Experiment 7-6: speech SFT acceptance campaign

This directory contains the reproducible local-GPU campaign and its retained evidence for both speech-training tracks described in the chapter:

• Orpheus cross-sentence voice/timbre consistency
• Sesame CSM control of <laughs>, <giggles>, and <sighs> events

The retained run is validation/exp7-6-20260804-v1/. It performed 60 optimizer updates for each LoRA, used disjoint held-out loss sets, generated matched base/adapted WAV comparisons, published the full adapters to Hugging Face, and kept explicit negative comparisons. See the run's REPORT.md for results and limitations.

The retained compatibility_failures.json also records the current Unsloth CSM pad-token rejection and Transformers bf16 codec/text merge mismatch. Sesame was therefore trained with standard PEFT in float32, without reducing the dataset, optimizer-step count, or comparison campaign.

Reproduce

Use a fresh environment because the two upstream notebooks move quickly:

python3 -m venv --system-site-packages .venv-exp7-6
.venv-exp7-6/bin/pip install -r chapter7/speech-sft-experiment/requirements.txt

.venv-exp7-6/bin/python chapter7/speech-sft-experiment/run_orpheus.py \
  --output chapter7/speech-sft-experiment/validation/my-run

.venv-exp7-6/bin/python chapter7/speech-sft-experiment/run_sesame.py \
  --output chapter7/speech-sft-experiment/validation/my-run

.venv-exp7-6/bin/python chapter7/speech-sft-experiment/analyze_campaign.py \
  --run chapter7/speech-sft-experiment/validation/my-run

The runners default to bojieli/... adapter repositories. Pass --hf-repo with a repository you can write, or modify the runners to skip publication for a private local reproduction. HF_TOKEN is required for publication.

Dataset provenance

The upstream notebooks name MrDragonFox/Elise. Hugging Face now marks that dataset disabled. The campaign therefore uses maxbsoft/mrdragonfox-elise at immutable revision 2cc657c3f94a83df18fcd968b7531ca1a19c7f88, a public non-disabled mirror of the 1,195-row Elise corpus. Both manifests record this substitution.

Interpretation

Execution acceptance and hypothesis support are separate. A run can be complete while an automatic quality proxy is negative. The MFCC statistic cosine used for Orpheus is a transparent timbre proxy. The AudioSet detector scores used for Sesame are event-presence proxies. Neither replaces a blinded human listening test, MOS, or enrolled-speaker verification, and the report does not claim perceptual quality from this bounded campaign.