ufal/whisper_streaming #161 documents the classic Whisper streaming
failure: on ambiguous audio the model falls into a prompt loop,
cascading a single token for 10+ words ("I I I I I I I I I I I…").
The chunk-boundary duplicate detector in live.rs doesn't catch
this — the repeat is within a single chunk, and the text is
technically novel so FTS is happy to keep it.
Fold the detection into is_hallucination as a third pass (after
HALLUCINATION_MARKERS substring-match and HALLUCINATION_TRAIL_PHRASES
exact-match). has_consecutive_repetition walks the token stream
(whitespace-split, lowercased) and returns true when any run of
≥REPETITION_RUN_THRESHOLD (4) identical tokens is found.
Threshold chosen deliberately: three consecutive matches appear in
normal speech ("no no no, that's wrong"), four almost never does.
Tests pin both sides — "I I I I I" detected, "no no no" allowed,
alternating patterns ("I am I am I am I am") allowed regardless of
length.
Phrase-level repetition ("thank you thank you thank you thank you")
is a documented companion failure mode but needs a sliding n-gram
matcher — deferred with a code comment flagging it.
No caller changes: post_process_segments already drops
is_hallucination hits when anti_hallucination is enabled.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Whisper was trained on subtitle corpora, so silence and room tone
trigger caption-style artefacts that the previous three-marker
blocklist ("[blank_audio]", "[music]", "[silence]") didn't catch:
"Thanks for watching!", "Please subscribe.", "ご視聴ありがとうござ
いました", "♪♪♪", etc. Documented in WhisperLive #185 / #246 and
ufal/whisper_streaming #121 as the top streaming-transcript-quality
issue after chunk-boundary repeats.
HALLUCINATION_MARKERS widens from 3 to 16 entries: all common
bracketed non-speech tags (applause / laughter / inaudible /
background noise / sounds), parens variants, and musical notation
(♪ / ♫). Still contains-match so the marker triggers even when
Whisper wraps it in other noise.
HALLUCINATION_TRAIL_PHRASES (renamed from AUTO_THANKS_PHRASES) jumps
from 4 to ~30 entries: YouTube sign-offs, subtitle-credit leakage,
and the two most common non-English variants (Japanese "thanks for
watching" + MBC Korean news sign-off). Stays exact-match so
legitimate dialogue containing "thanks" or "subscribe" mid-sentence
never gets dropped — a new regression test pins that invariant.
The <15-char length gate on trail phrases is removed; some of the
new entries (e.g. "please subscribe to our channel.") are longer.
Exact-match against a known list is safety enough.
No caller changes: post_process_segments already drops segments for
which is_hallucination returns true when anti_hallucination is on.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ai-formatting:
- rule_based.rs: collapse_repetitions() merges adjacent duplicate
tokens like 'I I can' -> 'I can' and 'think think that' -> 'think
that'. Normalises case and punctuation before comparison.
- pipeline.rs: post_process now calls collapse_repetitions when
format_mode is Clean or Smart. Added unit coverage.
audio:
- capture.rs: replace the seven deprecated cpal DeviceTrait::name()
call sites with a device_display_name() helper that uses the
non-deprecated description() path. Keeps identical behaviour,
silences compile warnings, ready for cpal upgrade.
Addresses the 'Christ. Christ.' live-transcription boundary duplicate
Jake saw during Group 1 dogfooding. Does not fix all cross-chunk
overlap cases (see live.rs OVERLAP_SAMPLES for the root cause) but
catches the common stutter pattern at post-processing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Normalise BRITISH_REPLACEMENTS: remove baked-in \b from entries so all
entries are plain base words; the function adds boundaries uniformly
- Replace O(n*m) while-loop double-space removal with single-pass collapse
- Add debug_assert! documenting ASCII assumption for byte-indexed suffix slicing
- Expand llm_client.rs module-level doc comment
- Run cargo fmt on ai-formatting crate
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Security fixes from code audit:
- CSP re-enabled in tauri.conf.json with strict directives
(was null — critical vulnerability)
- XSS fix in viewer highlightText(): HTML entities escaped before
inserting <mark> tags via {@html}
- Removed 3 unwrap() calls in rule_based.rs British English conversion
— replaced with safe let-else guards
- Removed unwrap() on main window lookup in lib.rs setup — now uses
if-let for graceful handling
- Wrapped JSON.parse in DictationPage transcription-result listener
with try/catch
Rebrand cleanup:
- Renamed all localStorage keys from ramble_* to kon_* across
7 files (stores, viewer, float, history)
12 tests passing, clippy clean.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
