RagLeapFirst release of ragleap-voice with speech engines. Pre-alpha, for testing.
What it adds: transcribe() and synthesize(); a Whisper large-v3 engine (pip install "ragleap-voice[whisper]"); an Indic Parler-TTS engine, which runs as an isolated worker from a Python environment you set up yourself. The core still has no runtime dependencies.
What it does not claim: every model is unverified and none is a default, so calls need allow_unverified=True. Each engine was run on only a few clips on one maintainer's machine (reports in #960 and #970). There is no accuracy, naturalness or emotion measurement, and no comparison with any other service. Not built: streaming, a voice-activity gate, resampling, voice cloning, emotion names.
Help wanted: #960 (test Whisper), #990 (test Parler, gated model), #970 (roadmap).