Whisper Large-v3 Turbo
Esta página de Soniqo documenta Whisper Large-v3 Turbo tal como está implementado en speech-swift / speech-core. Los enlaces a Hugging Face aparecen debajo de las notas de integración.
Primero página interna
Las tarjetas y menús apuntan primero aquí; los enlaces al modelo fuente y a los bundles siguen disponibles en esta página.
Resumen
| Modelo | Whisper Large-v3 Turbo |
|---|---|
| Rol | General multilingual speech-to-text |
| Backend | CoreML fp16 on CPU and Neural Engine |
| Salida | Text transcript |
| Idiomas | Whisper multilingual set, about 100 languages |
| Licencia | MIT weights from OpenAI |
| Estado | Ready through speech transcribe --engine whisper and the WhisperASR Swift product |
| Fuente | openai/whisper-large-v3-turbo |
| Producto Swift | WhisperASR |
| CLI / runtime | speech transcribe --engine whisper |
Uso
El fragmento siguiente refleja la API o el comando actual expuesto por speech-swift.
# Transcribe with the native CoreML Whisper runtime.
.build/release/speech transcribe recording.wav --engine whisper
.build/release/speech transcribe recording.wav --engine whisper --language enEnlaces del modelo
Notas de implementación
- LibriSpeech test-clean slice on an M5 Pro: 1.40% WER, mean RTF 0.089, 6.1 s model load, 384 MB peak RSS — versus 1.53% / 0.085 / 100.2 s / 507 MB for a direct WhisperKit run of the same model.
- The bundle splits into four CoreML models — mel spectrogram, audio encoder, decoder context prefill, and a KV-cached text decoder — with the encoder and decoder on the Neural Engine.
- Greedy no-timestamp decoding with automatic language detection or a --language hint; audio is processed in 30 s chunks with a repeated-word guard against greedy hallucination loops.
- Word timestamps, temperature fallback, and VAD-guided long-form seeking are not implemented yet; --model accepts turbo, large-v3-turbo, or any Hugging Face repo ID.