/**
 * @fileoverview Whisper Feature Extraction
 *
 * Implements WhisperFeatureExtractor compatible feature extraction for Smart Turn model.
 * Based on the official Python implementation and Rust reference implementation.
 *
 * The model expects log mel spectrograms with shape [batch, mel_bins, time_frames] = [1, 80, ~3000]
 * for 8 seconds of audio at 16kHz.
 *
 * @module @vowel.to/client/utils
 * @author vowel.to
 * @license Proprietary
 */
/**
 * Extract Whisper features from audio
 *
 * Converts raw audio samples to log mel spectrogram compatible with Whisper models.
 *
 * @param audio - Raw audio samples at 16kHz (Float32Array)
 * @returns Log mel spectrogram with shape [1, N_MELS, time_frames]
 */
export declare function extractWhisperFeatures(audio: Float32Array): Float32Array;
/**
 * Process audio for Smart Turn model
 *
 * Truncates/pads audio to 8 seconds and extracts Whisper features.
 * Output is padded/truncated to exactly 800 frames to match model expectation.
 *
 * @param audio - Raw audio samples at 16kHz (Float32Array)
 * @returns Feature tensor ready for ONNX model input with shape [1, 80, 800]
 */
export declare function processAudioForSmartTurn(audio: Float32Array): Float32Array;
//# sourceMappingURL=whisperFeatures.d.ts.map