Sampled Gemini voice matching could leave one person under separate IDs
across transcription chunks. Use hosted Precision-2 on the complete
recording and align its exclusive speaker timeline with MAI word timing.
Preserve transcription text, order, and timestamps; leave ambiguous word
assignments unknown instead of merging identities to satisfy a count.
Upload lossless FLAC, bound polling, retry transient status reads against
the same job, and validate results before emitting stdout. Keep optional
speaker-count and hosted model overrides. Disable MAI diarization and use
30-minute transcription uploads in both modes. Respect environment keys,
keep authentication out of process arguments, and save optional diagnostics
without audio, credentials, or signed upload URLs.
Test Plan:
- python3 -B -m unittest -v: all 101 regression tests passed, including
mocked provider upload, polling, alignment, failure, and cleanup paths.
- bash -n transcribe and shellcheck transcribe: passed.
- git diff --cached --check: passed.
- Prepared the 74-minute input locally and verified identical decoded PCM
hashes for the normalized WAV and full-recording lossless FLAC.
- Live API testing was explicitly deferred by the user; mocked results do
not establish speaker accuracy on the recording.
BREAKING CHANGE: Default diarization requires PYANNOTE_API_KEY or its
1Password reference. --speaker-model accepts hosted precision-2 or
community-1 instead of OpenRouter audio-chat model IDs.
The plain transcribe command still used chunk overlap speaker matching,
so users had to discover a model flag or supply a known speaker count to
get the improved voice comparison behavior.
Enable OpenRouter voice comparison with automatic speaker counting for
normal text and SRT invocations. Keep --speakers and --speaker-model as
optional overrides, and let --no-diarization disable both provider labels
and voice comparison. Make the basic command the documentation entry point
and clarify the defaults in CLI help and repository guidance.
Cover bare invocations across one and multiple uploads, default SRT,
optional count constraints, opt-out ordering, and atomic output when the
default comparison fails.
Test Plan:
- python3 -B -m unittest -v: all 85 tests passed.
- bash -n transcribe: passed.
- git diff --cached --check: passed.
- Tests use real audio encoding with fake provider responses. Live voice
accuracy remains unverified; no live API request was made for this change.
Short chunk overlaps often contain only one participant, so returning
speakers receive new transcript IDs. MAI can also split one voice into
several IDs within a single request, which overlap stitching cannot fix.
Add an opt-in audio comparison pass: --speakers N selects Gemini 3.8 Flash
with a known count, and --speaker-model allows another compatible model.
Keep request-local identities separate, select up to three clean excerpts
per ID, and compare all excerpts in one OpenRouter request. Gemini uses
Google AI Studio because the Vertex backend limits prompts to one audio
file. No local model or extra Python dependency is required.
Only reconcile identities whose sample assignments agree. Preserve
uncertain identities separately, reject incomplete or malformed replies,
and retain the original words and timestamps. Bound sample count and
request size, clip sampling to the real audio duration, and extend worker
cleanup and buffered output to the comparison step. Document the sampled
method's limitations and additional provider request.
Test Plan:
- python3 -B -m unittest -v: all 82 tests passed, including real MP3
encoding with fake transcription/comparison responses and cancellation.
- bash -n transcribe: passed.
- git diff --cached --check: passed.
- Live API and voice accuracy unverified: this workspace could not access
the configured 1Password credential. Mocked responses prove pipeline
behavior, not model accuracy on the supplied recording.
Process up to eight chunks concurrently by default and add --jobs 1..64,
with --jobs 1 retaining sequential operation. Give each worker separate
encoding and request files, release payloads after completion, and merge
responses by original chunk index regardless of completion order.
Use a bounded worker scheduler that handles jobs finishing before wait is
called. Stop scheduling on an observed worker failure and terminate worker
process groups on failure or interruption before removing temporary files.
Include jobs interrupted between launch and scheduler registration in cleanup.
Preserve all-or-nothing stdout for both text and SRT output.
Document the Bash 5.1 requirement, concurrency option, resource usage, and
limits of cancelling requests already accepted by a provider.
Test Plan:
- python3 -B -m unittest -v: all 52 tests passed.
- Mocked provider tests verified eight concurrent requests by default,
configurable limits, rolling worker replacement, isolated payloads,
ordered text, and identical sequential/parallel SRT output.
- Verified immediate encoder failure, concurrent API failure, SIGINT and
SIGTERM, child-process cancellation, cleanup, and no partial stdout.
- bash -n transcribe, ./transcribe --help, and
git diff --cached --check: passed.
- No live provider request; credentials are unavailable in this environment.
Render transcript-wide speaker numbers instead of exposing upload chunks in
reader-facing headers. Reconcile adjacent requests using at least two
unambiguous word-and-time matches in their shared audio, including when
provider speaker IDs change between requests. Reject conflicting mappings
and never treat matching numeric provider IDs as identity evidence.
Assign fresh IDs where continuity cannot be established and report that
uncertainty on stderr. Number only speakers retained after overlap trimming.
Document that overlap matching cannot identify a returning speaker who was
absent from the overlap, and add the reader-facing boundary to AGENTS.md.
Test Plan:
- python3 -B -m unittest -v: all 46 tests passed, including swapped IDs,
ambiguous evidence, continuity across three requests, and plain headers.
- git diff --cached --check: passed.
- Tests use synthetic speaker labels; live diarization accuracy was not
verified because provider credentials are unavailable in this environment.
Chunked text output previously joined every word into one line and discarded
speaker labels. Add a shared plain-text formatter that wraps lines, breaks
paragraphs at speaker changes and pauses, and splits long turns while
preserving word order. Display provider speaker IDs with chunk scope where
needed, without implying that IDs identify the same person across requests.
Request structured output for short diarized recordings too, and warn when
speaker labels are missing or incomplete. Keep SRT formatting separate and
preserve existing line breaks for short requests without diarization.
Test Plan:
- python3 -B -m unittest -q: all 43 tests passed.
- bash -n transcribe and git diff --cached --check: passed.
- Reflowed the existing transcript locally and verified all 15,888 words
were preserved; source audio and generated transcripts are not committed.
- A live excerpt check was blocked before upload by unavailable 1Password
access; actual provider diarization remains unverified.
Request speaker diarization through OpenRouter's Azure provider options for
all transcription uploads. Add --no-diarization to explicitly disable it for
both plain text and SRT output; returned speaker labels let existing SRT
cue generation split at speaker changes.
Default to 10-minute chunks when diarization is enabled because Microsoft
reports failures around 15 minutes and longer. Preserve the 30-minute
default with the opt-out and honor explicit --chunk-minutes values in both
modes. Document the options, duration limits, and output behavior.
Test Plan:
- python3 -B -m unittest -v: all 32 tests passed, including request options,
default upload limits, opt-out combinations, and SRT speaker boundaries.
- bash -n transcribe: passed.
- ./transcribe --help: verified the new option and chunk defaults.
- git diff --cached --check: passed.
- Provider responses were mocked; no live API request was made.
Sending entire recordings in one request can exceed provider limits.
Decode audio once to a shared PCM sample clock and split uploads at a
configurable maximum of 1 to 30 minutes, defaulting to 30 minutes.
Prefer nearby pauses and include six seconds of overlap to protect words
at audio cuts while keeping each encoded MP3 within the duration cap.
Request word timestamps for chunked text and SRT output. Match overlapping
words and times to retain each passage once, warn when a join falls back
to timestamps, and generate SRT cues on the original recording timeline.
Reuse subtitle timestamp validation without reordering regressing words.
Only emit output after every request and conversion succeeds.
Document the Python dependency, temporary storage, boundary limitations,
and lack of automatic retry or resume. Keep single-request text formatting.
Test Plan:
- python3 -B -m unittest -v: all 29 tests passed, including mocked provider
failures, native MP3 payload checks, boundary stitching, and SRT timing.
- bash -n transcribe and shellcheck transcribe: passed.
- git diff --cached --check: passed.
- Locally decoded and encoded the supplied 74-minute WebM into three MP3
chunks; ffprobe confirmed each duration was below 30 minutes.
- Live transcription was not verified: the configured 1Password credential
was unavailable.
ddidderr