80f0bf309f
feat: Apple Speech v3 + Freeflow polish + draft streaming
...
- Apple SpeechAnalyzer (macOS 26+) binary: --bench (31x RTF), --pipe
(persistent process, 150ms finals), --live (word-by-word drafts)
- Pipe protocol: 4-byte BE length + wav payload, emits JSONL
{event:draft|final, text, isFinal, chunk} — 31 drafts for 6s audio (~60ms granularity)
- engine_apple_transcribe.py: ApplePipeTranscriber with
transcribe() + transcribe_with_draft_callback(), VAD + draft
queue, new flags --apple-stream (on), --apple-stream-interval,
--apple-pipe (on). Fixes PIL/transformers import crash by lazy import.
- main_v3.py: engine selector {whisper,apple}, passthrough translate
when no -es/-fr/-ar, freeflow flags same as v2
- Freeflow polish: deterministic punctuation commands (comma,
question mark, new paragraph, at sign), filler stripping,
<keep> protection, skip-clean heuristic, freeflow/qwen/legacy
prompt styles. Much better final readability vs raw Apple/Whisper.
- main_v2.py, engine_llm.py, engine_distribute.py: integrate freeflow
- bench: Apple 2.12% WER vs Whisper Small 3.74% (Inscribe), CPU
0mW ANE (measured via powermetrics), 196M EN cryptex per locale.
- Verified: 31 word-by-word drafts, 2 finals, exit 0, bench regression ok.
Freeflow still much better for final polish — Apple wins on speed
and raw accuracy, freeflow wins on readable paragraph output.
Co-authored-by: internal-model
2026-07-13 21:08:18 -04:00
a2b108a5da
Refine caption display and final prompt experiments
2026-03-17 14:00:42 -04:00
7451d164b7
Tune paragraph prompts for multilingual noise
2026-03-17 11:51:53 -04:00
7da02a9697
Add LLM logging and prompt experiment harness
2026-03-17 11:13:22 -04:00
727b7701c1
Update project history with LLM debugging changes
2026-03-17 01:21:36 -04:00
96fe560f6f
Align Ollama system prompts with OpenAI
2026-03-17 01:19:28 -04:00
53060918e9
Remove unsupported OpenAI temperature override
2026-03-17 00:55:31 -04:00
487cb83080
Show OpenAI API error details for LLM failures
2026-03-17 00:54:56 -04:00
0dcbef7d34
Warm up local LLM before live corrections
2026-03-17 00:29:20 -04:00
9d9ee03ca1
Clarify LLM paragraph fallback behavior
2026-03-17 00:10:39 -04:00
4380c2ef99
Clarify pipeline labels and default to qwen3.5 0.8b
2026-03-17 00:06:47 -04:00
f79c4f7310
Fix main_v2 bridge and pipeline regressions
2026-03-16 21:15:15 -04:00
ffeb996d7e
Checkpoint current transcription pipeline state
2026-03-16 21:11:51 -04:00
50f58dbc22
feat: implement accumulative LLM paragraph engine and harden 4-process pipeline
2026-03-15 22:05:05 -04:00
ca07d96ff0
feat: add advanced tuning, diarization, and LLM support to multi-process engine
2026-03-15 19:53:36 -04:00
c474184169
feat: refactor transcription, translation, and distribution into multiple processes
2026-03-15 19:46:03 -04:00
d2c8d19c61
chore: update .gitignore to exclude .env and log files
2026-03-15 19:26:30 -04:00
5f2738c49c
feat: add speaker diarization and LLM paragraph structuring
2026-03-15 19:25:39 -04:00
7fe41ba88b
Document debug session issues and quick fixes
2026-03-09 21:54:00 -04:00
711a6ce692
Add session logging and tuned runtime preset
2026-03-09 21:33:11 -04:00
4115ddfa8b
feat: warm up ollama and keep post-correct model loaded
2026-03-06 20:48:11 -05:00
2b501f9cdd
fix: harden post-correct prompts and add overlap drift guard
2026-03-06 20:42:01 -05:00
617f2872e7
tune: increase default local llm timeouts for stability
2026-03-06 20:23:09 -05:00
11dbc5ce9a
feat: add optional speaker-change detection for line cuts
2026-03-06 20:02:38 -05:00
2881bd96dd
feat: add optional llm merge-decider for smart caption revisions
2026-03-06 19:56:07 -05:00
564fcaac1a
feat: add v2 post-correct mode with optional local llm
2026-03-06 19:36:08 -05:00
57366543ec
fix: harden arabic terminal rtl rendering
2026-03-06 19:25:12 -05:00
e3cd538dca
feat: load glossary corrections from file when available
2026-03-06 19:11:09 -05:00
221267a172
feat: add intelligent english caption correction v1
2026-03-06 19:05:56 -05:00
dc86ddedbd
feat: improve bridge context and translation decoding controls
2026-03-06 18:56:21 -05:00
6e94210105
docs: add potential Whisper optimizations to backlog in history.md
2026-03-01 11:48:15 -05:00
3a16e44409
docs: update history.md with Phase 13 and 14 for stability and multi-channel support
2026-03-01 11:39:25 -05:00
8f6d2a0303
feat: add multi-channel support and language filtering to focus on specific speakers
2026-03-01 11:39:03 -05:00
00231cc747
feat: add hallucination detection and watchdog for stuck transcriptions
2026-03-01 11:15:59 -05:00
2999502434
feat: include draft progress in ingestion payload when streaming is enabled
2026-02-28 21:08:55 -05:00
018539d41b
feat: parameterize min silence duration and increase default to 1000ms
2026-02-28 19:40:23 -05:00
81cea12e0c
Revert "feat: implement heuristic punctuation buffering for translations when context is enabled"
2026-02-28 19:39:39 -05:00
f68111123d
docs: update history.md with today's work (Phases 10-12)
2026-02-28 19:37:25 -05:00
8b88df8b17
feat: implement heuristic punctuation buffering for translations when context is enabled
2026-02-28 19:36:47 -05:00
ce4809a2b7
feat: implement prompt caching (-c) and source language bypass (--lang)
2026-02-28 19:21:16 -05:00
c519907846
fix: add missing sentencepiece dependency for translation models
2026-02-28 19:09:14 -05:00
6f60d49c7b
feat: implement streaming mode (-s) and 4-bit quantization (-q) for performance
2026-02-28 19:00:56 -05:00
3c874c113c
feat: add command-line arguments for lang selection, ingest toggle, and device management; add README and requirements
2026-02-28 18:55:37 -05:00
0ea605cc0c
Complete history.md with Phase 9: Server Ingest
2026-02-26 23:06:43 -05:00
1403604ce0
Implement server ingest with flat JSON, background worker, and exponential backoff
2026-02-26 23:06:24 -05:00
a4e17ce896
Add multiprocessing.freeze_support() to fix infinite loop in compiled binary
2026-02-26 22:02:25 -05:00
ceed77af14
Document build fix in history.md
2026-02-26 21:57:51 -05:00
7ba75d89dc
Fix mlx._reprlib_fix hidden import in build.sh
2026-02-26 21:57:43 -05:00
1b6f58a28e
Add build script and document compilation process in history.md
2026-02-26 21:55:33 -05:00
ac1582b6cd
Ignore build artifacts
2026-02-26 21:55:02 -05:00
Adolfo Reyna