feat(m15): SSML prosody tuning with research-validated Hebrew parameters

configs/examples/speaker_AGG_M_30-45_001.yaml

@@ -31,32 +31,32 @@ prosody_baseline:
 style_map:
   1:                                # Low tension — calm surface, controlled
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: 0               # semitones
+    rate_multiplier: 0.97           # M15: consensus Calm [-6%,+2%]; variant: slightly below mid
+    pitch_delta_st: 0               # M15: male pitch -3% to +2% at I1
     volume_delta_db: 0
     rms_target_dbfs: -28            # M3: quiet baseline
   2:                                # Moderate — slight edge, clipped responses
     style: "General"
-    rate_multiplier: 1.05
-    pitch_delta_st: 0               # M2a: AGG anger → faster + louder (M3), not higher pitch
-    volume_delta_db: +2
+    rate_multiplier: 1.03           # M15: consensus Neutral [0%,+4%]; variant: upper-mid
+    pitch_delta_st: 0               # M15: male pitch -2% to +2%; AGG anger → rate not pitch
+    volume_delta_db: 0
     rms_target_dbfs: -25            # M3: +3 dB
   3:                                # Active conflict — raised voice, contemptuous
-    style: "angry"
-    rate_multiplier: 1.15
-    pitch_delta_st: 0               # M2a: stay flat; rate carries urgency
-    volume_delta_db: +5
+    style: "General"
+    rate_multiplier: 1.06           # M15: consensus Irritation [+2%,+8%]; variant: mid-high
+    pitch_delta_st: +1              # M15: male pitch -1% to +3%
+    volume_delta_db: +3
     rms_target_dbfs: -22            # M3: +6 dB above I1
   4:                                # Escalated — aggressive, loud
-    style: "angry"
-    rate_multiplier: 1.20
-    pitch_delta_st: +1              # M2a: minimal rise only at extreme ends
-    volume_delta_db: +9
+    style: "General"
+    rate_multiplier: 1.10           # M15: consensus Anger [+4%,+12%]; variant: upper
+    pitch_delta_st: +2              # M15: male pitch -1% to +5%
+    volume_delta_db: +8
     rms_target_dbfs: -19            # M3: +9 dB above I1
   5:                                # Extreme — shouting, near-clipping
-    style: "angry"
-    rate_multiplier: 1.25
-    pitch_delta_st: +1              # M2a: hard cap; loudness carried by M3
+    style: "General"
+    rate_multiplier: 1.14           # M15: consensus Shouting [+8%,+15%]; variant: mid-high
+    pitch_delta_st: +3              # M15: male pitch 0% to +5%
     volume_delta_db: +13
     rms_target_dbfs: -15            # M3: +13 dB above I1 (satisfies ≥8 dB spec)
 

configs/examples/speaker_AGG_M_30-45_002.yaml

@@ -34,33 +34,33 @@ prosody_baseline:
 style_map:
   1:
     style: "General"
-    rate_multiplier: 1.0
+    rate_multiplier: 0.96           # M15: consensus Calm [-6%,+2%]; variant: lower-mid
     pitch_delta_st: 0
     volume_delta_db: 0
     rms_target_dbfs: -28
   2:
     style: "General"
-    rate_multiplier: 1.05
+    rate_multiplier: 1.01           # M15: consensus Neutral [0%,+4%]; variant: low
     pitch_delta_st: 0
-    volume_delta_db: +2
+    volume_delta_db: 0
     rms_target_dbfs: -25
   3:
     style: "General"
-    rate_multiplier: 1.15
-    pitch_delta_st: 0
-    volume_delta_db: +5
+    rate_multiplier: 1.04           # M15: consensus Irritation [+2%,+8%]; variant: low-mid
+    pitch_delta_st: +1
+    volume_delta_db: +4
     rms_target_dbfs: -22
   4:
     style: "General"
-    rate_multiplier: 1.20
-    pitch_delta_st: +1
+    rate_multiplier: 1.07           # M15: consensus Anger [+4%,+12%]; variant: mid
+    pitch_delta_st: +2
     volume_delta_db: +9
     rms_target_dbfs: -19
   5:
     style: "General"
-    rate_multiplier: 1.25
-    pitch_delta_st: +1
-    volume_delta_db: +13
+    rate_multiplier: 1.11           # M15: consensus Shouting [+8%,+15%]; variant: mid
+    pitch_delta_st: +2
+    volume_delta_db: +14
     rms_target_dbfs: -15
 
 # --- Disfluency profile ---

configs/examples/speaker_BEN_M_40-55_003.yaml

@@ -20,33 +20,33 @@ prosody_baseline:
 style_map:
   1:                                # Routine — somewhat guarded, transactional
     style: "General"
-    rate_multiplier: 1.0
+    rate_multiplier: 0.98           # M15: consensus Calm [-6%,+2%]; variant: mid
     pitch_delta_st: 0
     volume_delta_db: 0
     rms_target_dbfs: -28            # M3: quiet baseline
   2:                                # Frustrated — clipped, short answers
     style: "General"
-    rate_multiplier: 1.05
-    pitch_delta_st: 0               # M2a: anger → faster, not higher; M3 handles loudness
-    volume_delta_db: +2
+    rate_multiplier: 1.04           # M15: consensus Neutral [0%,+4%]; variant: upper
+    pitch_delta_st: 0               # M15: male pitch stays flat at low intensity
+    volume_delta_db: 0
     rms_target_dbfs: -25            # M3: +3 dB
   3:                                # Agitated — loud, accusatory, interrupting
-    style: "angry"
-    rate_multiplier: 1.15
-    pitch_delta_st: 0               # M2a: stay flat; rate carries urgency
-    volume_delta_db: +5
+    style: "General"
+    rate_multiplier: 1.07           # M15: consensus Irritation [+2%,+8%]; variant: high
+    pitch_delta_st: +1              # M15: male pitch -1% to +3%
+    volume_delta_db: +4
     rms_target_dbfs: -22            # M3: +6 dB above I1
   4:                                # Pre-attack — threatening, pacing energy
-    style: "angry"
-    rate_multiplier: 1.20
-    pitch_delta_st: +1              # M2a: minimal rise only at extreme ends
+    style: "General"
+    rate_multiplier: 1.11           # M15: consensus Anger [+4%,+12%]; variant: upper
+    pitch_delta_st: +2              # M15: male pitch -1% to +5%
     volume_delta_db: +9
     rms_target_dbfs: -19            # M3: +9 dB above I1
   5:                                # Attack — shouting, physical sounds added by augmenter
-    style: "angry"
-    rate_multiplier: 1.30
-    pitch_delta_st: +1              # M2a: hard cap; loudness carried by M3
-    volume_delta_db: +13
+    style: "General"
+    rate_multiplier: 1.15           # M15: consensus Shouting [+8%,+15%]; variant: upper
+    pitch_delta_st: +3              # M15: male pitch 0% to +5%
+    volume_delta_db: +14
     rms_target_dbfs: -15            # M3: +13 dB above I1 (satisfies ≥8 dB spec)
 
 disfluency:

configs/examples/speaker_SW_F_30-45_001.yaml

@@ -20,33 +20,33 @@ prosody_baseline:
 style_map:
   1:                                # Routine intake — calm, professional
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: -4              # M2a: lower baseline F0; prevents child-voice range at higher intensities
+    rate_multiplier: 1.0            # M15: consensus Calm [-6%,+2%]; variant: professional pace
+    pitch_delta_st: -1              # M15: female pitch -3% to +2%; variant: near neutral
     volume_delta_db: 0
     rms_target_dbfs: -26            # M3: professional baseline level
   2:                                # Slight tension — still professional, firmer tone
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: -3
-    volume_delta_db: +1
+    rate_multiplier: 1.02           # M15: consensus Neutral [0%,+4%]; variant: firm
+    pitch_delta_st: 0               # M15: female pitch -2% to +3%; variant: neutral
+    volume_delta_db: 0
     rms_target_dbfs: -27            # M3: slightly quieter (controlled de-escalation posture)
   3:                                # Conflict — de-escalation attempt, controlled stress
     style: "General"
-    rate_multiplier: 0.95           # Slows down deliberately (de-escalation technique)
-    pitch_delta_st: -3
+    rate_multiplier: 0.96           # M15: consensus Irritation; variant: deliberate slowing
+    pitch_delta_st: +1              # M15: female pitch 0% to +5%
     volume_delta_db: +2
     rms_target_dbfs: -28            # M3: quieter still (deliberate lowering of voice)
-  4:                                # Escalated — fear breaking through; pitch capped to avoid child-voice range
-    style: "sad"
-    rate_multiplier: 1.05
-    pitch_delta_st: -2              # M2a: cap pitch escalation; additional distress via rate/timing
+  4:                                # Escalated — fear breaking through
+    style: "General"
+    rate_multiplier: 1.0            # M15: consensus Anger; variant: trying to stay composed
+    pitch_delta_st: +2              # M15: female pitch +2% to +7%; fear raising
     volume_delta_db: +4
     rms_target_dbfs: -29            # M3: fear suppresses volume further
   5:                                # Attack — panic, screaming for help
-    style: "sad"
-    rate_multiplier: 1.20
-    pitch_delta_st: -1              # M2a: hard cap
-    volume_delta_db: +5
+    style: "General"
+    rate_multiplier: 1.12           # M15: consensus Shouting [+8%,+15%]; variant: high panic
+    pitch_delta_st: +5              # M15: female pitch +4% to +10%
+    volume_delta_db: +8
     rms_target_dbfs: -30            # M3: panic cry often perceived quieter than aggressor
 
 disfluency:

configs/examples/speaker_VIC_F_25-40_002.yaml

@@ -19,34 +19,34 @@ prosody_baseline:
 style_map:
   1:
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: -4              # M2a: lower baseline F0; prevents child-voice range at higher intensities
+    rate_multiplier: 0.95           # M15: consensus Calm [-6%,+2%]; variant: hesitant
+    pitch_delta_st: -2              # M15: female pitch -3% to +2%
     volume_delta_db: 0
     rms_target_dbfs: -26            # M3: baseline level (slightly quieter than AGG)
   2:                                # Tension — careful, watchful; slight vocal strain
     style: "General"
-    rate_multiplier: 0.95
-    pitch_delta_st: -3
-    volume_delta_db: +1
+    rate_multiplier: 0.96           # M15: consensus Neutral [0%,+4%]; variant: guarded
+    pitch_delta_st: -1              # M15: female pitch -2% to +3%
+    volume_delta_db: 0
     rms_target_dbfs: -27            # M3: voice drops slightly (guarded/hushed)
   3:                                # Conflict — defensive, shaking voice
-    style: "sad"                    # Azure "sad" approximates fear/distress for he-IL
-    rate_multiplier: 0.90
-    pitch_delta_st: -3
-    volume_delta_db: +2
+    style: "General"
+    rate_multiplier: 0.94           # M15: consensus Irritation; variant: very slow
+    pitch_delta_st: +2              # M15: female pitch 0% to +5%; tension
+    volume_delta_db: +3
     rms_target_dbfs: -28            # M3: still dropping (victim shrinks under pressure)
-  4:                                # Escalated — pleading, distress; pitch capped to avoid child-voice range
-    style: "sad"
-    rate_multiplier: 0.88
-    pitch_delta_st: -2              # M2a: cap pitch escalation; additional distress via rate/timing
-    volume_delta_db: +4
+  4:                                # Escalated — pleading, distress
+    style: "General"
+    rate_multiplier: 0.93           # M15: consensus Anger; variant: slow pleading
+    pitch_delta_st: +4              # M15: female pitch +2% to +7%; distress
+    volume_delta_db: +5
     rms_target_dbfs: -29            # M3: near-whisper pleading
   5:                                # Extreme — panic, screaming or sobbing
-    style: "sad"
-    rate_multiplier: 1.10           # Can speed up under panic
-    pitch_delta_st: -1              # M2a: hard cap; distress from rate instability, not further pitch rise
-    volume_delta_db: +5
-    rms_target_dbfs: -30            # M3: sobbing/panic often sounds quieter than shouting (mic overload avoided)
+    style: "General"
+    rate_multiplier: 1.04           # M15: consensus Shouting; variant: panic speeds up
+    pitch_delta_st: +6              # M15: female pitch +4% to +10%
+    volume_delta_db: +7
+    rms_target_dbfs: -30            # M3: sobbing/panic often sounds quieter than shouting
 
 disfluency:
   filled_pause_prob: 0.08           # Higher — victim is more hesitant, searches for words

configs/examples/speaker_VIC_F_25-40_003.yaml

@@ -31,33 +31,33 @@ prosody_baseline:
 style_map:
   1:
     style: "General"
-    rate_multiplier: 0.95
-    pitch_delta_st: -4
+    rate_multiplier: 0.98           # M15: consensus Calm [-6%,+2%]; variant: near-neutral
+    pitch_delta_st: -1              # M15: female pitch -3% to +2%; variant: mild
     volume_delta_db: 0
     rms_target_dbfs: -26
   2:
     style: "General"
-    rate_multiplier: 0.97
-    pitch_delta_st: -3
-    volume_delta_db: -1
+    rate_multiplier: 0.99           # M15: consensus Neutral [0%,+4%]; variant: near-baseline
+    pitch_delta_st: 0               # M15: female pitch -2% to +3%; variant: neutral
+    volume_delta_db: 0
     rms_target_dbfs: -27
   3:
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: -2
-    volume_delta_db: -2
+    rate_multiplier: 0.96           # M15: consensus Irritation; variant: moderate slowing
+    pitch_delta_st: +1              # M15: female pitch 0% to +5%
+    volume_delta_db: +2
     rms_target_dbfs: -28
   4:
     style: "General"
-    rate_multiplier: 1.02
-    pitch_delta_st: -1
-    volume_delta_db: -3
+    rate_multiplier: 0.97           # M15: consensus Anger; variant: less extreme slowing
+    pitch_delta_st: +3              # M15: female pitch +2% to +7%
+    volume_delta_db: +4
     rms_target_dbfs: -29
   5:
     style: "General"
-    rate_multiplier: 1.05
-    pitch_delta_st: -1
-    volume_delta_db: -4
+    rate_multiplier: 1.08           # M15: consensus Shouting; variant: higher panic speed
+    pitch_delta_st: +5              # M15: female pitch +4% to +10%
+    volume_delta_db: +6
     rms_target_dbfs: -30
 
 # --- Disfluency profile ---

configs/speakers/speaker_AGG_M_30-45_003.yaml

@@ -23,33 +23,33 @@ prosody_baseline:
 style_map:
   1:
     style: "General"
-    rate_multiplier: 1.0
-    pitch_delta_st: 0
+    rate_multiplier: 0.95           # M15: consensus Calm [-6%,+2%]; variant: deliberate
+    pitch_delta_st: 0               # M15: male pitch -3% to +2%
     volume_delta_db: 0
     rms_target_dbfs: -28
   2:
     style: "General"
-    rate_multiplier: 1.05
-    pitch_delta_st: 0
-    volume_delta_db: +2
+    rate_multiplier: 1.0            # M15: consensus Neutral [0%,+4%]; variant: baseline
+    pitch_delta_st: 0               # M15: male pitch stays flat
+    volume_delta_db: +1
     rms_target_dbfs: -25
   3:
-    style: "angry"
-    rate_multiplier: 1.12
-    pitch_delta_st: 0
-    volume_delta_db: +5
+    style: "General"
+    rate_multiplier: 1.05           # M15: consensus Irritation [+2%,+8%]; variant: mid
+    pitch_delta_st: +1              # M15: male pitch -1% to +3%
+    volume_delta_db: +3
     rms_target_dbfs: -22
   4:
-    style: "angry"
-    rate_multiplier: 1.18
-    pitch_delta_st: +1
-    volume_delta_db: +9
+    style: "General"
+    rate_multiplier: 1.09           # M15: consensus Anger [+4%,+12%]; variant: mid-high
+    pitch_delta_st: +2              # M15: male pitch -1% to +5%
+    volume_delta_db: +8
     rms_target_dbfs: -19
   5:
-    style: "angry"
-    rate_multiplier: 1.22
-    pitch_delta_st: +1
-    volume_delta_db: +13
+    style: "General"
+    rate_multiplier: 1.13           # M15: consensus Shouting [+8%,+15%]; variant: mid
+    pitch_delta_st: +2              # M15: male pitch 0% to +5%; hard cap
+    volume_delta_db: +12
     rms_target_dbfs: -15
 
 disfluency:

configs/speakers/speaker_BEN_M_40-55_004.yaml

@@ -21,32 +21,32 @@ prosody_baseline:
 style_map:
   1:
     style: "General"
-    rate_multiplier: 1.0
+    rate_multiplier: 0.97           # M15: consensus Calm [-6%,+2%]; variant: mid-low
     pitch_delta_st: 0
     volume_delta_db: 0
     rms_target_dbfs: -28
   2:
     style: "General"
-    rate_multiplier: 1.05
+    rate_multiplier: 1.02           # M15: consensus Neutral [0%,+4%]; variant: low-mid
     pitch_delta_st: 0
-    volume_delta_db: +2
+    volume_delta_db: +1
     rms_target_dbfs: -25
   3:
     style: "General"
-    rate_multiplier: 1.15
-    pitch_delta_st: 0
-    volume_delta_db: +5
+    rate_multiplier: 1.06           # M15: consensus Irritation [+2%,+8%]; variant: mid-high
+    pitch_delta_st: +1
+    volume_delta_db: +4
     rms_target_dbfs: -22
   4:
     style: "General"
-    rate_multiplier: 1.22
-    pitch_delta_st: +1
+    rate_multiplier: 1.10           # M15: consensus Anger [+4%,+12%]; variant: upper-mid
+    pitch_delta_st: +2
     volume_delta_db: +9
     rms_target_dbfs: -19
   5:
     style: "General"
-    rate_multiplier: 1.30
-    pitch_delta_st: +1
+    rate_multiplier: 1.14           # M15: consensus Shouting [+8%,+15%]; variant: mid-high
+    pitch_delta_st: +3              # M15: male pitch upper bound
     volume_delta_db: +13
     rms_target_dbfs: -15