Function semantic_downsample_factor

src/models/omnivoice/audio_tokenizer.cpp:467–478

}

int64_t semantic_downsample_factor(const OmniVoiceAudioTokenizerConfig & config) {
    const double factor =
        static_cast<double>(checked_positive(config.hop_length, "hop_length")) /
        (static_cast<double>(checked_positive(config.sample_rate, "sample_rate")) /
         static_cast<double>(checked_positive(config.semantic_sample_rate, "semantic_sample_rate"))) /
        static_cast<double>(checked_positive(config.downsample_factor, "downsample_factor"));
    const auto rounded = static_cast<int64_t>(std::llround(factor));
    if (rounded <= 0 || std::fabs(factor - static_cast<double>(rounded)) > 1.0e-6) {
        throw std::runtime_error("OmniVoice semantic_downsample_factor is not an integer");
    }
    return rounded;
}

int64_t semantic_feature_frames(const OmniVoiceAudioTokenizerConfig & config, int64_t semantic_samples) {
    const auto & semantic = config.semantic_model;