'44'gSP5R 4M:VP8'd)W@P8R!,8dSP5R"4\>;QJdV3R#lR(4F 'gK RM R M!V3R#lR(44pV'gV'dVP3W}WI4.pSFpVV9gK VP5V4K VR$,pV'dR%PAV4MRp\VVVR&7#))r}NrrlaughterTshortwords)r gN@g333333?r$rrpitchrrvoicedr(r*g?g?rrg ףp= ?r g?g333333?rrrc3,<"TF qS9xK R#5ir4r rgrhr s& rri&ProsodyReader._read..(sF3EaT 3EFrrc3,<"TF qS9xK R#5ir4r rs& rrir0s*)&'Dy)r:NNz, )rr r )rr)rr rrrrr)!rrsplitr%rrrastyper _MIN_SPEECH_Srrorrstdr _MIN_VOICED percentilemedianroundr)r_SEED_MIN_LOUD_SEED_Nrrrr/rrranyjoin)r7rrrzrrn_wordsrfrmsrrp25p75r pitch_medr seedablerloud_hitrrdeviatedseenrhr s&&&&& @rrProsodyReader._reads4=tzzr((*+ <bjj)AQ)F!G$!NOPD m #w!|zJ<*d9KLL'4'!BC CWns8S11% ;?999uTXXZ499;#567#U# :: $}}URH5HC39%Fbii./I!% %F 5:vuT1~ U9% v'9eFmt E$NHc%**o XsDGG  >)GGejjK.G 77W  6%%c4>!):tD4*5I Ir5I I  KK #HHD??S3H,HI a<^^d** Hf=~~,, F#  $"3"3"38S!2!2S!8888//C"77 ..445544  J' F# CCF3EFCCCF3EFFF KK  ___$(>!> KK 3*)*333*)**H  ! !#t :A} ABx"&tyyDDt599r)rrrrrrr4) r rrrrrrrrrr rrrrs#M= "W:W:rrgٿgrng333333g).r __future__rrr~ threading dataclassesrrnumpyrrrrrrrrrrrrcompilerrr5r=rCrHrNrSrWrsrtrdrmr#Lockrrr\r]rxrrrrr rrrsL<# (    S     JJM N  .. .$T*u Iu J    c63#vsc64$NXsFD9  $@@@4^^  k  $% 2-