Fix skipped voice lines: one clip per beat (reads whole line) + cap beats to 2 sentences so the clip is short and fully read 6efb75e Running verified HIMANSHUKUMARJHA commited on 27 days ago
Narrator no longer lists the cast by name (evokes scene only) — cuts the name clutter in the opening 67eb8ce verified HIMANSHUKUMARJHA commited on 28 days ago
Cleaner UI (strip duplicated speaker-name labels, no third-person self-naming) + faster (shorter one-sentence lines, max_new_tokens 140) 9ee9374 verified HIMANSHUKUMARJHA commited on 28 days ago
Live text streaming + per-sentence autoplay voice (generate_stream + voice.synthesize; avoids the warm-up worker-reuse stall) efd8b4b verified HIMANSHUKUMARJHA commited on 28 days ago
Option C: compose line, then reveal + autoplay voice sentence-by-sentence (text and voice arrive together) 2595296 verified HIMANSHUKUMARJHA commited on 28 days ago
Restore reliable streaming + per-line voice (revert combined-generator experiment that ZeroGPU would not dispatch) 0dd75d7 verified HIMANSHUKUMARJHA commited on 28 days ago
Lazy-load VoxCPM2 inside GPU (import pre-fork) so the main process stays CUDA-clean and @spaces.GPU generator forks work 1426284 verified HIMANSHUKUMARJHA commited on 28 days ago
Import voice at module load (pre-fork) to avoid import-lock deadlock inside ZeroGPU worker 847f4e6 verified HIMANSHUKUMARJHA commited on 28 days ago
Move combined text+voice streamer into model.py (importable) so ZeroGPU runs its @spaces.GPU generator body cc64c80 verified HIMANSHUKUMARJHA commited on 28 days ago
Add debug tracing to on_start/perform_stream to locate the stall 0fbb6b8 verified HIMANSHUKUMARJHA commited on 28 days ago
Use autoplay (non-streaming) audio output; push one voice clip per sentence as the line is written 843bb4f verified HIMANSHUKUMARJHA commited on 28 days ago
Stream voice with text: synthesize each sentence as the line is written; VoxCPM2 resident + disk-cached refs bb09223 verified HIMANSHUKUMARJHA commited on 28 days ago
Fix MiniCPM5-1B generate (transformers 5.x dict inputs); disable torch.compile so VoxCPM2 runs eager 44ebc1c verified HIMANSHUKUMARJHA commited on 28 days ago
Add per-character VoxCPM2 voices; switch actors to MiniCPM5-1B; draining memory meter; save-highlights a35f6a6 verified HIMANSHUKUMARJHA commited on 28 days ago
Fix UI layout: full-bleed dark background, responsive width, light-theme contrast (app.py only) 9c6739a verified HIMANSHUKUMARJHA commited on 29 days ago
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 83d0062 verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 49be274 verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory c0e1839 verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory c0f78de verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 3444d9e verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 63896c3 verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 2b0a293 verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 73d4abc verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory d5e88dc verified HIMANSHUKUMARJHA commited on Jun 9
Thousand-Token Theater: MiniCPM improv troupe with a hard 1,000-token memory 18150c7 verified HIMANSHUKUMARJHA commited on Jun 9