3 days ago
-
three takes for weak lines, keep the best
-
retake near-silent takes; whisper cap 15
-
retake near-silent takes; whisper cap 15
-
whisper lines capped at 18 chars; retake short takes
-
whisper lines capped at 18 chars; retake short takes
-
whisper lines capped at 18 chars; retake short takes
-
radio: read a director script file (was stale on the hub)
-
director -> tts pipeline with two interpreters
-
director train: reuse collected scripts
-
director: 160 training scripts + fixed block lookup
-
director: 160 training scripts + fixed block lookup
-
director: 40 topics + collect/train/eval driver
-
director: 40 topics + collect/train/eval driver
-
director: 40 topics + collect/train/eval driver
-
director: six prefixed lines instead of JSON
-
director: six prefixed lines instead of JSON
-
director: few-shot prompt + honest contract (8-40 chars)
-
director: few-shot prompt + honest contract (8-40 chars)
-
director probe: drop presence_penalty, use dtype=
-
director: run the probe in a fresh interpreter
-
director: import sys
-
director: hold huggingface_hub below 2 (httpx renamed)
-
director: resolve transformers5 + hub1 in one pip pass
-
director: Qwen3.5-0.8B probe against the TTS contract
-
director: Qwen3.5-0.8B probe against the TTS contract
-
radio: student reuses teacher frames; parse malformed script JSON
-
radio: require six longer lines
4 days ago
-
train the student on the whisper voice
-
mining: reject probe CER>0.5, drop unusable modulation gate
-
whisper slot a09 from buy1take2; radio cast updated
-
style probe: FREEFORM instruction and BASIC controls
-
style probe: FREEFORM instruction and BASIC controls
-
whisper probe: 6 low-voicing candidates
-
whisper probe: 6 low-voicing candidates
-
radio notebook: verify checkpoint length
-
radio demo: 4000-step student + end-to-end script
-
radio demo: 4000-step student + end-to-end script
-
voc: stop at the ISTFT head projection, ship spec
-
fix lm cache order and additive mask; probe audio VAE
-
export: drop removed kwarg, host-side numpy
-
export: put dummy tensors on the model device
-
browser export: lm/aggregator/voc with validation + int4
-
browser export: lm/aggregator/voc with validation + int4
-
add structure probe for the browser export
-
add structure probe for the browser export
-
release candidate: 4000 steps, seeded teacher, keep audio
-
release candidate: 4000 steps, seeded teacher, keep audio
-
run4: unpack cached traj, seed teacher, 8000 steps
-
run4: unpack cached traj, seed teacher, 8000 steps
-
run 3: 4000 steps, keep traj cache
