fix(stt): scope upload size limits to remote providers

[ypwcharles]

What does this PR do?

Fixes a bug where transcribe_audio() applies the 25 MiB remote upload cap to all providers, including local, local_command, and user-configured command providers that never upload files. A 26 MiB WAV file processed locally by faster-whisper gets rejected before the dispatcher is even reached.

The fix moves provider resolution before the size check: local/command providers skip the upload-cap enforcement, while remote built-ins, plugin providers, and unknown providers continue to enforce it.

Before (current main):

validate_audio_file()  ← enforces 25 MiB for ALL providers
  ↓
get_provider()
  ↓
local / openai / plugin / ...

After:

validate_audio_file(enforce_size_limit=False)  ← path/symlink/format only
  ↓
get_provider()
  ↓
if NOT local → validate_audio_file_size()  ← 25 MiB cap
  ↓
dispatch to provider

Related Issue

N/A — discovered during local STT workflow investigation with large meeting recordings.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/transcription_tools.py: Added _is_local_stt_provider() helper and _validate_audio_file_size() extracted from the original monolithic _validate_audio_file(). Size validation now happens after provider resolution, only for non-local providers.
• tests/tools/test_transcription_tools.py: Added oversized file tests for local and local_command providers confirming they reach the dispatcher.
• tests/tools/test_transcription_command_providers.py: Added oversized configured command provider test.
• tests/tools/test_transcription_plugin_dispatch.py: Refactored 9 E2E tests to use real tmp_path files instead of mocking _validate_audio_file() — the validation chain now actually runs. Added oversized plugin rejection test.

How to Test

1. scripts/run_tests.sh tests/tools/test_transcription.py tests/tools/test_transcription_tools.py tests/tools/test_transcription_command_providers.py tests/tools/test_transcription_plugin_dispatch.py -k "not test_timeout_returns_clean_error" -q — 215 passed, 0 failed.
2. The one excluded test (test_timeout_returns_clean_error) is a pre-existing baseline failure caused by the shared venv lacking psutil — triggers the live-system kill guard. Verified identical failure on clean upstream/main with no local changes.
3. Manual probe: 26 MiB sparse WAV + stt.provider: local → enters dispatcher. Same file + stt.provider: openai → returns "File too large" without calling _transcribe_openai.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this is not a duplicate
• My PR contains only changes related to this fix
• I've run the focused test suite — 215 passed, 0 failed
• I've added tests for my changes
• I've tested on: Ubuntu 24.04 (WSL2), Python 3.11.15, pytest 9.0.3

Documentation & Housekeeping

• I've updated relevant documentation — or N/A
• I've updated cli-config.yaml.example — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md — or N/A
• I've considered cross-platform impact (Windows, macOS) — or N/A
• I've updated tool descriptions/schemas — or N/A

Screenshots / Logs

# Manual probe output (26 MiB sparse WAV):
local provider:  success=True, calls=1, error=None
openai provider: success=False, calls=0, error="File too large: 25.0MB (max 25MB)"

[tonydwb]

Code Review Summary

Verdict: Approved

Scopes STT upload size limits to remote providers only (local file processing unaffected). Clean, well-scoped fix. No security concerns.

---

Reviewed by Hermes Agent

[teknium1]

Thanks for isolating a real ordering bug in transcribe_audio(): current main validates the 25 MiB limit before _get_provider() (tools/transcription_tools.py:1732-1745).

Problems

• tools/voice_mode.py:891-896 still chunks every oversized WAV before provider resolution. tests/tools/test_voice_mode.py:792-830 verifies this path while returning provider: "local"; the proposed transcription_tools.py change therefore does not give CLI local recordings the requested direct dispatch.
• The new command-provider exemption classifies all type: command providers as local. The documented contract supports curl-style pipelines (website/docs/user-guide/features/tts.md:542), so this also removes the cap for command providers that upload remotely.

Suggested changes

• Make voice-mode chunking provider-aware and add local-versus-remote oversized recording coverage.
• Limit the command exemption to provably local providers, or add an explicit documented configuration signal for command-provider upload behavior.

Automated hermes-sweeper review.

[ypwcharles]

Addressed review feedback: tightened _is_local_stt_provider() to only exempt local and local_command — configured command providers (including curl-style remote pipelines) now go through the upload size cap like any other non-local provider.

Changes:

• tools/transcription_tools.py: _is_local_stt_provider() simplified to a two-member whitelist
• tests/tools/test_transcription_command_providers.py: oversized command provider test now asserts rejection (not bypass)

Post-fix validation: 215 passed, 0 failed (canonical focused suite).

[ypwcharles]

Addressed voice-mode chunking feedback: transcribe_recording() now trusts transcribe_audio() provider-aware validation — always attempts direct transcription first, only chunks on "File too large" error. Local providers (faster-whisper, whisper.cpp) no longer get blindly chunked.

Changes:

• tools/voice_mode.py: removed pre-check chunking; lazy chunk on size error only
• tests/tools/test_voice_mode.py: updated 2 existing chunking tests + added 3 regression tests (local skip, remote chunk, other error no chunk)

Post-fix: 70 passed, 6 failed (6 pre-existing DetectAudioEnvironment/SubprocessTimeoutKill failures unrelated to this change). New head: d8e4a6866.

[teknium1]

Merged into main via consolidated salvage PR #73510 (merge c0c5dac531). Your change scoping the 25MB upload cap to remote providers only was cherry-picked with your authorship — local whisper no longer rejects big files.

Your contribution is credited to you in git history. Thank you! Closing this PR as merged-via-salvage.