fix(config): halve pitch escalation at I4–I5 to eliminate helium effect

[shaypal5]

Closes #64.

Summary

• Listening test revealed VIC female pitch at I5 reached +6 st (=36%), far exceeding the M15 research consensus of +4–10%. Root cause: semitone/percent unit conversion error in M15 calibration.
• Halves pitch_delta_st across all 10 speaker YAMLs (6 female VIC/SW, 4 male AGG/BEN)
• Reduces SpeakerState pitch drift targets so accumulated cross-turn pitch doesn't compound the problem
• Tightens MAX_F0_DRIFT_ST from 2.0 → 1.5 semitones

Before / After (VIC female, 210 Hz baseline, I5 after 3 escalating turns)

Component	Before	After
style_map pitch_delta_st	+6 st (36%)	+3 st (18%)
SpeakerState drift	+0.9 st	+0.6 st
Total pitch shift	+6.9 st → 297 Hz	+3.6 st → 255 Hz
Perceptual	Helium / oompa-loompa	Within natural female range

Changes per file

File	Change
configs/examples/speaker_VIC_F_25-40_002.yaml	I1:-2→0, I2:-1→0, I3:+2→+1, I4:+4→+2, I5:+6→+3
configs/examples/speaker_VIC_F_25-40_003.yaml	I1:-1→0, I4:+3→+2, I5:+5→+3
configs/speakers/speaker_VIC_F_25-40_004.yaml	I1:-2→0, I2:-1→0, I3:+2→+1, I4:+4→+2, I5:+6→+3
configs/examples/speaker_SW_F_30-45_001.yaml	I5:+5→+3
configs/speakers/speaker_SW_F_30-45_002.yaml	I4:+3→+2, I5:+5→+3
configs/speakers/speaker_SW_F_30-45_003.yaml	I1:-2→0, I2:-1→0, I3:+2→+1, I4:+4→+2, I5:+6→+3
configs/examples/speaker_AGG_M_30-45_001.yaml	I5:+3→+2
configs/examples/speaker_BEN_M_40-55_003.yaml	I5:+3→+2
configs/speakers/speaker_BEN_M_40-55_004.yaml	I5:+3→+2
configs/speakers/speaker_BEN_M_40-55_005.yaml	I5:+3→+2
synthbanshee/tts/speaker_state.py	AGG/VIC pitch drift targets reduced; MAX_F0_DRIFT_ST 2.0→1.5
tests/unit/test_speaker_state.py	Updated drift bound assertions

Test plan

• All 1435 tests pass
• ruff check — passed
• ruff format — passed
• mypy — passed
• Re-generate listening test clips after merge to verify improvement

🤖 Generated with Claude Code

[github-actions]

pr-agent-context report:

No unresolved review comments, failing checks, or actionable patch coverage gaps were found on PR #68 in repository https://github.com/DataHackIL/SynthBanshee. Treat this PR as all clear unless new signals appear.

Run metadata:

Tool ref: v4
Tool version: 4.0.21
Trigger: commit pushed
Workflow run: 25290315136 attempt 1
Comment timestamp: 2026-05-03T20:45:38.956618+00:00
PR head commit: 307f1297b7c45bf02c23032983ec820bea572214