Metadata-Version: 2.4
Name: inferencebench-voice
Version: 0.0.2
Summary: Voice transcription plugin for InferenceBench Suite (deterministic fixture-driven WER/CER skeleton)
Project-URL: Homepage, https://github.com/yobitelcomm/bench
Author-email: Yobitel Communications <bench@yobitel.com>
License: Apache-2.0
Keywords: ai,asr,benchmark,ml,transcription,voice,wer
Classifier: Development Status :: 2 - Pre-Alpha
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.12
Requires-Dist: inferencebench-envelope
Requires-Dist: inferencebench-harness
Requires-Dist: pydantic~=2.9
Requires-Dist: pyyaml~=6.0
Description-Content-Type: text/markdown

# inferencebench-voice

Voice transcription (ASR) plugin for the InferenceBench Suite.

Phase-2-quality skeleton: produces signed envelopes via deterministic fixture
processing, with placeholders for real ASR engine invocation that future
revisions wire to faster-whisper-server / OpenAI audio / Cohere transcribe.

Suite ID: `voice.transcription`

Bundled benchmarks:

- `voice.transcription.fleurs-mini` — 5 short utterances, WER scoring.
- `voice.transcription.long-form` — 3 longer references (40-100 words), WER scoring.

The skeleton does NOT actually decode audio. Each fixture's reference is
deterministically corrupted (drop last word, replace with `"end"`) to produce
a known-WER hypothesis. The plugin contract surface is the production one;
future revisions replace the stub `_synthesise_hypothesis` call with a real
audio invocation against a configured `EngineKind`.
