Proof
No visual proof cases have been published for this package yet. The registry summary below describes what the compat layer implements.whisper.rn
native-seam only: RNWhisper.install() mounts all fourteen JSI bindings so the package's own JS runs, but transcripts come from the host speech recognizer instead of a whisper model30%
API coverageWorking
package load and JSI binding handshake, getConstants, context and VAD context init/release lifecycle, transcribeFile/transcribeData resolving TranscribeResult shapes, onProgress and onNewSegments callbacks, abortTranscribe, bench, toggleNativeLog, and RealtimeTranscriber delivering real text through the browser SpeechRecognition API
Missing (8)
no whisper.cpp/ggml runtimeso the declared model file is never loaded — no CoreML or GPU accelerationmodel-derived segment timingstoken timestampstinydiarizebeam/temperature decoding optionslanguage auto-detector real VAD speech probabilities